From 565c992589ef2bf94a7fda75e857b3852702ee06 Mon Sep 17 00:00:00 2001 From: Jade Choghari Date: Thu, 11 Sep 2025 13:47:58 +0200 Subject: [PATCH] iterate on review --- examples/6_evaluate_libero_2.sh | 76 - examples/7_train_acc.sh | 93 - examples/7_train_libero_smolvla.sh | 89 - examples/8_train_smolvla_must.sh | 145 -- examples/9_evaluate_must.sh | 2811 ---------------------------- examples/checker.py | 27 - examples/checker2.py | 35 - examples/convert_data.py | 253 --- examples/convert_libero.py | 126 -- examples/evaluate_libero.py | 255 --- examples/requirements.in | 8 - examples/script2.py | 70 - examples/script3.py | 64 - examples/script4.py | 3 - log_text.txt | 1765 ----------------- src/lerobot/envs/libero copy.py | 326 ---- src/lerobot/envs/libero2.py | 308 --- tmux_log.txt | 2008 -------------------- 18 files changed, 8462 deletions(-) delete mode 100644 examples/6_evaluate_libero_2.sh delete mode 100644 examples/7_train_acc.sh delete mode 100644 examples/7_train_libero_smolvla.sh delete mode 100644 examples/8_train_smolvla_must.sh delete mode 100644 examples/9_evaluate_must.sh delete mode 100644 examples/checker.py delete mode 100644 examples/checker2.py delete mode 100644 examples/convert_data.py delete mode 100644 examples/convert_libero.py delete mode 100644 examples/evaluate_libero.py delete mode 100644 examples/requirements.in delete mode 100644 examples/script2.py delete mode 100644 examples/script3.py delete mode 100644 examples/script4.py delete mode 100644 log_text.txt delete mode 100644 src/lerobot/envs/libero copy.py delete mode 100644 src/lerobot/envs/libero2.py delete mode 100644 tmux_log.txt diff --git a/examples/6_evaluate_libero_2.sh b/examples/6_evaluate_libero_2.sh deleted file mode 100644 index 9d05c0330..000000000 --- a/examples/6_evaluate_libero_2.sh +++ /dev/null @@ -1,76 +0,0 @@ -#!/bin/bash - -# storage / caches -RAID=/raid/jade -export TRANSFORMERS_CACHE=$RAID/.cache/huggingface/transformers -export HF_HOME=$RAID/.cache/huggingface -export HF_DATASETS_CACHE=$RAID/.cache/huggingface/datasets -export HF_LEROBOT_HOME=$RAID/.cache/huggingface/lerobot -export WANDB_CACHE_DIR=$RAID/.cache/wandb -export TMPDIR=$RAID/.cache/tmp -mkdir -p $TMPDIR -export WANDB_MODE=offline -export HF_DATASETS_OFFLINE=1 -export HF_HUB_OFFLINE=1 -export TOKENIZERS_PARALLELISM=false -export MUJOCO_GL=egl -export CUDA_VISIBLE_DEVICES=3 - -# CONFIGURATION -POLICY_PATH="/raid/jade/logs/lerobot/lerobot_2_HuggingFaceVLA_libero_smolvla_lr1e-4bs32steps100000/checkpoints/100000/pretrained_model" -POLICY_PATH="AustineJohnBreaker/smolvla_stratch_libero_spatial" -TASK=libero_spatial -ENV_TYPE="libero" -BATCH_SIZE=10 -N_EPISODES=10 -USE_AMP=false -N_ACTION_STEPS=1 -SELF_ATTN_EVERY_N_LAYERS=2 -VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct -PAD_LANG_TO=longest -LOAD_VLM_WEIGHTS=true -NUM_VLM_LAYERS=16 -CHUNK_SIZE=50 -N_OBS_STEPS=1 -NUM_EXPERT_LAYERS=0 -EXPERT_WIDTH_MULTIPLIER=0.5 - - -# storage / caches -RAID=/raid/jade -export TRANSFORMERS_CACHE=$RAID/.cache/huggingface/transformers -export HF_HOME=$RAID/.cache/huggingface -export HF_DATASETS_CACHE=$RAID/.cache/huggingface/datasets -export HF_LEROBOT_HOME=$RAID/.cache/huggingface/lerobot -export WANDB_CACHE_DIR=$RAID/.cache/wandb -export TMPDIR=$RAID/.cache/tmp -mkdir -p $TMPDIR -export WANDB_MODE=offline -# export HF_DATASETS_OFFLINE=1 -# export HF_HUB_OFFLINE=1 -export TOKENIZERS_PARALLELISM=false -export MUJOCO_GL=egl -export MUJOCO_GL=egl -ADD_IMAGE_TOKENS=true -unset HF_HUB_OFFLINE -# RUN EVALUATION -python src/lerobot/scripts/eval.py \ - --policy.path="$POLICY_PATH" \ - --env.type="$ENV_TYPE" \ - --eval.batch_size="$BATCH_SIZE" \ - --eval.n_episodes="$N_EPISODES" \ - --env.multitask_eval=False \ - --env.task=$TASK \ - --policy.use_amp=$USE_AMP \ - --policy.n_action_steps=$N_ACTION_STEPS \ - # --policy.add_image_special_tokens=$ADD_IMAGE_TOKENS \ - --policy.attention_mode=$ATTN_MODE \ - --policy.self_attn_every_n_layers=$SELF_ATTN_EVERY_N_LAYERS \ - --policy.vlm_model_name=$VLM_NAME \ - --policy.pad_language_to=$PAD_LANG_TO \ - --policy.load_vlm_weights=$LOAD_VLM_WEIGHTS \ - --policy.num_vlm_layers=$NUM_VLM_LAYERS \ - --policy.chunk_size=$CHUNK_SIZE \ - --policy.n_obs_steps=$N_OBS_STEPS \ - --policy.num_expert_layers=$NUM_EXPERT_LAYERS \ - --policy.expert_width_multiplier=$EXPERT_WIDTH_MULTIPLIER \ diff --git a/examples/7_train_acc.sh b/examples/7_train_acc.sh deleted file mode 100644 index 27f445143..000000000 --- a/examples/7_train_acc.sh +++ /dev/null @@ -1,93 +0,0 @@ -#!/bin/bash -# smolvla training with accelerate - -set -euo pipefail - -# repo/env -cd ~/lerobot || exit 1 -# conda activate lerobot -export LC_ALL=C - -rm -f core-* - -# storage / caches -RAID=/raid/jade -export TRANSFORMERS_CACHE=$RAID/.cache/huggingface/transformers -export HF_HOME=$RAID/.cache/huggingface -export HF_DATASETS_CACHE=$RAID/.cache/huggingface/datasets -export HF_LEROBOT_HOME=$RAID/.cache/huggingface/lerobot -export WANDB_CACHE_DIR=$RAID/.cache/wandb -export TMPDIR=$RAID/.cache/tmp -mkdir -p $TMPDIR -export WANDB_MODE=offline -export HF_DATASETS_OFFLINE=1 -export HF_HUB_OFFLINE=1 -export TOKENIZERS_PARALLELISM=false -export MUJOCO_GL=egl - -# CONFIG -ENV=libero -TASK=libero_spatial -REPO_ID=physical-intelligence/libero - -POLICY=smolvla -VLM=HuggingFaceTB/SmolVLM2-500M-Instruct - -# Optim / scheduling -LR=1e-4 -DECAY_LR=2.5e-6 -DECAY_STEPS=30000 -USE_AMP=true # set to true for mixed precision -TRAIN_EXPERT_ONLY=true -N_ACTION_STEPS=1 -SEED=1000 - -# Training loop -OFFLINE_STEPS=100000 -BATCH_SIZE=32 -EVAL_FREQ=0 -SAVE_FREQ=20000 -EVAL_BATCH_SIZE=1 -NUM_EPISODES=1 - -# number of gpus to use -NUM_PROCESSES=2 -export CUDA_VISIBLE_DEVICES=1,3 -PORT=29522 - -# naming/output dir -TRAIN_DIR=$RAID/logs/lerobot/lerobot_2_${REPO_ID//\//_}_${POLICY}_lr${LR}bs${BATCH_SIZE}steps${OFFLINE_STEPS} -echo "Training dir: $TRAIN_DIR" - -rm -rf "$TRAIN_DIR" - -# RUN -python -m accelerate.commands.launch \ - --num_processes $NUM_PROCESSES \ - --num_machines 1 \ - --main_process_port $PORT \ - --mixed_precision=$( [ "$USE_AMP" = true ] && echo "bf16" || echo "no" ) \ - src/lerobot/scripts/train_accelerate.py \ - --policy.type=$POLICY \ - --policy.use_amp=True \ - --policy.vlm_model_name=$VLM \ - --dataset.repo_id=$REPO_ID \ - --dataset.root=$HF_DATASETS_CACHE \ - --env.type=$ENV \ - --env.task=$TASK \ - --output_dir=$TRAIN_DIR \ - --batch_size=$BATCH_SIZE \ - --steps=$OFFLINE_STEPS \ - --eval_freq=$EVAL_FREQ \ - --save_freq=$SAVE_FREQ \ - --eval.batch_size=$EVAL_BATCH_SIZE \ - --eval.n_episodes=$NUM_EPISODES \ - --policy.optimizer_lr=$LR \ - --policy.repo_id=None \ - --policy.scheduler_decay_lr=$DECAY_LR \ - --policy.scheduler_decay_steps=$DECAY_STEPS \ - --policy.n_action_steps=$N_ACTION_STEPS \ - --policy.train_expert_only=$TRAIN_EXPERT_ONLY \ - --policy.vlm_model_name=$VLM \ - --seed=$SEED \ - --wandb.enable=false diff --git a/examples/7_train_libero_smolvla.sh b/examples/7_train_libero_smolvla.sh deleted file mode 100644 index f0b9de4e5..000000000 --- a/examples/7_train_libero_smolvla.sh +++ /dev/null @@ -1,89 +0,0 @@ -#!/bin/bash -# smolvla training - -set -euo pipefail - -# repo/env -cd ~/lerobot || exit 1 -# conda activate lerobot -export LC_ALL=C - - -rm -f core-* - -# storage / caches (use RAID to avoid filling $HOME) -RAID=/raid/jade -export TRANSFORMERS_CACHE=$RAID/.cache/huggingface/transformers -export HF_HOME=$RAID/.cache/huggingface -export HF_DATASETS_CACHE=$RAID/.cache/huggingface/datasets -export HF_LEROBOT_HOME=$RAID/.cache/huggingface/lerobot -export WANDB_CACHE_DIR=$RAID/.cache/wandb -export TMPDIR=$RAID/.cache/tmp -mkdir -p $TMPDIR -export WANDB_MODE=offline -# export HF_DATASETS_OFFLINE=1 -# export HF_HUB_OFFLINE=1 -export TOKENIZERS_PARALLELISM=false -export MUJOCO_GL=egl - -# will only use if accelerate is used -PORT=29522 - -# =================== CONFIG =================== -ENV=libero -TASK=libero_object -REPO_ID=physical-intelligence/libero -ROOT=$RAID -POLICY=smolvla -VLM=HuggingFaceTB/SmolVLM2-500M-Instruct - -# Optim / scheduling -LR=1e-4 -DECAY_LR=2.5e-6 -DECAY_STEPS=30000 -USE_AMP=false -TRAIN_EXPERT_ONLY=true -N_ACTION_STEPS=1 -SEED=1000 - -# Training loop -OFFLINE_STEPS=100000 -BATCH_SIZE=32 -EVAL_FREQ=0 -SAVE_FREQ=300000 -EVAL_BATCH_SIZE=1 -NUM_EPISODES=1 - -# GPU selection 0, 1, 2, 3 -export CUDA_VISIBLE_DEVICES=0 - -# naming/output dir -TRAIN_DIR=$RAID/logs/lerobot/lerobot_solo_${REPO_ID//\//_}_${POLICY}_lr${LR}bs${BATCH_SIZE}steps${OFFLINE_STEPS} -echo "Training dir: $TRAIN_DIR" - -# train -rm -rf "$TRAIN_DIR" - -python src/lerobot/scripts/train.py \ - --policy.type=$POLICY \ - --policy.vlm_model_name=$VLM \ - --dataset.repo_id=$REPO_ID \ - --env.type=$ENV \ - --env.task=$TASK \ - --output_dir=$TRAIN_DIR \ - --batch_size=$BATCH_SIZE \ - --steps=$OFFLINE_STEPS \ - --eval_freq=$EVAL_FREQ \ - --save_freq=$SAVE_FREQ \ - --eval.batch_size=$EVAL_BATCH_SIZE \ - --eval.n_episodes=$NUM_EPISODES \ - --policy.use_amp=$USE_AMP \ - --policy.optimizer_lr=$LR \ - --policy.repo_id=None \ - --policy.scheduler_decay_lr=$DECAY_LR \ - --policy.scheduler_decay_steps=$DECAY_STEPS \ - --policy.n_action_steps=$N_ACTION_STEPS \ - --policy.train_expert_only=$TRAIN_EXPERT_ONLY \ - --policy.vlm_model_name=$VLM \ - --seed=$SEED \ - --wandb.enable=false diff --git a/examples/8_train_smolvla_must.sh b/examples/8_train_smolvla_must.sh deleted file mode 100644 index 828627d85..000000000 --- a/examples/8_train_smolvla_must.sh +++ /dev/null @@ -1,145 +0,0 @@ -#!/bin/bash -# smolvla training with accelerate - -set -euo pipefail - -# repo/env -cd ~/lerobot || exit 1 -# conda activate lerobot -export LC_ALL=C - -rm -f core-* - -# storage / caches -RAID=/raid/jade -export TRANSFORMERS_CACHE=$RAID/.cache/huggingface/transformers -export HF_HOME=$RAID/.cache/huggingface -export HF_DATASETS_CACHE=$RAID/.cache/huggingface/datasets -export HF_LEROBOT_HOME=$RAID/.cache/huggingface/lerobot -export WANDB_CACHE_DIR=$RAID/.cache/wandb -export TMPDIR=$RAID/.cache/tmp -mkdir -p $TMPDIR -export WANDB_MODE=offline -# export HF_DATASETS_OFFLINE=1 -# export HF_HUB_OFFLINE=1 -export TOKENIZERS_PARALLELISM=false -export MUJOCO_GL=egl - -# CONFIG -ENV=libero -TASK=libero_spatial -REPO_ID=HuggingfaceVLA/libero - -POLICY=smolvla -VLM=HuggingFaceTB/SmolVLM2-500M-Instruct - -# Optim / scheduling -LR=1e-4 -DECAY_LR=2.5e-6 -DECAY_STEPS=30000 -USE_AMP=true # set to true for mixed precision -TRAIN_EXPERT_ONLY=true -N_ACTION_STEPS=1 -SEED=1000 -LOAD_VLM_WEIGHTS=true -# Training loop -OFFLINE_STEPS=100000 -BATCH_SIZE=32 -EVAL_FREQ=0 -SAVE_FREQ=20000 -EVAL_BATCH_SIZE=1 -NUM_EPISODES=1 -ADD_IMAGE_TOKENS=tru -N_OBS_STEPS=1 -ATTN_MODE=cross_attn -EXPERT_WIDTH_MULTIPLIER=0.5 -# number of gpus to use -NUM_PROCESSES=2 -NUM_VLM_LAYERS=0 -SELF_ATTN_EVERY_N_LAYERS=0 -CHUNK_SIZE=50 -export CUDA_VISIBLE_DEVICES=1 -PORT=29522 -PREFIX_LENGTH=0 -LOAD_VLM_WEIGHTS=true -MAX_ACTION_DIM=32 -MAX_STATE_DIM=32 -# naming/output dir -TRAIN_DIR=$RAID/logs/lerobot/lerobot_new_sep11_v2_${REPO_ID//\//_}_${POLICY}_lr${LR}bs${BATCH_SIZE}steps${OFFLINE_STEPS} -echo "Training dir: $TRAIN_DIR" - -rm -rf "$TRAIN_DIR" - -# RUN -# python -m accelerate.commands.launch \ -# --num_processes $NUM_PROCESSES \ -# --num_machines 1 \ -# --main_process_port $PORT \ -# --mixed_precision=$( [ "$USE_AMP" = true ] && echo "bf16" || echo "no" ) \ -# src/lerobot/scripts/train_accelerate.py \ -# --policy.type=$POLICY \ -# --policy.use_amp=True \ -# --policy.vlm_model_name=$VLM \ -# --dataset.repo_id=$REPO_ID \ -# --dataset.root=$HF_DATASETS_CACHE \ -# --env.type=$ENV \ -# --env.task=$TASK \ -# --output_dir=$TRAIN_DIR \ -# --batch_size=$BATCH_SIZE \ -# --steps=$OFFLINE_STEPS \ -# --eval_freq=$EVAL_FREQ \ -# --save_freq=$SAVE_FREQ \ -# --eval.batch_size=$EVAL_BATCH_SIZE \ -# --eval.n_episodes=$NUM_EPISODES \ -# --policy.optimizer_lr=$LR \ -# --policy.repo_id=None \ -# --policy.scheduler_decay_lr=$DECAY_LR \ -# --policy.scheduler_decay_steps=$DECAY_STEPS \ -# --policy.n_action_steps=$N_ACTION_STEPS \ -# --policy.train_expert_only=$TRAIN_EXPERT_ONLY \ -# --policy.vlm_model_name=$VLM \ -# --policy.n_obs_steps=$N_OBS_STEPS \ -# --policy.attention_mode=$ATTN_MODE \ -# --policy.prefix_length=$PREFIX_LENGTH \ -# --policy.num_vlm_layers=$NUM_VLM_LAYERS \ -# --policy.chunk_size=$CHUNK_SIZE \ -# --policy.expert_width_multiplier=$EXPERT_WIDTH_MULTIPLIER \ -# --policy.self_attn_every_n_layers=$SELF_ATTN_EVERY_N_LAYERS \ -# --seed=$SEED \ -# --wandb.enable=false - - -python src/lerobot/scripts/train.py \ - --policy.type=$POLICY \ - --policy.use_amp=False \ - --policy.vlm_model_name=$VLM \ - --dataset.repo_id=$REPO_ID \ - --dataset.root='/raid/jade/.cache/huggingface/lerobot/HuggingFaceVLA/libero/data' \ - --env.type=$ENV \ - --env.task=$TASK \ - --output_dir=$TRAIN_DIR \ - --batch_size=$BATCH_SIZE \ - --steps=$OFFLINE_STEPS \ - --eval_freq=$EVAL_FREQ \ - --save_freq=$SAVE_FREQ \ - --eval.batch_size=$EVAL_BATCH_SIZE \ - --eval.n_episodes=$NUM_EPISODES \ - --policy.optimizer_lr=$LR \ - --policy.repo_id=None \ - --policy.scheduler_decay_lr=$DECAY_LR \ - --policy.scheduler_decay_steps=$DECAY_STEPS \ - --policy.n_action_steps=$N_ACTION_STEPS \ - --policy.train_expert_only=$TRAIN_EXPERT_ONLY \ - --policy.vlm_model_name=$VLM \ - --policy.n_obs_steps=$N_OBS_STEPS \ - --policy.attention_mode=$ATTN_MODE \ - --policy.prefix_length=$PREFIX_LENGTH \ - --policy.num_vlm_layers=$NUM_VLM_LAYERS \ - --policy.chunk_size=$CHUNK_SIZE \ - --policy.load_vlm_weights=$LOAD_VLM_WEIGHTS \ - --policy.expert_width_multiplier=$EXPERT_WIDTH_MULTIPLIER \ - --policy.self_attn_every_n_layers=$SELF_ATTN_EVERY_N_LAYERS \ - --policy.max_action_dim=$MAX_ACTION_DIM \ - --policy.max_state_dim=$MAX_STATE_DIM \ - --seed=$SEED \ - --wandb.enable=false diff --git a/examples/9_evaluate_must.sh b/examples/9_evaluate_must.sh deleted file mode 100644 index 534153330..000000000 --- a/examples/9_evaluate_must.sh +++ /dev/null @@ -1,2811 +0,0 @@ -#!/bin/bash - -#SBATCH --job-name=lerobot_eval_smolpi0_libero_eval10ep_ca_sa2_16vlm_w075_smolvlm2b_lr7e5 -#SBATCH --nodes=1 -#SBATCH --ntasks=1 -#SBATCH --gpus-per-node=1 -#SBATCH --mail-type=END,FAIL -#SBATCH --output=/lustre/fswork/projects/rech/dyf/ugz83ue/logs/slurm/lerobot_eval_smolpi0_libero_eval10ep_ca_sa2_16vlm_w075_smolvlm2b_lr7e5.out -###SBATCH --nodelist=jean-zay-a101 -#SBATCH --cpus-per-task=45 -###SBATCH --exclusive -#SBATCH --time=15:00:00 -#SBATCH --mail-user=mustafa.shukor@isir.upmc.fr - -##SBATCH --partition=gpu_p2 -##SBATCH --qos=qos_gpu-t3 -###SBATCH -C v100-32g -##SBATCH -A dyf@v100 - -##SBATCH --partition=gpu_p5 -##SBATCH -C a100 -###SBATCH -A dyf@a100 -##SBATCH -A lqm@a100 -##SBATCH --qos=qos_gpu_a100-dev -##SBATCH --qos=qos_gpu_a100-t3 - -#SBATCH --partition=gpu_p6 -#SBATCH -C h100 -#SBATCH -A lqm@h100 -###SBATCH --qos=qos_gpu_h100-dev -#SBATCH --qos=qos_gpu_h100-t3 - -###SBATCH --begin=now+2hour - -# cd ~/lerobot_pi -# source ~/.bashrc -# source activate lerobot -# export LC_ALL=C - -# rm core-* -export CUDA_VISIBLE_DEVICES=3 -# storage / caches -RAID=/raid/jade -export TRANSFORMERS_CACHE=$RAID/.cache/huggingface/transformers -export HF_HOME=$RAID/.cache/huggingface -export HF_DATASETS_CACHE=$RAID/.cache/huggingface/datasets -export HF_LEROBOT_HOME=$RAID/.cache/huggingface/lerobot -export WANDB_CACHE_DIR=$RAID/.cache/wandb -export TMPDIR=$RAID/.cache/tmp -mkdir -p $TMPDIR -export WANDB_MODE=offline -export HF_DATASETS_OFFLINE=1 -# export HF_HUB_OFFLINE=1 -export TOKENIZERS_PARALLELISM=false -export MUJOCO_GL=egl -export CUDA_VISIBLE_DEVICES=3 - -PORT=29512 - -## then later -## wandb sync wandb/offline-run-* - - -ENV=libero - -# TASK=libero_10 -TASK=libero_spatial -# TASK=libero_spatial -# TASK=libero_10 -# TASK=libero_spatial - - -POLICY_NAME=smolpi0 - -POLICY=smolpi0 -ENV=libero - - - - - - -CKPT_KEYS_MAPPING=model._orig_mod.//model. -LOAD_VLM_WEIGHTS=true -PEFT_METHOD=freeze -SELF_ATTN_ONLY_ACTIONS=false -CAUSAL_ATTENTION_ON_HISTORY=false - -PREDICT_RELATIVE_ACTIONS=false -RELATIVE_ACTIONS_MODE=first -SHUFFLE_CAMERA_POSITIONS=false - -VLM_IMG_SIZE=-1 -REGRESSION_LOSS=false - - -# ## Baseline for ablation study -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=max_length -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4_bs8_steps100000_gpus2_freeze32_onlyexpert_1act_promptfalse_imgtoktrue_nobs1_compiletrue_cross_attn_pref0_gap1_localimgfalse_reverseimgorderfalse_statetopreftrue/checkpoints/last/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-500M-Video-Instruct - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=max_length -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4_bs8_steps100000_gpus2_freeze32_onlyexpert_1act_promptfalse_imgtoktrue_nobs1_compiletrue_cross_attn/checkpoints/last/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=false -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-500M-Video-Instruct - - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=max_length -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4_bs8_steps100000_gpus2_freeze32_onlyexpert_1act_promptfalse_imgtoktrue_nobs1_compiletrue_self_attn/checkpoints/last/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=self_attn -# STATE_TO_PREFIX=false -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-500M-Video-Instruct - - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_self_attn_gap1_localimgfalse_statetopreffalse_explay0_vlml0_causalacttrue_sa0/checkpoints/last/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=self_attn -# STATE_TO_PREFIX=false -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-500M-Video-Instruct - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr5e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalactfalse_sa0_smolvlm2250/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-256M-Video-Instruct - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr8e-5bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalactfalse_sa0_smolvlm22b/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-2.2B-Instruct - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4_bs8_steps100000_gpus2_freeze32_onlyexpert_1act_promptfalse_imgtoktrue_nobs1_compiletrue_self_attn_pref0_gap1_localimgfalse_reverseimgorderfalse_statetopreftrue_toklongest_explay0_vlml0_causalacttrue/checkpoints/last/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=self_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-500M-Video-Instruct - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalacttrue_sa2/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM2-500M-Video-Instruct - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4_bs8_steps100000_gpus2_freeze32_onlyexpert_1act_promptfalse_imgtoktrue_nobs1_compiletrue_cross_attn_pref0_gap1_localimgfalse_reverseimgorderfalse_statetopreffalse/checkpoints/last/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=false -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-500M-Video-Instruct - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_self_attn_gap1_localimgfalse_statetopreffalse_explay0_vlml0_causalacttrue_sa0_smolvlm2500/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=self_attn -# STATE_TO_PREFIX=false -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-500M-Video-Instruct - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_self_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalactfalse_sa0_smolvlm2500/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=self_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-500M-Video-Instruct - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=8 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml8_causalactfalse_sa0/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-500M-Video-Instruct - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalactfalse_sa0/checkpoints/last/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-500M-Video-Instruct - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=24 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml24_causalactfalse_sa0/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-500M-Video-Instruct - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtokfalse_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalactfalse_sa0_smolvlm2500/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=false -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-500M-Video-Instruct - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=100 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalactfalse_sa0_smolvlm2500_chunk100/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-500M-Video-Instruct - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=30 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalactfalse_sa0_smolvlm2500_chunk30/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-500M-Video-Instruct - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=10 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalactfalse_sa0_smolvlm2500_chunk10/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-500M-Video-Instruct - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=1 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalactfalse_sa0_smolvlm2500_chunk1/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-500M-Video-Instruct - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=16 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay16_vlml0_causalactfalse_sa0_smolvlm2500_chunk50_nobs1/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-500M-Video-Instruct - - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=2 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalactfalse_sa0_smolvlm2500_chunk50_nobs2/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-500M-Video-Instruct - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=3 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4_bs8_steps100000_gpus2_freeze32_onlyexpert_1act_promptfalse_imgtoktrue_nobs3_compiletrue_cross_attn_pref0_gap1_localimgfalse_reverseimgorderfalse_statetopreftrue_toklongest/checkpoints/last/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-500M-Video-Instruct - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="observation.state" -# N_OBS_STEPS=3 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalactfalse_sa0_smolvlm2500_chunk50_nobs3_paststates/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-500M-Video-Instruct - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="observation.state,image" -# N_OBS_STEPS=3 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalactfalse_sa0_smolvlm2500_chunk50_nobs3_paststatesimgs/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-500M-Video-Instruct - - -# EXPERT_WIDTH_MULTIPLIER=1 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr9.5e-5bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalactfalse_sa0_smolvlm2500_chunk50_nobs1_expw1/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-500M-Video-Instruct - - -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr9.5e-5bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalactfalse_sa0_smolvlm2500_chunk50_nobs1_expw0.75/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-500M-Video-Instruct - - -# EXPERT_WIDTH_MULTIPLIER=0.25 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr2e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalactfalse_sa0_smolvlm2500_chunk50_nobs1_expw0.25/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-500M-Video-Instruct - - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="observation.state,image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtokfalse_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalactfalse_sa0_smolvlm2500_chunk50_nobs1_expw0.5/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=false -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-500M-Video-Instruct - - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="observation.state,image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs16steps100000gpus2freeze32_imgtokfalse_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalactfalse_sa0_smolvlm2500_chunk50_nobs1_expw0.5/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=false -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-500M-Video-Instruct - - - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="observation.state,image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_imgtokfalse_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalactfalse_sa0_smolvlm2500_chunk50_nobs1_expw0.5/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=false -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM2-500M-Video-Instruct - - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr5e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalactfalse_sa0_smolvlm1250_chunk50_nobs1_expw0.5/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-256M-Instruct - - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr8e-5bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalactfalse_sa0_smolvlm12b_chunk50_nobs1_expw0.5/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-Instruct - - - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalactfalse_sa0_smolvlm1500_chunk50_nobs1_expw0.5/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr8e-5bs8steps100000gpus2freeze32_imgtokfalse_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalactfalse_sa0_smolvlm1500_chunk50_nobs1_expw0.5/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=false -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr8e-5bs8steps100000gpus2freeze32_imgtokfalse_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalactfalse_sa0_smolvlm1500_chunk50_nobs1_expw0.5/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=false -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalactfalse_sa0_smolvlm2500_chunk50_nobs1_expw0.5/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.5/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.5/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalactfalse_sa0_smolvlm2500_chunk50_nobs1_expw0.5/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalactfalse_sa0_smolvlm2500_chunk50_nobs1_expw0.5_rep/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.5/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalactfalse_sa2_smolvlm2500_chunk50_nobs1_expw0.5_rep/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_imgtokfalse_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.5/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=false -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2full8_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalacttrue_sa0_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-6/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2full8_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalacttrue_sa0_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-5/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2full8_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalacttrue_sa0_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# PEFT_METHOD=lora -# PEFT_TARGET_MODEL=text -# LORA_TARGET_MODULES=q_proj,v_proj,k_proj -# LORA_R=32 -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2lora32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalacttrue_sa0_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4_loraqkv/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# PEFT_METHOD=lora -# PEFT_TARGET_MODEL=text -# LORA_TARGET_MODULES=q_proj,v_proj,k_proj -# LORA_R=32 -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2lora32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalacttrue_sa0_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-5_loraqkv/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# PEFT_METHOD=lora -# PEFT_TARGET_MODEL=text -# LORA_TARGET_MODULES=q_proj,v_proj,k_proj,up_proj,down_proj,gate_proj -# LORA_R=32 -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2lora32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalacttrue_sa0_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# PEFT_METHOD=lora -# PEFT_TARGET_MODEL=text -# LORA_TARGET_MODULES=q_proj,v_proj,k_proj,up_proj,down_proj,gate_proj -# LORA_R=32 -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2lora32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalacttrue_sa0_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-5/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# PEFT_METHOD=lora -# PEFT_TARGET_MODEL=text -# LORA_TARGET_MODULES=q_proj,v_proj,k_proj,up_proj,down_proj,gate_proj -# LORA_R=32 -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2lora32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalacttrue_sa0_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-6/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2lora32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalacttrue_sa0_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-5/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_imgtokfalse_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa0_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4_loraqkv/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=false -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_imgtokfalse_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=false -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs16steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs16steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs16steps100000gpus2freeze32_imgtokfalse_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=false -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - - -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtokfalse_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalacttrue_sa0_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=false -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# SELF_ATTN_ONLY_ACTIONS=true -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4_saacttrue/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=max_length -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4_saactfalse_droptrue_max_length/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=max_length -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4_saactfalse_dropfalse_max_length/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=max_length -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr9.5e-5bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.75_lrvlm1e-4_saactfalse_dropfalse_max_length/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr9.5e-5bs32steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.75_lrvlm1e-4_saactfalse_dropfalse_longest/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.75_lrvlm1e-4_saactfalse_dropfalse_longest/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4_longest_ptdroidfull/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4_longest_ptcomv3freeze/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4_longest_ptcomv1v2full/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4_longest_ptcomv1v2freeze/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr2e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4_longest_pt_trans1true/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr2e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4_longest_pt_trans3true/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_imgtoktrue_self_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalacttrue_sa0_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4_longest_pt_trans1false/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=self_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_imgtoktrue_self_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalactfalse_sa0_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4_longest_pt_trans1false/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=self_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=false -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4_longest_pt_trans1false/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalacttrue_sa0_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4_longest_pt_trans1false/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_ptcomv3freeze25_trans1false/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_ptcomv3freeze50_trans1false/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_ptcomv3freeze75_trans1false/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_ptcomv3freeze100_trans1false/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4_longest_pt_trans6true/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4_longest_pt_trans4true/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4_longest_pt_trans7true/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4_longest_pt_trans5true/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4_longest_pt_trans2true/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans1true/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4_longest_pt_trans8true/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4_longest_pt_trans9true/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4_longest_ptcomv1v2full/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.25 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalacttrue_sa0_smolvlm2500_chunk50_nobs1_expw0.25_lrvlm1e-4_longest_pt_trans0false/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=1 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalacttrue_sa0_smolvlm2500_chunk50_nobs1_expw1_lrvlm1e-4_longest_pt_trans0false/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalacttrue_sa0_smolvlm2500_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans0false/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalacttrue_sa0_smolvlm2500_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans0false/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.25 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalacttrue_sa0_smolvlm2500_chunk50_nobs1_expw0.25_lrvlm1e-4_longest_pt_trans0false/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=1 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml0_causalacttrue_sa0_smolvlm2500_chunk50_nobs1_expw1_lrvlm1e-4_longest_pt_trans0false/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="observation.state" -# N_OBS_STEPS=3 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs3statestrue_expw0.75_lrvlm1e-4_longest_pt_trans0false/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image,observation.state" -# N_OBS_STEPS=3 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs3_expw0.75_lrvlm1e-4_longest_pt_trans0false/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image,observation.state" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1statestrue_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr1e-5100000/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image,observation.state" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1statestrue_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr1e-530000/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image,observation.state" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1statestrue_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr5e-6100000/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1statestrue_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-6100000/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1statestrue_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1statestrue_expw0.75_lrvlm1e-4_longest_pt_trans0true_decaylr2.5e-630000/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_imgtoktrue_cross_attn_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1statestrue_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr1e-5200000/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_imgtoktrue_cross_attn_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1statestrue_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr5e-6200000/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_imgtoktrue_cross_attn_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1statestrue_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr1e-530000/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_imgtoktrue_cross_attn_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1statestrue_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-6200000/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_imgtoktrue_cross_attn_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1statestrue_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-6100000/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_imgtoktrue_cross_attn_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1statestrue_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_imgtoktrue_cross_attn_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1statestrue_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr5e-6100000/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_vlml16_causalacttrue_sa2_smolvla500base_chunk50_nobs1statestrue_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relactfalsefirst_camfalse/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLA-500M-Base - - - -# PREDICT_RELATIVE_ACTIONS=true -# RELATIVE_ACTIONS_MODE=relative -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_imgtoktrue_cross_attn_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1statestrue_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relacttruerelative/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - - -# PREDICT_RELATIVE_ACTIONS=true -# RELATIVE_ACTIONS_MODE=first -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_imgtoktrue_cross_attn_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1statestrue_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relacttruefirst/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_imgtoktrue_cross_attn_vlml16_causalacttrue_sa2_smolvla500base_chunk50_nobs1statestrue_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relactfalsefirst_camfalse/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLA-500M-Base - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr8e-5bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1statestrue_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-6100000_relactfalsefirst_camfalse/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr6e-5bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1statestrue_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-6100000_relactfalsefirst_camfalse/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr4e-5bs8steps100000gpus2freeze32_imgtoktrue_cross_attn_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1statestrue_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-6100000_relactfalsefirst_camfalse/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_cross_attn_vlml16_sa2_smolvlm2500_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_ptcomv1v2freezebs64transv0_trans0false_decaylr2.5e-630000_relactfalsefirst_camfalse_vim-1/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_cross_attn_vlml16_sa2_smolvlm2500_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans1true_decaylr2.5e-630000_relactfalsefirst_camfalse_vim-1/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# REGRESSION_LOSS=true -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs8steps100000gpus2freeze32_cross_attn_vlml0_sa0_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relactfalsefirst_camfalse_vim-1_regtrue/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# REGRESSION_LOSS=true -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_cross_attn_vlml16_sa2_smolvlm2500_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relactfalsefirst_camfalse_vim-1_regtrue/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_cross_attn_vlml16_sa2_smolvlm2500_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-6100000_relactfalsefirst_camfalse_vim-1/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr2e-4bs32steps100000gpus2freeze32_cross_attn_vlml16_sa2_smolvlm2500_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-6100000_relactfalsefirst_camfalse_vim-1/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr3e-4bs32steps100000gpus2freeze32_cross_attn_vlml16_sa2_smolvlm2500_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-6100000_relactfalsefirst_camfalse_vim-1/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# REGRESSION_LOSS=true -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr9e-5bs8steps100000gpus2freeze32_cross_attn_vlml0_sa0_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relactfalsefirst_camfalse_vim-1_regtrue/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - - -# REGRESSION_LOSS=true -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.5 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=0 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr2e-4bs8steps100000gpus2freeze32_cross_attn_vlml0_sa0_smolvlm2500_chunk50_nobs1_expw0.5_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relactfalsefirst_camfalse_vim-1_regtrue/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=0 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr9e-5bs32steps100000gpus2freeze32_cross_attn_vlml16_sa2_smolvlm2500_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-6100000_relactfalsefirst_camfalse_vim-1_regfalse/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr5e-4bs8steps100000gpus2freeze32_cross_attn_vlml16_sa2_smolvlm2250_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relactfalsefirst_camfalse_vim-1_regfalse/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM2-256M-Video-Instruct - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr8e-5bs8steps100000gpus2freeze32_cross_attn_vlml16_sa2_smolvlm22b_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relactfalsefirst_camfalse_vim-1_regfalse/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM2-2.2B-Instruct - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr5e-4bs32steps100000gpus2freeze32_cross_attn_vlml16_sa2_smolvlm1250_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relactfalsefirst_camfalse_vim-1_regfalse/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM2-256M-Video-Instruct - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr4e-4bs32steps100000gpus2freeze32_cross_attn_vlml16_sa2_smolvlm1250_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relactfalsefirst_camfalse_vim-1_regfalse/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM2-256M-Video-Instruct - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr3e-4bs32steps100000gpus2freeze32_cross_attn_vlml16_sa2_smolvlm1250_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relactfalsefirst_camfalse_vim-1_regfalse/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM2-256M-Video-Instruct - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr6e-5bs32steps100000gpus2freeze32_cross_attn_vlml16_sa2_smolvlm12b_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relactfalsefirst_camfalse_vim-1_regfalse/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM2-2.2B-Instruct - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_cross_attn_vlml16_sa2_smolvlm12b_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relactfalsefirst_camfalse_vim-1_regfalse/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM2-2.2B-Instruct - - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr9e-5bs32steps100000gpus2freeze32_cross_attn_vlml16_sa2_smolvlm12b_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relactfalsefirst_camfalse_vim-1_regfalse/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM2-2.2B-Instruct - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr7e-5bs32steps100000gpus2freeze32_cross_attn_vlml16_sa2_smolvlm12b_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relactfalsefirst_camfalse_vim-1_regfalse/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM2-2.2B-Instruct - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr8e-5bs32steps100000gpus2freeze32_cross_attn_vlml16_sa2_smolvlm12b_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relactfalsefirst_camfalse_vim-1_regfalse/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM2-2.2B-Instruct - - -CAUSAL_ATTENTION_ON_HISTORY=true -SELF_ATTN_ONLY_ACTIONS=false -EXPERT_WIDTH_MULTIPLIER=0.75 -PAST_OBS_KEYS="image" -N_OBS_STEPS=1 -NUM_EXPERT_LAYERS=0 -CHUNK_SIZE=50 -NUM_VLM_LAYERS=16 -PAD_LANG_TO=longest -EVAL_CKPT=/raid/jade/models/smolvlamust -ADD_IMAGE_TOKENS=true -ATTN_MODE=cross_attn -STATE_TO_PREFIX=true -CAUSAL_ACTION_ATTENTION_MASK=true -SELF_ATTN_EVERY_N_LAYERS=2 -VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr6e-5bs32steps100000gpus2freeze32_cross_attn_vlml16_sa2_smolvlm22b_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relactfalsefirst_camfalse_vim-1_regfalse_compilefalse/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM2-2.2B-Instruct - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr9e-5bs32steps100000gpus2freeze32_cross_attn_vlml16_sa2_smolvlm22b_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relactfalsefirst_camfalse_vim-1_regfalse_compilefalse/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM2-2.2B-Instruct - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr8e-5bs32steps100000gpus2freeze32_cross_attn_vlml16_sa2_smolvlm22b_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relactfalsefirst_camfalse_vim-1_regfalse_compilefalse/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM2-2.2B-Instruct - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_cross_attn_vlml16_sa2_smolvlm22b_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relactfalsefirst_camfalse_vim-1_regfalse_compilefalse/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM2-2.2B-Instruct - - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr7e-4bs32steps100000gpus2freeze32_cross_attn_vlml16_sa2_smolvlm1250_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relactfalsefirst_camfalse_vim-1_regfalse/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM2-256M-Video-Instruct - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr1e-4bs32steps100000gpus2freeze32_cross_attn_vlml16_sa2_smolvlm2250_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relactfalsefirst_camfalse_vim-1_regfalse_compilefalse/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM2-256M-Video-Instruct - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr3e-4bs32steps100000gpus2freeze32_cross_attn_vlml16_sa2_smolvlm2250_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relactfalsefirst_camfalse_vim-1_regfalse_compilefalse/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM2-256M-Video-Instruct - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr5e-4bs32steps100000gpus2freeze32_cross_attn_vlml16_sa2_smolvlm2250_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relactfalsefirst_camfalse_vim-1_regfalse_compilefalse/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM2-256M-Video-Instruct - - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr7e-4bs32steps100000gpus2freeze32_cross_attn_vlml16_sa2_smolvlm2250_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relactfalsefirst_camfalse_vim-1_regfalse_compilefalse/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM2-256M-Video-Instruct - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr6e-4bs32steps100000gpus2freeze32_cross_attn_vlml16_sa2_smolvlm2250_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relactfalsefirst_camfalse_vim-1_regfalse_compilefalse/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM2-256M-Video-Instruct - -# CAUSAL_ATTENTION_ON_HISTORY=true -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=$WORK/logs/lerobot/lerobot_physical_intelligence_libero_smolpi0_lr4e-4bs32steps100000gpus2freeze32_cross_attn_vlml16_sa2_smolvlm2250_chunk50_nobs1_expw0.75_lrvlm1e-4_longest_pt_trans0false_decaylr2.5e-630000_relactfalsefirst_camfalse_vim-1_regfalse_compilefalse/checkpoints/best/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM2-256M-Video-Instruct - - -# # TASK=libero_spatial -# MULTITASK_EVAL=false -# N_EPISODES=50 - -# TASK=libero_spatial,libero_object,libero_goal,libero_10 -MULTITASK_EVAL=true -# N_EPISODES=5 -N_EPISODES=1 - -# MAX_PARRALLEL_TASKS=5 -# MAX_PARRALLEL_TASKS=2 -MAX_PARRALLEL_TASKS=1 - -# NUM_EVALS=2 -# SEEDS=(1000 5000) -SEEDS=(5000) -# ACTION_STEPS_LIST=(1 10 30 50) -ACTION_STEPS_LIST=(1) -# ACTION_STEPS_LIST=(50) -TASK_LIST=(libero_spatial libero_object libero_goal libero_10) -TASK_LIST=(libero_spatial) -for SEED in "${SEEDS[@]}"; do - for N_ACTION_STEPS in "${ACTION_STEPS_LIST[@]}"; do - for TASK in "${TASK_LIST[@]}"; do - echo "$TASK Evaluating: $EVAL_CKPT | N_ACTION_STEPS=$N_ACTION_STEPS | EVAL SEED=$SEED" - python src/lerobot/scripts/eval.py \ - --output_dir=/raid/jade/logs/lerobot/tmp \ - --env.type=$ENV \ - --env.task=$TASK \ - --eval.batch_size=$N_EPISODES \ - --eval.n_episodes=$N_EPISODES \ - --seed=$SEED \ - --policy.use_amp=false \ - --policy.path=$EVAL_CKPT \ - --policy.n_action_steps=$N_ACTION_STEPS \ - --policy.checkpoint_path=$EVAL_CKPT \ - --env.multitask_eval=$MULTITASK_EVAL --env.max_parallel_tasks=$MAX_PARRALLEL_TASKS \ - --policy.add_image_special_tokens=$ADD_IMAGE_TOKENS \ - --policy.attention_mode=$ATTN_MODE \ - --policy.causal_action_attention_mask=$CAUSAL_ACTION_ATTENTION_MASK \ - --policy.state_to_prefix=$STATE_TO_PREFIX \ - --policy.self_attn_every_n_layers=$SELF_ATTN_EVERY_N_LAYERS \ - --policy.pad_language_to=$PAD_LANG_TO \ - --policy.load_vlm_weights=$LOAD_VLM_WEIGHTS \ - --policy.vlm_model_name=$VLM_NAME \ - --policy.num_vlm_layers=$NUM_VLM_LAYERS \ - --policy.chunk_size=$CHUNK_SIZE \ - --policy.n_obs_steps=$N_OBS_STEPS \ - --policy.past_obs_keys=$PAST_OBS_KEYS \ - --policy.num_expert_layers=$NUM_EXPERT_LAYERS \ - --policy.expert_width_multiplier=$EXPERT_WIDTH_MULTIPLIER \ - --policy.peft_method=$PEFT_METHOD \ - --policy.self_attn_only_actions=$SELF_ATTN_ONLY_ACTIONS \ - --policy.causal_attention_on_history=$CAUSAL_ATTENTION_ON_HISTORY \ - --policy.predict_relative_actions=$PREDICT_RELATIVE_ACTIONS --policy.relative_actions_mode=$RELATIVE_ACTIONS_MODE --policy.shuffle_camera_positions=$SHUFFLE_CAMERA_POSITIONS \ - --policy.vlm_img_size=$VLM_IMG_SIZE \ - --policy.regression_loss=$REGRESSION_LOSS - # --policy.peft_config.r=$LORA_R --policy.peft_config.target_modules=$LORA_TARGET_MODULES --policy.peft_method=$PEFT_METHOD --policy.peft_target_model=$PEFT_TARGET_MODEL - - echo "Done with: $EVAL_CKPT | Steps=$N_ACTION_STEPS | EVAL SEED=$SEED" - echo "------------------------------------------------------" - done - done -done - - -# ############################################################################################################################################ -# ############################################################################################################################################ -# ############################################################################################################################################ -# ########### Offline eval - - -# # ############################ -# # # Community datasets V1 -# # # REPO_ID=pranavsaroha/so100_legos4,pranavsaroha/so100_onelego2,jpata/so100_pick_place_tangerine,pranavsaroha/so100_onelego3,pranavsaroha/so100_carrot_2,pranavsaroha/so100_carrot_5,pandaRQ/pick_med_1,HITHY/so100_strawberry,vladfatu/so100_above,koenvanwijk/orange50-1,koenvanwijk/orange50-variation-2,FeiYjf/new_GtoR,CSCSXX/pick_place_cube_1.18,vladfatu/so100_office,dragon-95/so100_sorting,dragon-95/so100_sorting_1,nbaron99/so100_pick_and_place4,Beegbrain/pick_place_green_block,Ityl/so100_recording2,dragon-95/so100_sorting_2,dragon-95/so100_sorting_3,aractingi/push_cube_offline_data,HITHY/so100_peach3,HITHY/so100_peach4,shreyasgite/so100_legocube_50,shreyasgite/so100_base_env,triton7777/so100_dataset_mix,Deason11/Open_the_drawer_to_place_items,Deason11/PLACE_TAPE_PUSH_DRAWER,NONHUMAN-RESEARCH/SOARM100_TASK_VENDA,mikechambers/block_cup_14,samsam0510/tooth_extraction_3,samsam0510/tooth_extraction_4,samsam0510/cube_reorientation_2,samsam0510/cube_reorientation_4,samsam0510/glove_reorientation_1,DorayakiLin/so100_pick_charger_on_tissue,zijian2022/noticehuman3,liuhuanjim013/so100_th -# # # Inconsistent actions dim: Deason11/Open_the_drawer_to_place_items, Deason11/PLACE_TAPE_PUSH_DRAWER -# # # Filtered datasets -# # REPO_ID=pranavsaroha/so100_onelego2,pranavsaroha/so100_onelego3,pranavsaroha/so100_carrot_2,vladfatu/so100_above,koenvanwijk/orange50-1,CSCSXX/pick_place_cube_1.18,dragon-95/so100_sorting,dragon-95/so100_sorting_1,nbaron99/so100_pick_and_place4,Beegbrain/pick_place_green_block,dragon-95/so100_sorting_3,HITHY/so100_peach3,shreyasgite/so100_legocube_50,triton7777/so100_dataset_mix,NONHUMAN-RESEARCH/SOARM100_TASK_VENDA,mikechambers/block_cup_14,samsam0510/tooth_extraction_3,samsam0510/tooth_extraction_4,samsam0510/cube_reorientation_2,samsam0510/cube_reorientation_4,samsam0510/glove_reorientation_1,vladfatu/so100_office,pranavsaroha/so100_legos4,Ityl/so100_recording2,FeiYjf/new_GtoR,dragon-95/so100_sorting_2,HITHY/so100_peach4,jpata/so100_pick_place_tangerine,HITHY/so100_strawberry,shreyasgite/so100_base_env,koenvanwijk/orange50-variation-2,pranavsaroha/so100_carrot_5,pandaRQ/pick_med_1,aractingi/push_cube_offline_data,DorayakiLin/so100_pick_charger_on_tissue,zijian2022/noticehuman3,liuhuanjim013/so100_th -# # SAMPLING_WEIGHTS= -# # DATASET_NAME=so100_community_v1 - - -# # # Community datasets V2 -# # # Inconsistent actions: 1g0rrr/sam_openpi_solder1, 1g0rrr/sam_openpi03, 1g0rrr/sam_openpi_solder2 -# # # Other issues: pierfabre/rabbit bensprenger/right_arm_p_brick_in_box_with_y_noise_v0 pierfabre/horse pierfabre/pig2 pierfabre/pig3 pierfabre/cow2,pierfabre/sheep -# # # REPO_ID=Chojins/chess_game_009_white,sihyun77/suho_3_17_1,sihyun77/sihyun_3_17_2,sihyun77/suho_3_17_3,sihyun77/sihyun_3_17_5,Odog16/so100_cube_drop_pick_v1,sihyun77/sihyun_main_2,sihyun77/suho_main_2,Bartm3/dice2,sihyun77/sihyun_main_3,Loki0929/so100_duck,pietroom/holdthis,pietroom/actualeasytask,Beegbrain/pick_lemon_and_drop_in_bowl,Beegbrain/sweep_tissue_cube,zijian2022/321,gxy1111/so100_pick_place,Odog16/so100_cube_stacking_v1,sihyun77/mond_1,andlyu/so100_indoor_1,andlyu/so100_indoor_3,frk2/so100large,lirislab/sweep_tissue_cube,lirislab/lemon_into_bowl,lirislab/red_cube_into_green_lego_block,lirislab/red_cube_into_blue_cube,00ri/so100_battery,frk2/so100largediffcam,FsqZ/so100_1,ZGGZZG/so100_drop0,Chojins/chess_game_000_white_red,smanni/train_so100_fluffy_box,ganker5/so100_push_20250328,ganker5/so100_dataline_0328,ganker5/so100_color_0328,CrazyYhang/A1234-B-C_mvA2B,RasmusP/so100_Orange2Green,sixpigs1/so100_pick_cube_in_box,ganker5/so100_push_20250331,ganker5/so100_dataline_20250331,lirislab/put_caps_into_teabox,lirislab/close_top_drawer_teabox,lirislab/open_top_drawer_teabox,lirislab/unfold_bottom_right,lirislab/push_cup_target,lirislab/put_banana_bowl,Chojins/chess_game_001_blue_stereo,Chojins/chess_game_001_red_stereo,ganker5/so100_toy_20250402,Gano007/so100_medic,00ri/so100_battery_bin_center,paszea/so100_whale_2,lirislab/fold_bottom_right,lirislab/put_coffee_cap_teabox,therarelab/so100_pick_place_2,paszea/so100_whale_3,paszea/so100_whale_4,paszea/so100_lego,LemonadeDai/so100_coca,zijian2022/backgrounda,zijian2022/backgroundb,356c/so100_nut_sort_1,Mwuqiu/so100_0408_muti,aimihat/so100_tape,lirislab/so100_demo,356c/so100_duck_reposition_1,zijian2022/sort1,weiye11/so100_410_zwy,VoicAndrei/so100_banana_to_plate_only,sixpigs1/so100_stack_cube_error,isadev/bougies3,zijian2022/close3,bensprenger/left_arm_yellow_brick_in_box_v0,lirislab/guess_who_so100,bensprenger/left_arm_yellow_brick_in_box_with_purple_noise_v0,roboticshack/team16-can-stacking,zijian2022/insert2,roboticshack/team-7-right-arm-grasp-tape,Jiangeng/so100_413,roboticshack/team9-pick_cube_place_static_plate,AndrejOrsula/lerobot_double_ball_stacking_random,roboticshack/left-arm-grasp-lego-brick,roboticshack/team-7-left-arm-grasp-motor,roboticshack/team9-pick_chicken_place_plate,roboticshack/team13-two-balls-stacking,tkc79/so100_lego_box_1,roboticshack/team13-three-balls-stacking,pierfabre/chicken,roboticshack/team16-water-pouring,ad330/cubePlace,Jiafei1224/so100_pa222per,paszea/so100_lego_2cam,bensprenger/chess_game_001_blue_stereo,Mohamedal/put_banana,tkc79/so100_lego_box_2,samanthalhy/so100_herding_1,jlesein/TestBoulon7 -# # REPO_ID=pierfabre/rabbit,bensprenger/right_arm_p_brick_in_box_with_y_noise_v0,pierfabre/horse,pierfabre/pig2,pierfabre/pig3,pierfabre/cow2,pierfabre/sheep,Chojins/chess_game_009_white,sihyun77/suho_3_17_1,sihyun77/sihyun_3_17_2,sihyun77/suho_3_17_3,sihyun77/sihyun_3_17_5,Odog16/so100_cube_drop_pick_v1,sihyun77/sihyun_main_2,sihyun77/suho_main_2,Bartm3/dice2,sihyun77/sihyun_main_3,Loki0929/so100_duck,pietroom/holdthis,pietroom/actualeasytask,Beegbrain/pick_lemon_and_drop_in_bowl,Beegbrain/sweep_tissue_cube,zijian2022/321,gxy1111/so100_pick_place,Odog16/so100_cube_stacking_v1,sihyun77/mond_1,andlyu/so100_indoor_1,andlyu/so100_indoor_3,frk2/so100large,lirislab/sweep_tissue_cube,lirislab/lemon_into_bowl,lirislab/red_cube_into_green_lego_block,lirislab/red_cube_into_blue_cube,00ri/so100_battery,frk2/so100largediffcam,FsqZ/so100_1,ZGGZZG/so100_drop0,Chojins/chess_game_000_white_red,smanni/train_so100_fluffy_box,ganker5/so100_push_20250328,ganker5/so100_dataline_0328,ganker5/so100_color_0328,CrazyYhang/A1234-B-C_mvA2B,RasmusP/so100_Orange2Green,sixpigs1/so100_pick_cube_in_box,ganker5/so100_push_20250331,ganker5/so100_dataline_20250331,lirislab/put_caps_into_teabox,lirislab/close_top_drawer_teabox,lirislab/open_top_drawer_teabox,lirislab/unfold_bottom_right,lirislab/push_cup_target,lirislab/put_banana_bowl,Chojins/chess_game_001_blue_stereo,Chojins/chess_game_001_red_stereo,ganker5/so100_toy_20250402,Gano007/so100_medic,00ri/so100_battery_bin_center,paszea/so100_whale_2,lirislab/fold_bottom_right,lirislab/put_coffee_cap_teabox,therarelab/so100_pick_place_2,paszea/so100_whale_3,paszea/so100_whale_4,paszea/so100_lego,LemonadeDai/so100_coca,zijian2022/backgrounda,zijian2022/backgroundb,356c/so100_nut_sort_1,Mwuqiu/so100_0408_muti,aimihat/so100_tape,lirislab/so100_demo,356c/so100_duck_reposition_1,zijian2022/sort1,weiye11/so100_410_zwy,VoicAndrei/so100_banana_to_plate_only,sixpigs1/so100_stack_cube_error,isadev/bougies3,zijian2022/close3,bensprenger/left_arm_yellow_brick_in_box_v0,lirislab/guess_who_so100,bensprenger/left_arm_yellow_brick_in_box_with_purple_noise_v0,roboticshack/team16-can-stacking,zijian2022/insert2,roboticshack/team-7-right-arm-grasp-tape,Jiangeng/so100_413,roboticshack/team9-pick_cube_place_static_plate,AndrejOrsula/lerobot_double_ball_stacking_random,roboticshack/left-arm-grasp-lego-brick,roboticshack/team-7-left-arm-grasp-motor,roboticshack/team9-pick_chicken_place_plate,roboticshack/team13-two-balls-stacking,tkc79/so100_lego_box_1,roboticshack/team13-three-balls-stacking,pierfabre/chicken,roboticshack/team16-water-pouring,ad330/cubePlace,Jiafei1224/so100_pa222per,paszea/so100_lego_2cam,bensprenger/chess_game_001_blue_stereo,Mohamedal/put_banana,tkc79/so100_lego_box_2,samanthalhy/so100_herding_1,jlesein/TestBoulon7 -# # SAMPLING_WEIGHTS= -# # DATASET_NAME=so100_community_v2 - -# # Community datasets V1+V2 -# # REPO_ID=pierfabre/rabbit,bensprenger/right_arm_p_brick_in_box_with_y_noise_v0,pierfabre/horse,pierfabre/pig2,pierfabre/pig3,pierfabre/cow2,pierfabre/sheep,Chojins/chess_game_009_white,sihyun77/suho_3_17_1,sihyun77/sihyun_3_17_2,sihyun77/suho_3_17_3,sihyun77/sihyun_3_17_5,Odog16/so100_cube_drop_pick_v1,sihyun77/sihyun_main_2,sihyun77/suho_main_2,Bartm3/dice2,sihyun77/sihyun_main_3,Loki0929/so100_duck,pietroom/holdthis,pietroom/actualeasytask,Beegbrain/pick_lemon_and_drop_in_bowl,Beegbrain/sweep_tissue_cube,zijian2022/321,gxy1111/so100_pick_place,Odog16/so100_cube_stacking_v1,sihyun77/mond_1,andlyu/so100_indoor_1,andlyu/so100_indoor_3,frk2/so100large,lirislab/sweep_tissue_cube,lirislab/lemon_into_bowl,lirislab/red_cube_into_green_lego_block,lirislab/red_cube_into_blue_cube,00ri/so100_battery,frk2/so100largediffcam,FsqZ/so100_1,ZGGZZG/so100_drop0,Chojins/chess_game_000_white_red,smanni/train_so100_fluffy_box,ganker5/so100_push_20250328,ganker5/so100_dataline_0328,ganker5/so100_color_0328,CrazyYhang/A1234-B-C_mvA2B,RasmusP/so100_Orange2Green,sixpigs1/so100_pick_cube_in_box,ganker5/so100_push_20250331,ganker5/so100_dataline_20250331,lirislab/put_caps_into_teabox,lirislab/close_top_drawer_teabox,lirislab/open_top_drawer_teabox,lirislab/unfold_bottom_right,lirislab/push_cup_target,lirislab/put_banana_bowl,Chojins/chess_game_001_blue_stereo,Chojins/chess_game_001_red_stereo,ganker5/so100_toy_20250402,Gano007/so100_medic,00ri/so100_battery_bin_center,paszea/so100_whale_2,lirislab/fold_bottom_right,lirislab/put_coffee_cap_teabox,therarelab/so100_pick_place_2,paszea/so100_whale_3,paszea/so100_whale_4,paszea/so100_lego,LemonadeDai/so100_coca,zijian2022/backgrounda,zijian2022/backgroundb,356c/so100_nut_sort_1,Mwuqiu/so100_0408_muti,aimihat/so100_tape,lirislab/so100_demo,356c/so100_duck_reposition_1,zijian2022/sort1,weiye11/so100_410_zwy,VoicAndrei/so100_banana_to_plate_only,sixpigs1/so100_stack_cube_error,isadev/bougies3,zijian2022/close3,bensprenger/left_arm_yellow_brick_in_box_v0,lirislab/guess_who_so100,bensprenger/left_arm_yellow_brick_in_box_with_purple_noise_v0,roboticshack/team16-can-stacking,zijian2022/insert2,roboticshack/team-7-right-arm-grasp-tape,Jiangeng/so100_413,roboticshack/team9-pick_cube_place_static_plate,AndrejOrsula/lerobot_double_ball_stacking_random,roboticshack/left-arm-grasp-lego-brick,roboticshack/team-7-left-arm-grasp-motor,roboticshack/team9-pick_chicken_place_plate,roboticshack/team13-two-balls-stacking,tkc79/so100_lego_box_1,roboticshack/team13-three-balls-stacking,pierfabre/chicken,roboticshack/team16-water-pouring,ad330/cubePlace,Jiafei1224/so100_pa222per,paszea/so100_lego_2cam,bensprenger/chess_game_001_blue_stereo,Mohamedal/put_banana,tkc79/so100_lego_box_2,samanthalhy/so100_herding_1,jlesein/TestBoulon7,pranavsaroha/so100_onelego2,pranavsaroha/so100_onelego3,pranavsaroha/so100_carrot_2,vladfatu/so100_above,koenvanwijk/orange50-1,CSCSXX/pick_place_cube_1.18,dragon-95/so100_sorting,dragon-95/so100_sorting_1,nbaron99/so100_pick_and_place4,Beegbrain/pick_place_green_block,dragon-95/so100_sorting_3,HITHY/so100_peach3,shreyasgite/so100_legocube_50,triton7777/so100_dataset_mix,NONHUMAN-RESEARCH/SOARM100_TASK_VENDA,mikechambers/block_cup_14,samsam0510/tooth_extraction_3,samsam0510/tooth_extraction_4,samsam0510/cube_reorientation_2,samsam0510/cube_reorientation_4,samsam0510/glove_reorientation_1,vladfatu/so100_office,pranavsaroha/so100_legos4,Ityl/so100_recording2,FeiYjf/new_GtoR,dragon-95/so100_sorting_2,HITHY/so100_peach4,jpata/so100_pick_place_tangerine,HITHY/so100_strawberry,shreyasgite/so100_base_env,koenvanwijk/orange50-variation-2,pranavsaroha/so100_carrot_5,pandaRQ/pick_med_1,aractingi/push_cube_offline_data,DorayakiLin/so100_pick_charger_on_tissue,zijian2022/noticehuman3,liuhuanjim013/so100_th -# REPO_ID=pierfabre/rabbit,bensprenger/right_arm_p_brick_in_box_with_y_noise_v0,pierfabre/horse,pierfabre/pig2 -# SAMPLING_WEIGHTS= - -# # # Community V3 -# # # issues, yskim2025/unitylerobot (version), cranberrysoft/so100 (don't exist),29 datasets different actions: nguyen-v/so100_rotate_red_button satvikahuja/mixer_on_off_new_1 ... -# # REPO_ID=satvikahuja/mixer_on_off_new_1,aergogo/so100_pick_place,andy309/so100_0314_fold_cloths,jchun/so100_pickplace_small_20250323_120056,astroyat/cube,Ofiroz91/so_100_cube2bowl,HappyPablo/dec3_data2,ZCM5115/so100_1210,francescocrivelli/orange_feeding,francescocrivelli/carrot_eating,0x00raghu/toffee_red,0x00raghu/toffee_red_2,0x00raghu/toffee_red_3__,0x00raghu/toffee_blue,0x00raghu/toffee_blue_2,0x00raghu/toffee_to_hand_1,0x00raghu/toffee_to_hand_2,liyitenga/so100_bi_hello,liyitenga/so100_bi_giveme5,ZCM5115/so100_2Arm3cameras_movebox,pranavsaroha/so100_carrot_1,pranavsaroha/so100_carrot_3,pranavsaroha/so100_carrot_4,maximilienroberti/so100_lego_red_box,pranavsaroha/so100_squishy,rabhishek100/so100_train_dataset,pranavsaroha/so100_squishy100,swarajgosavi/kikobot_pusht_real_v2,pandaRQ/pickmed,swarajgosavi/act_kikobot_pusht_real,pranavsaroha/so100_squishy2colors,pranavsaroha/so100_squishy2colors_1,Chojins/chess_game_001_white,jmrog/so100_sweet_pick,Chojins/chess_game_002_white,pranavsaroha/so100_squishy2colors_2_new,Chojins/chess_game_003_white,aractingi/pick_place_lego_cube,Chojins/chess_game_004_white,Chojins/chess_game_005_white,Chojins/chess_game_006_white,Chojins/chess_game_007_white,koenvanwijk/blue2,jlitch/so100multicam3,koenvanwijk/blue52,jlitch/so100multicam6,aractingi/pick_place_lego_cube_1,jlitch/so100multicam7,vladfatu/so100_ds,Chojins/chess_game_000_white,HITHY/so100-kiwi,HITHY/so100_peach1,HITHY/so100_redstrawberry,satvikahuja/orange_mixer_1,satvikahuja/mixer_on_off,satvikahuja/orange_pick_place_new1,satvikahuja/mixer_on_off_new,danmac1/real_real332,FeiYjf/Makalu_push,liyitenga/so100_pick_taffy1,chmadran/so100_dataset04,FeiYjf/Maklu_dataset,FeiYjf/new_Dataset,liyitenga/so100_pick_taffy2,satvikahuja/mixer_on_off_new_4,CSCSXX/pick_place_cube_1.17,liyitenga/so100_pick_taffy3,liyitenga/so100_pick_taffy4,yuz1wan/so100_pick_pink,yuz1wan/so100_pick_wahaha,yuz1wan/so100_pp_pink,yuz1wan/so100_pour_cup,liyitenga/so100_pick_taffy5,liyitenga/so100_pick_taffy6,yuz1wan/so100_button,yuz1wan/so100_pickplace,liyitenga/so100_pick_taffy7,FeiYjf/push_gg,FeiYjf/push_0094,swarajgosavi/act_kikobot_block_real,liyitenga/so100_pick_taffy8,phospho-ai/OrangeBrick3Cameras,vaishanthr/toy_pick_place,SeanLMH/so100_picknplace_v2,pepijn223/yellow_lego_in_box1,DimiSch/so100_50ep_2,DimiSch/so100_50ep_3,SeanLMH/so100_picknplace,nbaron99/so100_pick_and_place2,chmadran/so100_dataset08,vaishanthr/toy_pickplace_50ep,Beegbrain/pick_place_green_block_lr,Ityl/so100_recording1,vaishanthr/toy_pickplace,ad330/so100_box_pickPlace,Beegbrain/so100_put_cube_cup,aractingi/push_green_cube_hf,aractingi/push_green_cube_hf_cropped_resized,carpit680/giraffe_task,carpit680/giraffe_sock_demo_1,DimiSch/so100_terra_50_2,carpit680/giraffe_sock_demo_2,aractingi/push_cube_to_face_reward,aractingi/push_cube_to_face_reward_cropped_resized,aractingi/push_cube_reward_data,aractingi/push_cube_reward_data_cropped_resized,aractingi/push_cube_offline_data_cropped_resized,aractingi/push_cube_front_side_reward,aractingi/push_cube_front_side_reward_cropped_resized,aractingi/push_cube_front_side_reward_long,aractingi/push_cube_front_side_reward_long_cropped_resized,aractingi/push_cube_reward,aractingi/push_cube_reward_cropped_resized,aractingi/push_cube_square_reward_cropped_resized,aractingi/push_cube_square_reward_1,aractingi/push_cube_square_reward_1_cropped_resized,aractingi/push_cube_square_light_reward,aractingi/push_cube_square_light_offline_demo,aractingi/push_cube_square_light_offline_demo_cropped_resized,denghj/dataset_red_tape01,aractingi/push_cube_square_offline_demo,aractingi/push_cube_square_offline_demo_cropped_resized,Beegbrain/stack_two_cubes,FeiYjf/Test_NNNN,LegrandFrederic/Orange-brick-lower-resolution,aractingi/pick_place_lego_cube_cropped_resized,aractingi/push_cube_overfit,aractingi/push_cube_overfit_cropped_resized,HITHY/so100_peach,zaringleb/so100_cube_2,andreasBihlmaier/dual_arm_transfer_2025_02_16,zaringleb/so100_cube_4_binary,1g0rrr/reward_pickplace1,1g0rrr/reward_pickplace1_cropped_resized,FeiYjf/Hold_Pieces,FeiYjf/Grab_Pieces,hegdearyandev/so100_eraser_cup_v1,jbraumann/so100_1902,liyitenga/so100_pick_taffy10,mikechambers/block_cup_5,zaringleb/so100_cube_5_linear,yuz1wan/so100_pickplace_0223_2,yuz1wan/so100_pickplace_0223_3,samsam0510/mj_data_temp,samsam0510/tape_insert_1,samsam0510/tape_insert_2,pengjunkun/so100_push_to_hole,Deason11/Random_Kitchen,1g0rrr/reward_dataset_name2,1g0rrr/reward_dataset_name2_cropped_resized,1g0rrr/offline_dataset_name2,1g0rrr/offline_dataset_name2_cropped_resized,aractingi/push_cube_simp_cropped_resized,danielkr452/so100_work6,Loki0929/so100_100,yuz1wan/so100_fold_0227_1,yuz1wan/so100_fold_0227_2,speedyyoshi/so100_grasp_pink_block,lirislab/stack_two_red_cubes,lirislab/red_cube_into_mug,lirislab/green_lego_block_into_mug,lirislab/green_lego_block_into_mug_easy,kevin510/lerobot-cat-toy-placement,NONHUMAN-RESEARCH/SOARM100_TASK_VENDA_BOX,wangjl1512/pour_water,airthebear/so100_GL,zijian2022/noticehuman1,zijian2022/noticehuman2,kantine/so100_kapla_tower6,zijian2022/noticehuman5,zijian2022/llm40,Ashton3/lerobot-aloha,zijian2022/noticehuman50,AaronNewman/screwdriver_task_batch1,AaronNewman/screwdriver_task_batch2,AaronNewman/screwdriver_task_batch3,zijian2022/noticehuman60,zijian2022/noticehuman70,Bartm3/tape_to_bin,liuhuanjim013/so100_th_1,Pi-robot/barbecue_flip,Pi-robot/barbecue_put,wangjl1512/doll,sshh11/so100_orange_50ep_1,sshh11/so100_orange_50ep_2,DorayakiLin/so100_pick_cube_in_box,Bartm3/tape_to_bin2,luke250305/play_dice_250311.1,andy309/so100_0311_1152,sihyun77/suho_so100,sihyun77/si_so100,shreyasgite/so100_base_left,sihyun77/suho_red,liuhuanjim013/so100_block,andy309/so100_0313_no_wrist_camera,zijian2022/l9,zijian2022/n1_2,DorayakiLin/so100_stack_cube,andy309/so100_0313_no_wrist_camera_with_two_arms_cloths,joaoocruz00/so100_makeitD1,zijian2022/l10_1,zijian2022/l10_5,sihyun77/suho_red2,sihyun77/suho_angel,sihyun77/sihyun_king,acrampette/third_arm_01,Winster/so100_cube,1g0rrr/sam_openpi03,thedevansh/mar16_1336,hkphoooey/throw_stuffie,doujiangwang/task1_10epi_100000step,sihyun77/sihyun_3_17_1,acrampette/third_arm_02,imsyed00/so100_yellowbowl_pickplace_1,kumarhans/so100_tape_task,sihyun77/sihyun_main,doujiangwang/task2_10epi_100000step,kantine/industrial_robothon_buttons_expert,kantine/industrial_robothon_buttons_anomaly,kantine/industrial_robothon_hatchAndProbe_expert,kantine/industrial_robothon_hatchAndProbe_anomaly,Odog16/so100_tea_towel_folding_v1,zijian2022/so100_318,zijian2022/so100_318_1,Congying1112/so100_place_blue_bottle_with_two_cameras,Congying1112/so100_place_blue_bottle_with_two_cameras2,Congying1112/so100_place_blue_bottle_with_single_camera,pietroom/first_task_short,kantine/industrial_screws_sorting_expert,kantine/industrial_screws_sorting_anomaly,pietroom/second_task,zijian2022/c0,doujiangwang/task4_10epi_100000step,Congying1112/so100_switch_with_onhand_camera,HYAIYN/so100_get_orange_10epi,doujiangwang/task5_10epi_100000step,1g0rrr/sam_openpi_cube_low10,1g0rrr/sam_openpi_cube_top10,1g0rrr/sam_openpi_wire10,1g0rrr/sam_openpi_solder1,1g0rrr/sam_openpi_solder2,wcode/so100_put_pen_50,jchun/so100_pickplace_small_20250322_193929,bnarin/so100_tic_tac_toe_we_do_it_live,dc2ac/so100-t5,chmadran/so100_home_dataset,baladhurgesh97/so100_final_picking_3,bnarin/so100_tic_tac_toe_move_0_0,bnarin/so100_tic_tac_toe_move_1_0,bnarin/so100_tic_tac_toe_move_2_1,bnarin/so100_tic_tac_toe_move_4_0,zaringleb/so100_cube_6_2d,andlyu/so100_indoor_0,andlyu/so100_indoor_2,Winster/so100_sim,badwolf256/so100_twin_cam_duck,Congying1112/so100_simplepick_with_2_cameras_from_top,andlyu/so100_indoor_4,Zak-Y/so100_grap_dataset,kantine/domotic_pouringCoffee_expert,kantine/domotic_pouringCoffee_anomaly,lucasngoo/so100_strawberry_grape,kantine/domotic_makingCoffee_expert,kantine/domotic_makingCoffee_anomaly,ZGGZZG/so100_drop1,kantine/industrial_soldering_expert,kantine/industrial_soldering_anomaly,Yotofu/so100_sweeper_shoes,kantine/domotic_dishTidyUp_expert,kantine/domotic_dishTidyUp_anomaly,kantine/domotic_groceriesSorting_expert,kantine/domotic_groceriesSorting_anomaly,badwolf256/so100_twin_cam_duck_v2,kantine/domotic_vegetagblesAndFruitsSorting_expert,kantine/domotic_vegetagblesAndFruitsSorting_anomaly,kantine/domotic_setTheTable_expert,kantine/domotic_setTheTable_anomaly,therarelab/so100_pick_place,abhisb/so100_51_ep,andlyu/so100_indoor_val_0,allenchienxxx/so100Test,lizi178119985/so100_jia,badwolf256/so100_twin_cam_duck_v3,andrewcole712/so100_tape_bin_place,Gano007/so100_lolo,Zak-Y/so100_three_cameras_dataset,Gano007/so100_doliprane,XXRRSSRR/so100_v3_num_episodes_50,zijian2022/assemblyarm2,ganker5/so100_action_20250403,andlyu/so100_indoor_val2,Gano007/so100_gano,paszea/so100_whale_grab,paszea/so100_whale,Clementppr/lerobot_pick_and_place_dataset_world_model,andlyu/so100_indoor_10,RasmusP/so100_dataset50ep_a,RasmusP/so100_dataset50ep,Gano007/so100_second,zaringleb/so100_cude_linear_and_2d_comb,dsfsg/grasp_pens,zijian2022/digitalfix,zijian2022/digitalfix2,zijian2022/digitalfix3,T1g3rGE/so100_pickplace_small_20250407_171912,sihyun77/mond_13,abokinala/sputnik_100_11_pick_place_container,dsfsg/bring_bottle,duthvik/sputnik_100_13_pick_place_container,abokinala/sputnik_100_12_pick_place_container,Mwuqiu/so100_0408,AK51/4090_01,356c/so100_rope_reposition_1,paszea/so100_lego_mix,abokinala/sputnik_100_14_pick_place_container,abokinala/sputnik_100_23_pick_place_surface,jiajun001/eraser00_2,jlesein/TestBoulon2,duthvik/sputnik_100_31_pour_liquid,duthvik/sputnik_100_24_pick_place_surface,duthvik/sputnik_100_25_pick_place_surface,duthvik/sputnik_100_17_pick_place_container,duthvik/sputnik_100_26_pick_place_surface,VoicAndrei/so100_banana_to_plate_rebel_full,isadev/bougies1,danaaubakirova/so100_task_1,danaaubakirova/so100_task_2,danaaubakirova/so100_task_3,danaaubakirova/so100_task_4,sixpigs1/so100_pick_cube_in_box_error,sixpigs1/so100_push_cube_error,sixpigs1/so100_pull_cube_error,isadev/bougies2,therarelab/med_dis_rare_6,duthvik/sputnik_100_27_pick_place_surface,zijian2022/closer3,duthvik/sputnik_100_41_custom_tasks,duthvik/sputnik_100_42_custom_tasks,duthvik/sputnik_100_43_custom_tasks,duthvik/sputnik_100_44_custom_tasks,duthvik/sputnik_100_51_kitchen_tasks,duthvik/sputnik_100_52_kitchen_tasks,duthvik/sputnik_100_53_kitchen_tasks,duthvik/sputnik_100_45_custom_tasks,duthvik/sputnik_100_32_pour_liquid,duthvik/sputnik_100_29_pick_place_surface,duthvik/sputnik_100_18_pick_place_container,sixpigs1/so100_pull_cube_by_tool_error,sixpigs1/so100_insert_cylinder_error,abokinala/sputnik_100_54_kitchen_tasks,abokinala/sputnik_100_55_kitchen_tasks,m1b/so100_bluelego,abokinala/sputnik_100_46_custom_tasks,m1b/so100_bluelego_updt,kantine/flip_A0,kantine/flip_A1,kantine/flip_A2,kantine/flip_A3,lirislab/guess_who_no_cond,kantine/flip_A4,kantine/flip_A5,lirislab/guess_who_lighting,nguyen-v/so100_press_red_button,nguyen-v/so100_bimanual_grab_lemon_put_in_box2,pierfabre/cow,nguyen-v/press_red_button_new,nguyen-v/so100_rotate_red_button,raghav-katta-1/lerobot2,Cidoyi/so100_all_notes,roboticshack/team10-red-block,Cidoyi/so100_all_notes_1,roboticshack/team_5-QuiEstCe_everyBox,roboticshack/team11_pianobot,roboticshack/team2-guess_who_so100,roboticshack/team2-guess_who_so100_light,roboticshack/team2-guess_who_so100_edge_case,roboticshack/team2-guess_who_less_ligth,Cidoyi/so100_all_notes_3,dsfsg/grasp_pen_and_bottle,abokinala/sputnik_100_60_kitchen_tasks,abokinala/sputnik_100_58_kitchen_tasks,danaaubakirova/so100_v2_task_1,danaaubakirova/so100_v2_task_2,danaaubakirova/so100_v2_task_3,danaaubakirova/so100_v2_task_4,zijian2022/force1,zijian2022/force2,zijian2022/force3,jiajun001/eraser00_3,zijian2022/bi2,zijian2022/bi1,zijian2022/hand1,Setchii/so100_grab_ball,MossProphet/so100_square-1-2-3.2 -# # SAMPLING_WEIGHTS= -# # DATASET_NAME=so100_community_v3 - -# ########################## - -# ROBOT=so100 -# export TOKENIZERS_PARALLELISM=false -# export MUJOCO_GL=egl - - - -# SAMPLING_WEIGHTS= -# FEATURES_VERSION=2 -# NUM_IMAGE_TRANSFORMS=10 -# TRAIN_ON_ALL_FEATURES=true -# NORM_PER_ROBOT=true -# USE_IMAGENET_STATS=false - -# MAX_STATE_DIM=6 -# MAX_ACTION_DIM=6 -# MAX_NUM_IMAGES=3 -# MAX_IMAGE_DIM=256 - - -# SEED=5000 -# BATCH_SIZE=32 -# # EVAL_STEPS=1000 -# EVAL_STEPS=100 - - - - - -# SELF_ATTN_ONLY_ACTIONS=false -# EXPERT_WIDTH_MULTIPLIER=0.75 -# PAST_OBS_KEYS="image" -# N_OBS_STEPS=1 -# NUM_EXPERT_LAYERS=0 -# CHUNK_SIZE=50 -# NUM_VLM_LAYERS=16 -# PAD_LANG_TO=longest -# EVAL_CKPT=/lustre/fswork/projects/rech/dyf/ugz83ue/logs/lerobot/lerobot_so100_community_v1_v2_smolpi0_lr1e-4bs64steps200000gpus4freeze32_imgtoktrue_cross_attn_gap1_localimgfalse_statetopreftrue_explay0_vlml16_causalacttrue_sa2_smolvlm2500_chunk50_nobs1_expw0.75_feat2_lrvlm1e-4_droptrue_max_length/checkpoints/080000/pretrained_model/ -# ADD_IMAGE_TOKENS=true -# ATTN_MODE=cross_attn -# STATE_TO_PREFIX=true -# CAUSAL_ACTION_ATTENTION_MASK=true -# SELF_ATTN_EVERY_N_LAYERS=2 -# VLM_NAME=HuggingFaceTB/SmolVLM-500M-Instruct - - -# python lerobot/scripts/offline_inference.py \ -# --output_dir=$WORK/logs/lerobot/tmp \ -# --batch_size=$BATCH_SIZE \ -# --seed=$SEED \ -# --eval_steps=$EVAL_STEPS \ -# --use_amp=false \ -# --device=cuda \ -# --dataset.repo_id=$REPO_ID --dataset.local_files_only=true --dataset.sampling_weights=$SAMPLING_WEIGHTS --dataset.use_imagenet_stats=$USE_IMAGENET_STATS --policy.normalize_per_robot_type=$NORM_PER_ROBOT \ -# --dataset.image_transforms.max_num_transforms=$NUM_IMAGE_TRANSFORMS --dataset.image_transforms.enable=true --dataset.train_on_all_features=$TRAIN_ON_ALL_FEATURES \ -# --dataset.max_action_dim=$MAX_ACTION_DIM --dataset.max_state_dim=$MAX_STATE_DIM --dataset.max_num_images=$MAX_NUM_IMAGES --dataset.max_image_dim=$MAX_IMAGE_DIM --dataset.features_version=$FEATURES_VERSION \ -# --policy.type=$POLICY \ -# --policy.checkpoint_path=$EVAL_CKPT \ -# --policy.checkpoint_keys_mapping=$CKPT_KEYS_MAPPING \ -# --policy.add_image_special_tokens=$ADD_IMAGE_TOKENS \ -# --policy.attention_mode=$ATTN_MODE \ -# --policy.causal_action_attention_mask=$CAUSAL_ACTION_ATTENTION_MASK \ -# --policy.state_to_prefix=$STATE_TO_PREFIX \ -# --policy.self_attn_every_n_layers=$SELF_ATTN_EVERY_N_LAYERS \ -# --policy.vlm_model_name=$VLM_NAME \ -# --policy.pad_language_to=$PAD_LANG_TO \ -# --policy.load_vlm_weights=$LOAD_VLM_WEIGHTS \ -# --policy.num_vlm_layers=$NUM_VLM_LAYERS \ -# --policy.chunk_size=$CHUNK_SIZE \ -# --policy.n_obs_steps=$N_OBS_STEPS \ -# --policy.past_obs_keys=$PAST_OBS_KEYS \ -# --policy.num_expert_layers=$NUM_EXPERT_LAYERS \ -# --policy.expert_width_multiplier=$EXPERT_WIDTH_MULTIPLIER \ -# --policy.peft_method=$PEFT_METHOD \ -# --policy.self_attn_only_actions=$SELF_ATTN_ONLY_ACTIONS \ -# --policy.robot_type=$ROBOT - - - - -# MULTITASK_EVAL=true -# N_EPISODES=5 -# MAX_PARRALLEL_TASKS=1 -# ACTION_STEPS_LIST=(1) -# TASK_LIST=(libero_10) -# for N_ACTION_STEPS in "${ACTION_STEPS_LIST[@]}"; do -# for TASK in "${TASK_LIST[@]}"; do -# echo "$TASK Evaluating: $EVAL_CKPT | N_ACTION_STEPS=$N_ACTION_STEPS" -# python lerobot/scripts/eval.py \ -# --output_dir=$WORK/logs/lerobot/tmp \ -# --env.type=$ENV \ -# --env.task=$TASK \ -# --eval.batch_size=$N_EPISODES \ -# --eval.n_episodes=$N_EPISODES \ -# --use_amp=false \ -# --device=cuda \ -# --policy.n_action_steps=$N_ACTION_STEPS \ -# --policy.type=$POLICY \ -# --policy.checkpoint_path=$EVAL_CKPT \ -# --policy.checkpoint_keys_mapping=$CKPT_KEYS_MAPPING \ -# --env.multitask_eval=$MULTITASK_EVAL --env.max_parallel_tasks=$MAX_PARRALLEL_TASKS \ -# --policy.add_image_special_tokens=$ADD_IMAGE_TOKENS \ -# --policy.attention_mode=$ATTN_MODE \ -# --policy.causal_action_attention_mask=$CAUSAL_ACTION_ATTENTION_MASK \ -# --policy.state_to_prefix=$STATE_TO_PREFIX \ -# --policy.self_attn_every_n_layers=$SELF_ATTN_EVERY_N_LAYERS \ -# --policy.vlm_model_name=$VLM_NAME \ -# --policy.load_vlm_weights=$LOAD_VLM_WEIGHTS \ -# --policy.num_vlm_layers=$NUM_VLM_LAYERS \ -# --policy.chunk_size=$CHUNK_SIZE - -# echo "Done with: $EVAL_CKPT | Steps=$N_ACTION_STEPS" -# echo "------------------------------------------------------" -# done -# done - diff --git a/examples/checker.py b/examples/checker.py deleted file mode 100644 index 12377e9b9..000000000 --- a/examples/checker.py +++ /dev/null @@ -1,27 +0,0 @@ -from huggingface_hub import HfApi -api = HfApi() -# api.upload_large_folder( -# repo_id="HuggingFaceVLA/libero", -# repo_type="dataset", -# folder_path="/raid/jade/.cache/huggingface/lerobot/HuggingFaceVLA/libero", -# ) -api.upload_large_folder( - repo_id="HuggingFaceVLA/metaworld_mt50", - repo_type="dataset", - folder_path="/raid/jade/.cache/huggingface/lerobot/metaworld_mt50", -) -# repo_id="HuggingFaceVLA/libero" -# # Upload extra files -# api.upload_file( -# repo_id=repo_id, -# repo_type="dataset", -# path_or_fileobj="/raid/jade/libero_converted/README.md", -# path_in_repo="README.md" -# ) - -# api.upload_folder( -# repo_id=repo_id, -# repo_type="dataset", -# folder_path="/raid/jade/libero_converted/meta", -# path_in_repo="meta" -# ) diff --git a/examples/checker2.py b/examples/checker2.py deleted file mode 100644 index a5825d87f..000000000 --- a/examples/checker2.py +++ /dev/null @@ -1,35 +0,0 @@ -import pyarrow.parquet as pq - -# # First parquet (cached HF version) -meta1 = pq.read_metadata("/raid/jade/.cache/huggingface/datasets/data/chunk-000/episode_000000.parquet") -meta1 = pq.read_metadata("//raid/jade/.cache/huggingface/lerobot/HuggingFaceVLA/libero/data/chunk-000/episode_000019.parquet") -print("First parquet key_value_metadata:") -print(meta1.metadata) # low-level file metadata -# print() -print("Second") -# Second parquet (your converted version) -meta2 = pq.read_metadata("//raid/jade/.cache/huggingface/lerobot/HuggingFaceVLA/libero/data/chunk-000/episode_000019.parquet") -print("\nSecond parquet key_value_metadata:") -# print(meta2.metadata) - -# from datasets import load_dataset -# root_dir = "/raid/jade/libero_converted" - -# # Load all parquet files under the root_dir recursively -# ds = load_dataset("parquet", data_files=f"{root_dir}/**/*.parquet") - -# print(ds) # prints split info -# print(ds["train"].features) # check schema/features - -# # Peek at one row -# example = ds["train"][0] -# print(example.keys()) -# print(type(example["observation.images.image"])) -# print(type(example["observation.images.image2"])) - -import pyarrow.parquet as pq - -for ep in ["episode_000019.parquet", "episode_000021.parquet", "episode_000026.parquet"]: - path = f"/raid/jade/.cache/huggingface/lerobot/HuggingFaceVLA/libero/data/chunk-000/{ep}" - schema = pq.read_schema(path) - print(ep, schema.names) diff --git a/examples/convert_data.py b/examples/convert_data.py deleted file mode 100644 index 96ce58cb1..000000000 --- a/examples/convert_data.py +++ /dev/null @@ -1,253 +0,0 @@ -#!/usr/bin/env python3 -""" -Convert local LeRobot datasets from v2.0 to v2.1 format. -This script adapts the official converter to work with local datasets. -""" - -import sys -import argparse -import logging -from pathlib import Path - -# Add lerobot to path -sys.path.insert(0, '/home/jade_choghari/lerobot/src') - -from lerobot.datasets.lerobot_dataset import CODEBASE_VERSION, LeRobotDataset -from lerobot.datasets.utils import EPISODES_STATS_PATH, STATS_PATH, load_stats, write_info -from lerobot.datasets.v21.convert_stats import check_aggregate_stats, convert_stats - -logging.basicConfig(level=logging.INFO) -logger = logging.getLogger(__name__) - -def convert_local_dataset( - dataset_path: str, - num_workers: int = 4, - skip_if_converted: bool = True -): - """ - Convert a local dataset from v2.0 to v2.1 format. - - Args: - dataset_path: Path to the local dataset directory - num_workers: Number of workers for parallel processing - skip_if_converted: Skip if already has episodes_stats.jsonl - """ - dataset_path = Path(dataset_path) - - print(f"🔄 Converting local dataset: {dataset_path}") - - # Check if already converted - episodes_stats_path = dataset_path / "meta" / "episodes_stats.jsonl" - if episodes_stats_path.exists() and skip_if_converted: - # Check if file is empty - file_size = episodes_stats_path.stat().st_size - if file_size == 0: - print(f" ⚠️ episodes_stats.jsonl is empty, will regenerate") - else: - # Check if file has content - with open(episodes_stats_path, 'r') as f: - content = f.read().strip() - if not content: - print(f" ⚠️ episodes_stats.jsonl has no content, will regenerate") - else: - print(f" ⏭️ Already has episodes_stats.jsonl, skipping") - return True - - try: - # Check if this is a v2.0 dataset that needs conversion - episodes_stats_path = dataset_path / "meta" / "episodes_stats.jsonl" - stats_path = dataset_path / "meta" / "stats.json" - - if not episodes_stats_path.exists() and stats_path.exists(): - print(f" 🔄 Detected v2.0 dataset, creating temporary episodes_stats.jsonl...") - # Create empty episodes_stats.jsonl to allow loading - episodes_stats_path.touch() - created_temp_file = True - else: - created_temp_file = False - - # Load dataset from local path with pyav video backend - print(f" 📂 Loading dataset from local path...") - # Use a dummy repo_id since we're loading locally - dummy_repo_id = f"{dataset_path.parent.name}/{dataset_path.name}" - dataset = LeRobotDataset( - dummy_repo_id, - root=str(dataset_path), - # video_backend="pyav", - # local_files_only=True - ) - - # Remove temporary file if we created it - if created_temp_file and episodes_stats_path.exists() and episodes_stats_path.stat().st_size == 0: - episodes_stats_path.unlink() - print(f" 🗑️ Removed temporary episodes_stats.jsonl") - - # Remove existing episodes_stats if present (ensure clean conversion) - episodes_stats_path = dataset_path / "meta" / "episodes_stats.jsonl" - if episodes_stats_path.exists(): - episodes_stats_path.unlink() - print(f" 🗑️ Removed existing episodes_stats.jsonl") - - # Check if video directory exists before conversion - videos_dir = dataset_path / "videos" - if not videos_dir.exists(): - print(f" ⚠️ No videos directory found - will skip video statistics") - - # Convert stats - print(f" 📊 Computing episode statistics...") - convert_stats(dataset, num_workers=num_workers) - - # Load reference stats for validation if they exist - stats_path = dataset.root / STATS_PATH - if stats_path.exists(): - print(f" ✅ Validating against reference stats...") - try: - ref_stats = load_stats(dataset.root) - check_aggregate_stats(dataset, ref_stats) - print(f" ✅ Stats validation passed!") - except AssertionError as e: - print(f" ⚠️ Stats validation failed with minor differences: {e}") - print(f" ⚠️ This is likely due to floating-point precision, continuing anyway...") - # Check if the error is just a small numerical difference - if "Max absolute difference:" in str(e) and "Max relative difference:" in str(e): - print(f" ✅ Treating as acceptable numerical precision difference") - else: - raise e - - # Remove old stats.json file - print(f" 🗑️ Removing old stats.json") - stats_path.unlink() - else: - print(f" ⚠️ No reference stats found, skipping validation") - - # Update codebase version - dataset.meta.info["codebase_version"] = CODEBASE_VERSION - write_info(dataset.meta.info, dataset.root) - - print(f" ✅ Successfully converted to v2.1") - return True - - except Exception as e: - print(f" ❌ Failed to convert: {e}") - logger.exception("Conversion failed") - return False - -def convert_multiple_datasets( - base_dirs: list[str], - max_datasets: int = None, - num_workers: int = 4 -): - """Convert multiple datasets from base directories.""" - - datasets_to_convert = [] - - # Scan for datasets needing conversion - for base_dir in base_dirs: - base_path = Path(base_dir) - if not base_path.exists(): - print(f"⚠️ Directory not found: {base_dir}") - continue - - print(f"🔍 Scanning: {base_dir}") - - # Walk through author/dataset structure - for author_dir in sorted(base_path.iterdir()): - if not author_dir.is_dir(): - continue - - for dataset_dir in sorted(author_dir.iterdir()): - if not dataset_dir.is_dir(): - continue - - # Check if needs conversion - episodes_stats_path = dataset_dir / "meta" / "episodes_stats.jsonl" - info_path = dataset_dir / "meta" / "info.json" - - needs_conversion = False - if info_path.exists(): - if not episodes_stats_path.exists(): - needs_conversion = True - print(f" 📝 Found (missing): {author_dir.name}/{dataset_dir.name}") - else: - # Check if episodes_stats file is empty - try: - file_size = episodes_stats_path.stat().st_size - if file_size == 0: - needs_conversion = True - print(f" 📝 Found (empty): {author_dir.name}/{dataset_dir.name}") - else: - # Check if file has content - with open(episodes_stats_path, 'r') as f: - content = f.read().strip() - if not content: - needs_conversion = True - print(f" 📝 Found (no content): {author_dir.name}/{dataset_dir.name}") - except Exception as e: - # If we can't read the file, consider it needs conversion - needs_conversion = True - print(f" 📝 Found (read error): {author_dir.name}/{dataset_dir.name}") - - if needs_conversion: - datasets_to_convert.append(dataset_dir) - - if not datasets_to_convert: - print("🎉 No datasets need conversion!") - return - - if max_datasets: - datasets_to_convert = datasets_to_convert[:max_datasets] - - print(f"\n🚀 Converting {len(datasets_to_convert)} datasets...") - - successful = 0 - failed = 0 - - for i, dataset_path in enumerate(datasets_to_convert, 1): - print(f"\n[{i}/{len(datasets_to_convert)}] {dataset_path.parent.name}/{dataset_path.name}") - - success = convert_local_dataset(dataset_path, num_workers=num_workers) - if success: - successful += 1 - else: - failed += 1 - - print(f"\n📊 Conversion Summary:") - print(f" ✅ Successful: {successful}") - print(f" ❌ Failed: {failed}") - print(f" 📈 Success rate: {successful}/{len(datasets_to_convert)} ({100*successful/len(datasets_to_convert):.1f}%)") - - -def main(): - parser = argparse.ArgumentParser(description="Convert local LeRobot datasets to v2.1 format") - parser.add_argument("--dataset", type=str, help="Single dataset path to convert") - parser.add_argument("--base-dirs", nargs="+", - default=["/fsx/dana_aubakirova/vla/community_dataset_v1"], - help="Base directories to scan for datasets") - parser.add_argument("--max-datasets", type=int, help="Maximum number of datasets to convert") - parser.add_argument("--num-workers", type=int, default=4, help="Number of workers for stats computation") - parser.add_argument("--all", action="store_true", help="Convert all datasets in base directories") - - args = parser.parse_args() - - if args.dataset: - # Convert single dataset - success = convert_local_dataset(args.dataset, num_workers=args.num_workers) - if success: - print(f"\n🎉 Successfully converted: {args.dataset}") - else: - print(f"\n💥 Failed to convert: {args.dataset}") - sys.exit(1) - - elif args.all: - # Convert all datasets - convert_multiple_datasets( - args.base_dirs, - max_datasets=args.max_datasets, - num_workers=args.num_workers - ) - - else: - parser.print_help() - -if __name__ == "__main__": - main() \ No newline at end of file diff --git a/examples/convert_libero.py b/examples/convert_libero.py deleted file mode 100644 index 7bfc50eae..000000000 --- a/examples/convert_libero.py +++ /dev/null @@ -1,126 +0,0 @@ -import os -import pyarrow.parquet as pq -import tempfile -import shutil - -# Root directory of converted data -root_dir = "/raid/jade/libero_converted" - -# No renaming -rename_map = { - -} - -# Hugging Face features metadata (constant across all files) -HF_METADATA = { - b"huggingface": b'{"info": {"features": {"observation.images.image": {"_type": "Image"}, "observation.images.image2": {"_type": "Image"}, "state": {"feature": {"dtype": "float32", "_type": "Value"}, "length": 8, "_type": "Sequence"}, "actions": {"feature": {"dtype": "float32", "_type": "Value"}, "length": 7, "_type": "Sequence"}, "timestamp": {"dtype": "float32", "_type": "Value"}, "frame_index": {"dtype": "int64", "_type": "Value"}, "episode_index": {"dtype": "int64", "_type": "Value"}, "index": {"dtype": "int64", "_type": "Value"}, "task_index": {"dtype": "int64", "_type": "Value"}}}}' -} - -def patch_parquet(parquet_path, hf_metadata): - try: - table = pq.read_table(parquet_path) - - # Merge metadata - new_meta = dict(table.schema.metadata or {}) - new_meta.update(hf_metadata) - - # Apply metadata to table - table = table.replace_schema_metadata(new_meta) - - # Write safely via temp file - tmp_fd, tmp_path = tempfile.mkstemp(suffix=".parquet") - os.close(tmp_fd) - pq.write_table(table, tmp_path) - shutil.move(tmp_path, parquet_path) - - print(f"✅ Patched: {parquet_path}") - return True - except Exception as e: - print(f"❌ Failed on {parquet_path}: {e}") - return False - -# Walk through all chunk dirs and patch parquet files -for dirpath, _, filenames in os.walk(root_dir): - for fname in filenames: - if fname.endswith(".parquet"): - fpath = os.path.join(dirpath, fname) - patch_parquet(fpath, HF_METADATA)#!/usr/bin/env python3 - -#!/usr/bin/env python3 -import os -import pyarrow.parquet as pq -import tempfile -import shutil - -# Explicit list of files to patch -FILES_TO_PATCH = [ - "/raid/jade/.cache/huggingface/lerobot/HuggingFaceVLA/libero/data/chunk-000/episode_000021.parquet", - "/raid/jade/.cache/huggingface/lerobot/HuggingFaceVLA/libero/data/chunk-000/episode_000022.parquet", - "/raid/jade/.cache/huggingface/lerobot/HuggingFaceVLA/libero/data/chunk-000/episode_000023.parquet", - "/raid/jade/.cache/huggingface/lerobot/HuggingFaceVLA/libero/data/chunk-000/episode_000024.parquet", - "/raid/jade/.cache/huggingface/lerobot/HuggingFaceVLA/libero/data/chunk-000/episode_000025.parquet", -] - -# Optional renaming map (fill in as needed) -rename_map = { - # "old_column_name": "new_column_name", - "image": "observation.images.image", - "image2": "observation.images.image2", - "actions": "action", -} - -# Hugging Face features metadata (constant across all files) -HF_METADATA = { - b"huggingface": b'{"info": {"features": {' - b'"observation.images.image": {"_type": "Image"}, ' - b'"observation.images.image2": {"_type": "Image"}, ' - b'"state": {"feature": {"dtype": "float32", "_type": "Value"}, "length": 8, "_type": "Sequence"}, ' - b'"actions": {"feature": {"dtype": "float32", "_type": "Value"}, "length": 7, "_type": "Sequence"}, ' - b'"timestamp": {"dtype": "float32", "_type": "Value"}, ' - b'"frame_index": {"dtype": "int64", "_type": "Value"}, ' - b'"episode_index": {"dtype": "int64", "_type": "Value"}, ' - b'"index": {"dtype": "int64", "_type": "Value"}, ' - b'"task_index": {"dtype": "int64", "_type": "Value"}}}}' -} - -def patch_parquet(parquet_path, hf_metadata, rename_map): - try: - # Load parquet table - table = pq.read_table(parquet_path) - - # If renaming is needed - if rename_map: - schema = table.schema - new_names = [ - rename_map.get(name, name) for name in schema.names - ] - table = table.rename_columns(new_names) - - # Merge schema metadata - new_meta = dict(table.schema.metadata or {}) - new_meta.update(hf_metadata) - - # Replace metadata in table - table = table.replace_schema_metadata(new_meta) - - # Write safely via temp file - tmp_fd, tmp_path = tempfile.mkstemp(suffix=".parquet") - os.close(tmp_fd) - pq.write_table(table, tmp_path) - - # Replace original file - shutil.move(tmp_path, parquet_path) - - print(f"✅ Patched: {parquet_path}") - return True - except Exception as e: - print(f"❌ Failed on {parquet_path}: {e}") - return False - - -if __name__ == "__main__": - for fpath in FILES_TO_PATCH: - if os.path.exists(fpath): - patch_parquet(fpath, HF_METADATA, rename_map) - else: - print(f"⚠️ File not found: {fpath}") diff --git a/examples/evaluate_libero.py b/examples/evaluate_libero.py deleted file mode 100644 index bf99994b6..000000000 --- a/examples/evaluate_libero.py +++ /dev/null @@ -1,255 +0,0 @@ -""" -This script demonstrates how to evaluate a pretrained smolVLA policy on the LIBERO benchmark. -""" - -import collections -import dataclasses -import logging -import math -import pathlib - -import cv2 -import draccus -import imageio -import numpy as np -import torch -from libero.libero import benchmark, get_libero_path -from libero.libero.envs import OffScreenRenderEnv -from tqdm import tqdm - -from lerobot.policies.smolvla.modeling_smolvla import SmolVLAPolicy -from lerobot.policies.pi0.modeling_pi0 import PI0Policy - -LIBERO_DUMMY_ACTION = [0.0] * 6 + [-1.0] -LIBERO_ENV_RESOLUTION = 256 # resolution used to render training data - -@dataclasses.dataclass -class Args: - """ - Evaluation arguments for smolVLA on LIBERO. - """ - - # --- Hugging Face arguments --- - policy_path: str = "lerobot/smolvla_base" - """Path to the pretrained policy on the Hugging Face Hub or local directory.""" - - # --- LIBERO environment-specific parameters --- - task_suite_name: str = "libero_spatial" - """Task suite. Options: libero_spatial, libero_object, libero_goal, libero_10, libero_90""" - num_steps_wait: int = 10 - """Number of steps to wait for objects to stabilize in sim.""" - num_trials_per_task: int = 50 - """Number of rollouts per task.""" - - # --- Evaluation arguments --- - video_out_path: str = "data/libero/videos" - """Path to save videos.""" - device: str = "cuda" - """Device to use for evaluation.""" - - seed: int = 7 - """Random Seed (for reproducibility)""" - - -@draccus.wrap() -def eval_libero(args: Args) -> None: - # Set random seed - torch.manual_seed(args.seed) - np.random.seed(args.seed) - - # --- Load Policy --- - policy = SmolVLAPolicy.from_pretrained(args.policy_path) - policy.to(args.device) - policy.eval() - - # --- Initialize LIBERO task suite --- - benchmark_dict = benchmark.get_benchmark_dict() - try: - task_suite = benchmark_dict[args.task_suite_name]() - except KeyError: - raise ValueError( - f"Unknown task suite: {args.task_suite_name}. " - f"Available options are: {list(benchmark_dict.keys())}" - ) - num_tasks_in_suite = task_suite.n_tasks - logging.info(f"Task suite: {args.task_suite_name}") - - pathlib.Path(args.video_out_path).mkdir(parents=True, exist_ok=True) - - if args.task_suite_name == "libero_spatial": - max_steps = 220 # longest training demo has 193 steps - elif args.task_suite_name == "libero_object": - max_steps = 280 # longest training demo has 254 steps - elif args.task_suite_name == "libero_goal": - max_steps = 300 # longest training demo has 270 steps - elif args.task_suite_name == "libero_10": - max_steps = 520 # longest training demo has 505 steps - elif args.task_suite_name == "libero_90": - max_steps = 400 # longest training demo has 373 steps - else: - # Fallback for custom task suites - max_steps = 520 - - # --- Evaluation Loop --- - total_episodes, total_successes = 0, 0 - for task_id in tqdm(range(num_tasks_in_suite), desc="Tasks"): - # Get task - task = task_suite.get_task(task_id) - - # Get default LIBERO initial states - initial_states = task_suite.get_task_init_states(task_id) - - # Initialize LIBERO environment and task description - env, task_description = _get_libero_env(task, LIBERO_ENV_RESOLUTION, args.seed) - - # Start episodes - task_episodes, task_successes = 0, 0 - for episode_idx in tqdm( - range(min(args.num_trials_per_task, len(initial_states))), - desc=f"Task {task_id}: {task.language}", - leave=False, - ): - logging.info(f"\nTask: {task_description}") - - # Reset environment and policy - env.reset() - policy.reset() - - # Set initial states - obs = env.set_init_state(initial_states[episode_idx]) - - # IMPORTANT: Do nothing for the first few timesteps because the simulator drops objects - # and we need to wait for them to fall - for _ in range(args.num_steps_wait): - obs, _, _, _ = env.step(LIBERO_DUMMY_ACTION) - - # Setup - t = 0 - frames = [] - done = False - - # Add initial frame - agentview_image = np.ascontiguousarray(obs["agentview_image"][::-1, ::-1]) - # frames.append(agentview_image) - # import ipdb; ipdb.set_trace() - logging.info(f"Starting episode {task_episodes+1}...") - while t < max_steps: - try: - # Get preprocessed image - # IMPORTANT: rotate 180 degrees to match train preprocessing - wrist_img = np.ascontiguousarray(obs["robot0_eye_in_hand_image"][::-1, ::-1]) - agentview_image = np.ascontiguousarray(obs["agentview_image"][::-1, ::-1]) - frames.append(agentview_image) - - # Prepare observations dict - state = np.concatenate( - ( - obs["robot0_eef_pos"], - _quat2axisangle(obs["robot0_eef_quat"]), - obs["robot0_gripper_qpos"], - ) - ) - observation = { - "observation.images.image": torch.from_numpy(agentview_image / 255.0) - .permute(2, 0, 1) - .to(torch.float32) - .to(args.device).unsqueeze(0), - "observation.images.image2": torch.from_numpy(wrist_img / 255.0) - .permute(2, 0, 1) - .to(torch.float32) - .to(args.device).unsqueeze(0), - "observation.state": torch.from_numpy(state).to(torch.float32).to(args.device).unsqueeze(0), - "task": task_description, - } - - # Query model to get action - with torch.inference_mode(): - action_tensor = policy.select_action(observation) - action = action_tensor.cpu().numpy()[0] - action[-1] = 1 - action[-1] - - # Execute action in environment - obs, _, done, _ = env.step(action) - if done: - task_successes += 1 - total_successes += 1 - break - t += 1 - - except Exception as e: - logging.error(f"Caught exception: {e}") - break - - task_episodes += 1 - total_episodes += 1 - - # Save a replay video of the episode - suffix = "success" if done else "failure" - task_segment = task_description.replace(" ", "_").replace("/", "_") - video_path = ( - pathlib.Path(args.video_out_path) / f"rollout_task_{task_id}_episode_{episode_idx}_{task_segment}_{suffix}.mp4" - ) - fps = 30 - writer = imageio.get_writer(video_path, fps=fps) - - for image in frames: - writer.append_data(image) - writer.close() - logging.info(f"Saved video to {video_path}") - - # Log current results - logging.info(f"Success: {done}") - if total_episodes > 0: - logging.info(f"# episodes completed so far: {total_episodes}") - logging.info(f"# successes: {total_successes} ({total_successes / total_episodes * 100:.1f}%)") - - # Log final results for the task - if task_episodes > 0: - logging.info(f"Task {task_id} success rate: {float(task_successes) / float(task_episodes):.2f}") - if total_episodes > 0: - logging.info(f"Cumulative success rate: {float(total_successes) / float(total_episodes):.2f}") - - logging.info("--- Evaluation finished ---") - if total_episodes > 0: - logging.info(f"Total success rate: {float(total_successes) / float(total_episodes):.2f}") - logging.info(f"Total episodes: {total_episodes}") - logging.info(f"Total successes: {total_successes}") - cv2.destroyAllWindows() - - -def _get_libero_env(task, resolution, seed): - """Initializes and returns the LIBERO environment, along with the task description.""" - task_description = task.language - task_bddl_file = pathlib.Path(get_libero_path("bddl_files")) / task.problem_folder / task.bddl_file - env_args = { - "bddl_file_name": str(task_bddl_file), - "camera_heights": resolution, - "camera_widths": resolution, - } - env = OffScreenRenderEnv(**env_args) - env.seed(seed) # IMPORTANT: seed seems to affect object positions even when using fixed initial state - return env, task_description - - -def _quat2axisangle(quat): - """ - Copied from robosuite: - https://github.com/ARISE-Initiative/robosuite/blob/eafb81f54ffc104f905ee48a16bb15f059176ad3/robosuite/utils/transform_utils.py#L490C1-L512C55 - """ - # clip quaternion - if quat[3] > 1.0: - quat[3] = 1.0 - elif quat[3] < -1.0: - quat[3] = -1.0 - - den = np.sqrt(1.0 - quat[3] * quat[3]) - if math.isclose(den, 0.0): - # This is (close to) a zero degree rotation, immediately return - return np.zeros(3) - - return (quat[:3] * 2.0 * math.acos(quat[3])) / den - - -if __name__ == "__main__": - logging.basicConfig(level=logging.INFO) - eval_libero() \ No newline at end of file diff --git a/examples/requirements.in b/examples/requirements.in deleted file mode 100644 index 25664608a..000000000 --- a/examples/requirements.in +++ /dev/null @@ -1,8 +0,0 @@ -imageio[ffmpeg] -numpy==1.22.4 -tqdm -tyro -PyYaml -opencv-python==4.6.0.66 -robosuite==1.4.1 -matplotlib==3.5.3 \ No newline at end of file diff --git a/examples/script2.py b/examples/script2.py deleted file mode 100644 index cbd4da913..000000000 --- a/examples/script2.py +++ /dev/null @@ -1,70 +0,0 @@ -#!/usr/bin/env python3 -import os -import pyarrow.parquet as pq -import tempfile -import shutil - -FILES_TO_PATCH = [ - "/raid/jade/.cache/huggingface/lerobot/HuggingFaceVLA/libero/data/chunk-000/episode_000021.parquet", - "/raid/jade/.cache/huggingface/lerobot/HuggingFaceVLA/libero/data/chunk-000/episode_000022.parquet", - "/raid/jade/.cache/huggingface/lerobot/HuggingFaceVLA/libero/data/chunk-000/episode_000023.parquet", - "/raid/jade/.cache/huggingface/lerobot/HuggingFaceVLA/libero/data/chunk-000/episode_000024.parquet", - "/raid/jade/.cache/huggingface/lerobot/HuggingFaceVLA/libero/data/chunk-000/episode_000025.parquet", -] - -# Column renaming map -rename_map = { - "wrist_image": "observation.images.image2", - "actions": "action", -} - -# Hugging Face metadata -HF_METADATA = { - b"huggingface": b'{"info": {"features": {' - b'"observation.images.image": {"_type": "Image"}, ' - b'"observation.images.image2": {"_type": "Image"}, ' - b'"state": {"feature": {"dtype": "float32", "_type": "Value"}, "length": 8, "_type": "Sequence"}, ' - b'"action": {"feature": {"dtype": "float32", "_type": "Value"}, "length": 7, "_type": "Sequence"}, ' - b'"timestamp": {"dtype": "float32", "_type": "Value"}, ' - b'"frame_index": {"dtype": "int64", "_type": "Value"}, ' - b'"episode_index": {"dtype": "int64", "_type": "Value"}, ' - b'"index": {"dtype": "int64", "_type": "Value"}, ' - b'"task_index": {"dtype": "int64", "_type": "Value"}}}}' -} - -def patch_parquet(parquet_path, hf_metadata, rename_map): - try: - table = pq.read_table(parquet_path) - - # Apply column renames if needed - if rename_map: - schema = table.schema - new_names = [rename_map.get(name, name) for name in schema.names] - table = table.rename_columns(new_names) - - # Merge schema metadata - new_meta = dict(table.schema.metadata or {}) - new_meta.update(hf_metadata) - - # Replace metadata - table = table.replace_schema_metadata(new_meta) - - # Write via temp file - tmp_fd, tmp_path = tempfile.mkstemp(suffix=".parquet") - os.close(tmp_fd) - pq.write_table(table, tmp_path) - - shutil.move(tmp_path, parquet_path) - print(f"✅ Patched: {parquet_path}") - return True - except Exception as e: - print(f"❌ Failed on {parquet_path}: {e}") - return False - - -if __name__ == "__main__": - for fpath in FILES_TO_PATCH: - if os.path.exists(fpath): - patch_parquet(fpath, HF_METADATA, rename_map) - else: - print(f"⚠️ File not found: {fpath}") diff --git a/examples/script3.py b/examples/script3.py deleted file mode 100644 index 7b4d7957a..000000000 --- a/examples/script3.py +++ /dev/null @@ -1,64 +0,0 @@ -#!/usr/bin/env python3 -import os -import pyarrow.parquet as pq -import tempfile -import shutil - -# Root directory containing all parquet files -ROOT_DIR = "/raid/jade/.cache/huggingface/lerobot/HuggingFaceVLA/libero/data" - -# Column renaming map (normalize schema to what training expects) -rename_map = { - "state": "observation.state", -} - -# Hugging Face metadata (aligned with expected feature names) -HF_METADATA = { - b"huggingface": b'{"info": {"features": {' - b'"observation.images.image": {"_type": "Image"}, ' - b'"observation.images.image2": {"_type": "Image"}, ' - b'"observation.state": {"feature": {"dtype": "float32", "_type": "Value"}, "length": 8, "_type": "Sequence"}, ' - b'"action": {"feature": {"dtype": "float32", "_type": "Value"}, "length": 7, "_type": "Sequence"}, ' - b'"timestamp": {"dtype": "float32", "_type": "Value"}, ' - b'"frame_index": {"dtype": "int64", "_type": "Value"}, ' - b'"episode_index": {"dtype": "int64", "_type": "Value"}, ' - b'"index": {"dtype": "int64", "_type": "Value"}, ' - b'"task_index": {"dtype": "int64", "_type": "Value"}}}}' -} - -def patch_parquet(parquet_path, hf_metadata, rename_map): - try: - # Read the parquet table - table = pq.read_table(parquet_path) - - # Apply renames if necessary - if rename_map: - new_names = [rename_map.get(name, name) for name in table.schema.names] - if new_names != table.schema.names: - table = table.rename_columns(new_names) - - # Update metadata - new_meta = dict(table.schema.metadata or {}) - new_meta.update(hf_metadata) - table = table.replace_schema_metadata(new_meta) - - # Write to temp file then atomically move back - tmp_fd, tmp_path = tempfile.mkstemp(suffix=".parquet") - os.close(tmp_fd) - pq.write_table(table, tmp_path) - shutil.move(tmp_path, parquet_path) - - # Debug print - print(f"✅ Patched: {parquet_path}") - print(" Columns:", table.schema.names) - return True - except Exception as e: - print(f"❌ Failed on {parquet_path}: {e}") - return False - -if __name__ == "__main__": - for dirpath, _, filenames in os.walk(ROOT_DIR): - for fname in filenames: - if fname.endswith(".parquet"): - fpath = os.path.join(dirpath, fname) - patch_parquet(fpath, HF_METADATA, rename_map) diff --git a/examples/script4.py b/examples/script4.py deleted file mode 100644 index 2eed60a94..000000000 --- a/examples/script4.py +++ /dev/null @@ -1,3 +0,0 @@ -from huggingface_hub import HfApi -hub_api = HfApi() -hub_api.create_tag("HuggingFaceVLA/libero", tag="v2.1", repo_type="dataset") diff --git a/log_text.txt b/log_text.txt deleted file mode 100644 index 6676df0eb..000000000 --- a/log_text.txt +++ /dev/null @@ -1,1765 +0,0 @@ - self.vlm_with_expert = SmolVLMWithExpertModel( - File "/home/jade_choghari/lerobot/src/lerobot/policies/smolvla/smolvlm_with_expert.py", line 88, in __init__ - self.processor = AutoProcessor.from_pretrained(model_id) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/transformers/models/auto/processing -_auto.py", line 300, in from_pretrained - config_dict, _ = ProcessorMixin.get_processor_dict(pretrained_model_name_or_path, **kwargs) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/transformers/processing_utils.py", -line 944, in get_processor_dict - resolved_raw_chat_template_file = cached_file( - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/transformers/utils/hub.py", line 32 -1, in cached_file - file = cached_files(path_or_repo_id=path_or_repo_id, filenames=[filename], **kwargs) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/transformers/utils/hub.py", line 47 -8, in cached_files - hf_hub_download( - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/huggingface_hub/utils/_validators.p -y", line 114, in _inner_fn - return fn(*args, **kwargs) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/huggingface_hub/file_download.py", -line 1010, in hf_hub_download - return _hf_hub_download_to_cache_dir( - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/huggingface_hub/file_download.py", -line 1073, in _hf_hub_download_to_cache_dir - (url_to_download, etag, commit_hash, expected_size, xet_file_data, head_call_error) = _get_metadata_or_catch_err -or( - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/huggingface_hub/file_download.py", -line 1546, in _get_metadata_or_catch_error - metadata = get_hf_file_metadata( - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/huggingface_hub/utils/_validators.p -y", line 114, in _inner_fn - return fn(*args, **kwargs) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/huggingface_hub/file_download.py", -line 1463, in get_hf_file_metadata - r = _request_wrapper( - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/huggingface_hub/file_download.py", -line 286, in _request_wrapper - response = _request_wrapper( - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/huggingface_hub/file_download.py", -line 309, in _request_wrapper - response = http_backoff(method=method, url=url, **params, retry_on_exceptions=(), retry_on_status_codes=(429,)) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/huggingface_hub/utils/_http.py", li -ne 310, in http_backoff - response = session.request(method=method, url=url, **kwargs) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/requests/sessions.py", line 589, in - request - resp = self.send(prep, **send_kwargs) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/requests/sessions.py", line 703, in - send - r = adapter.send(request, **kwargs) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/huggingface_hub/utils/_http.py", li -ne 96, in send - return super().send(request, *args, **kwargs) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/requests/adapters.py", line 644, in - send - resp = conn.urlopen( - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/urllib3/connectionpool.py", line 78 -7, in urlopen - response = self._make_request( - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/urllib3/connectionpool.py", line 53 -4, in _make_request - response = conn.getresponse() - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/urllib3/connection.py", line 565, i -n getresponse - httplib_response = super().getresponse() - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/http/client.py", line 1375, in getresponse - response.begin() - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/http/client.py", line 318, in begin - version, status, reason = self._read_status() - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/http/client.py", line 279, in _read_status - line = str(self.fp.readline(_MAXLINE + 1), "iso-8859-1") - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/socket.py", line 717, in readinto - return self._sock.recv_into(b) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/ssl.py", line 1307, in recv_into - return self.read(nbytes, buffer) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/ssl.py", line 1163, in read - return self._sslobj.read(len, buffer) -KeyboardInterrupt -clea -(lerobot) jade_choghari@hf-dgx-01:~/lerobot$ clear -(lerobot) jade_choghari@hf-dgx-01:~/lerobot$ bash examples/8_train_smolvla_must.sh -Training dir: /raid/jade/logs/lerobot/lerobot_2_physical-intelligence_libero_smolvla_lr1e-4bs32steps100000 -/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/transformers/utils/hub.py:111: FutureWarnin -g: Using `TRANSFORMERS_CACHE` is deprecated and will be removed in v5 of Transformers. Use `HF_HOME` instead. - warnings.warn( -INFO 2025-09-09 15:50:52 ils/utils.py:48 Cuda backend detected, using cuda. -WARNING 2025-09-09 15:50:52 /policies.py:81 Device 'None' is not available. Switching to 'cuda'. -INFO 2025-09-09 15:50:52 ts/train.py:137 {'batch_size': 32, - 'dataset': {'episodes': None, - 'image_transforms': {'enable': False, - 'max_num_transforms': 3, - 'random_order': False, - 'tfs': {'brightness': {'kwargs': {'brightness': [0.8, - 1.2]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'contrast': {'kwargs': {'contrast': [0.8, - 1.2]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'hue': {'kwargs': {'hue': [-0.05, - 0.05]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'saturation': {'kwargs': {'saturation': [0.5, - 1.5]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'sharpness': {'kwargs': {'sharpness': [0.5, - 1.5]}, - 'type': 'SharpnessJitter', - 'weight': 1.0}}}, - 'repo_id': 'physical-intelligence/libero', - 'revision': None, - 'root': '/raid/jade/.cache/huggingface/datasets', - 'use_imagenet_stats': True, - 'video_backend': 'torchcodec'}, - 'env': {'camera_name': 'agentview_image,robot0_eye_in_hand_image', - 'episode_length': 520, - 'features': {'action': {'shape': [7], - 'type': }, - 'agent_pos': {'shape': [8], - 'type': }, - 'pixels/agentview_image': {'shape': [360, 360, 3], - 'type': }, - 'pixels/robot0_eye_in_hand_image': {'shape': [360, - 360, - 3], - 'type': }}, - 'features_map': {'action': 'action', - 'agent_pos': 'observation.state', - 'pixels/agentview_image': 'observation.images.image', - 'pixels/robot0_eye_in_hand_image': 'observation.images.image2'}, - 'fps': 30, - 'init_states': True, - 'max_parallel_tasks': 5, - 'multitask_eval': True, - 'obs_type': 'pixels_agent_pos', - 'render_mode': 'rgb_array', - 'task': 'libero_spatial', - 'type': 'libero'}, - 'eval': {'batch_size': 1, 'n_episodes': 1, 'use_async_envs': False}, - 'eval_freq': 0, - 'job_name': 'libero_smolvla', - 'log_freq': 200, - 'num_workers': 4, - 'optimizer': {'betas': [0.9, 0.95], - 'eps': 1e-08, - 'grad_clip_norm': 10, - 'lr': 0.0001, - 'type': 'adamw', - 'weight_decay': 1e-10}, - 'output_dir': '/raid/jade/logs/lerobot/lerobot_2_physical-intelligence_libero_smolvla_lr1e-4bs32steps100000', - 'policy': {'adapt_to_pi_aloha': False, - 'add_image_special_tokens': False, - 'attention_mode': 'cross_attn', - 'chunk_size': 50, - 'device': 'cuda', - 'empty_cameras': 0, - 'expert_width_multiplier': 0.5, - 'freeze_vision_encoder': True, - 'gradient_accumulation_steps': 1, - 'input_features': {}, - 'license': None, - 'load_vlm_weights': False, - 'max_action_dim': 32, - 'max_period': 4.0, - 'max_state_dim': 32, - 'min_period': 0.004, - 'n_action_steps': 1, - 'n_obs_steps': 1, - 'normalization_mapping': {'ACTION': , - 'STATE': , - 'VISUAL': }, - 'num_expert_layers': -1, - 'num_steps': 10, - 'num_vlm_layers': 16, - 'optimizer_betas': [0.9, 0.95], - 'optimizer_eps': 1e-08, - 'optimizer_grad_clip_norm': 10, - 'optimizer_lr': 0.0001, - 'optimizer_weight_decay': 1e-10, - 'output_features': {}, - 'pad_language_to': 'longest', - 'prefix_length': 0, - 'private': None, - 'push_to_hub': True, - 'repo_id': 'None', - 'resize_imgs_with_padding': [512, 512], - 'scheduler_decay_lr': 2.5e-06, - 'scheduler_decay_steps': 30000, - 'scheduler_warmup_steps': 1000, - 'self_attn_every_n_layers': 2, - 'tags': None, - 'tokenizer_max_length': 48, - 'train_expert_only': True, - 'train_state_proj': True, - 'type': 'smolvla', - 'use_amp': True, - 'use_cache': True, - 'use_delta_joint_actions_aloha': False, - 'vlm_model_name': 'HuggingFaceTB/SmolVLM2-500M-Instruct'}, - 'resume': False, - 'save_checkpoint': True, - 'save_freq': 20000, - 'scheduler': {'decay_lr': 2.5e-06, - 'num_decay_steps': 30000, - 'num_warmup_steps': 1000, - 'peak_lr': 0.0001, - 'type': 'cosine_decay_with_warmup'}, - 'seed': 1000, - 'steps': 100000, - 'use_policy_training_preset': True, - 'wandb': {'disable_artifact': False, - 'enable': False, - 'entity': None, - 'mode': None, - 'notes': None, - 'project': 'lerobot', - 'run_id': None}} -INFO 2025-09-09 15:50:52 ts/train.py:143 Logs will be saved locally. -INFO 2025-09-09 15:50:52 ts/train.py:153 Creating dataset -WARNING 2025-09-09 15:50:52 ts/utils.py:302 -The dataset you requested (physical-intelligence/libero) is in 2.0 format. -While current version of LeRobot is backward-compatible with it, the version of your dataset still uses global -stats instead of per-episode stats. Update your dataset stats to the new format using this command: -``` -python -m lerobot.datasets.v21.convert_dataset_v20_to_v21 --repo-id=physical-intelligence/libero -``` - -If you encounter a problem, contact LeRobot maintainers on [Discord](https://discord.com/invite/s3KuuzsPFb) -or open an [issue on GitHub](https://github.com/huggingface/lerobot/issues/new/choose). - -WARNING 2025-09-09 15:50:52 ts/utils.py:302 -The dataset you requested (physical-intelligence/libero) is in 2.0 format. -While current version of LeRobot is backward-compatible with it, the version of your dataset still uses global -stats instead of per-episode stats. Update your dataset stats to the new format using this command: -``` -python -m lerobot.datasets.v21.convert_dataset_v20_to_v21 --repo-id=physical-intelligence/libero -``` - -If you encounter a problem, contact LeRobot maintainers on [Discord](https://discord.com/invite/s3KuuzsPFb) -or open an [issue on GitHub](https://github.com/huggingface/lerobot/issues/new/choose). - -Resolving data files: 100%|████████████████████████████████████████████████████████| 1693/1693 [00:00<00:00, 67057.8 -5it/s] -Loading dataset shards: 100%|███████████████████████████████████████████████████████████| 70/70 [00:00<00:00, 5343.9 -4it/s] -INFO 2025-09-09 15:50:57 ts/train.py:163 Creating policy -Fetching 2 files: 100%|██████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 47393.2 -7it/s] -Fetching 2 files: 100%|███████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 3797.4 -7it/s] -Fetching 2 files: 100%|██████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 44384.1 -7it/s] -Fetching 2 files: 100%|███████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 6533.1 -8it/s] -Reducing the number of VLM layers to 16 ... -INFO 2025-09-09 15:51:30 ts/train.py:168 Creating optimizer and scheduler -INFO 2025-09-09 15:51:30 ts/train.py:180 Output dir: /raid/jade/logs/lerobot/lerobot_2_physical-intelligence_libero_ -smolvla_lr1e-4bs32steps100000 -INFO 2025-09-09 15:51:30 ts/train.py:182 cfg.env.task='libero_spatial' -INFO 2025-09-09 15:51:30 ts/train.py:183 cfg.steps=100000 (100K) -INFO 2025-09-09 15:51:30 ts/train.py:184 dataset.num_frames=273465 (273K) -INFO 2025-09-09 15:51:30 ts/train.py:185 dataset.num_episodes=1693 -INFO 2025-09-09 15:51:30 ts/train.py:186 num_learnable_params=49103712 (49M) -INFO 2025-09-09 15:51:30 ts/train.py:187 num_total_params=399268924 (399M) -INFO 2025-09-09 15:51:30 ts/train.py:225 Start offline training on a fixed dataset -> /home/jade_choghari/lerobot/src/lerobot/scripts/train.py(230)train() --> train_tracker.dataloading_s = time.perf_counter() - start_time -(Pdb) batch.keys() -dict_keys(['image', 'wrist_image', 'state', 'actions', 'timestamp', 'frame_index', 'episode_index', 'index', 'task_i -ndex', 'task']) -(Pdb) policy.config.input_features -{'image': PolicyFeature(type=, shape=(3, 256, 256)), 'wrist_image': PolicyFeature(type -=, shape=(3, 256, 256))} -(Pdb) quit() -Traceback (most recent call last): - File "/home/jade_choghari/lerobot/src/lerobot/scripts/train.py", line 343, in - main() - File "/home/jade_choghari/lerobot/src/lerobot/scripts/train.py", line 339, in main - train() - File "/home/jade_choghari/lerobot/src/lerobot/configs/parser.py", line 225, in wrapper_inner - response = fn(cfg, *args, **kwargs) - File "/home/jade_choghari/lerobot/src/lerobot/scripts/train.py", line 230, in train - train_tracker.dataloading_s = time.perf_counter() - start_time - File "/home/jade_choghari/lerobot/src/lerobot/scripts/train.py", line 230, in train - train_tracker.dataloading_s = time.perf_counter() - start_time - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/bdb.py", line 90, in trace_dispatch - return self.dispatch_line(frame) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/bdb.py", line 115, in dispatch_line - if self.quitting: raise BdbQuit -bdb.BdbQuit -clear -^[[A(lerobot) jade_choghari@hf-dgx-01:~/lerobot$ clear -(lerobot) jade_choghari@hf-dgx-01:~/lerobot$ bash examples/8_train_smolvla_must.sh -Training dir: /raid/jade/logs/lerobot/lerobot_2_physical-intelligence_libero_smolvla_lr1e-4bs32steps100000 -/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/transformers/utils/hub.py:111: FutureWarnin -g: Using `TRANSFORMERS_CACHE` is deprecated and will be removed in v5 of Transformers. Use `HF_HOME` instead. - warnings.warn( -INFO 2025-09-09 15:53:49 ils/utils.py:48 Cuda backend detected, using cuda. -WARNING 2025-09-09 15:53:49 /policies.py:81 Device 'None' is not available. Switching to 'cuda'. -INFO 2025-09-09 15:53:49 ts/train.py:137 {'batch_size': 32, - 'dataset': {'episodes': None, - 'image_transforms': {'enable': False, - 'max_num_transforms': 3, - 'random_order': False, - 'tfs': {'brightness': {'kwargs': {'brightness': [0.8, - 1.2]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'contrast': {'kwargs': {'contrast': [0.8, - 1.2]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'hue': {'kwargs': {'hue': [-0.05, - 0.05]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'saturation': {'kwargs': {'saturation': [0.5, - 1.5]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'sharpness': {'kwargs': {'sharpness': [0.5, - 1.5]}, - 'type': 'SharpnessJitter', - 'weight': 1.0}}}, - 'repo_id': 'physical-intelligence/libero', - 'revision': None, - 'root': '/raid/jade/.cache/huggingface/datasets', - 'use_imagenet_stats': True, - 'video_backend': 'torchcodec'}, - 'env': {'camera_name': 'agentview_image,robot0_eye_in_hand_image', - 'episode_length': 520, - 'features': {'action': {'shape': [7], - 'type': }, - 'agent_pos': {'shape': [8], - 'type': }, - 'pixels/agentview_image': {'shape': [360, 360, 3], - 'type': }, - 'pixels/robot0_eye_in_hand_image': {'shape': [360, - 360, - 3], - 'type': }}, - 'features_map': {'action': 'action', - 'agent_pos': 'observation.state', - 'pixels/agentview_image': 'observation.images.image', - 'pixels/robot0_eye_in_hand_image': 'observation.images.image2'}, - 'fps': 30, - 'init_states': True, - 'max_parallel_tasks': 5, - 'multitask_eval': True, - 'obs_type': 'pixels_agent_pos', - 'render_mode': 'rgb_array', - 'task': 'libero_spatial', - 'type': 'libero'}, - 'eval': {'batch_size': 1, 'n_episodes': 1, 'use_async_envs': False}, - 'eval_freq': 0, - 'job_name': 'libero_smolvla', - 'log_freq': 200, - 'num_workers': 4, - 'optimizer': {'betas': [0.9, 0.95], - 'eps': 1e-08, - 'grad_clip_norm': 10, - 'lr': 0.0001, - 'type': 'adamw', - 'weight_decay': 1e-10}, - 'output_dir': '/raid/jade/logs/lerobot/lerobot_2_physical-intelligence_libero_smolvla_lr1e-4bs32steps100000', - 'policy': {'adapt_to_pi_aloha': False, - 'add_image_special_tokens': False, - 'attention_mode': 'cross_attn', - 'chunk_size': 50, - 'device': 'cuda', - 'empty_cameras': 0, - 'expert_width_multiplier': 0.5, - 'freeze_vision_encoder': True, - 'gradient_accumulation_steps': 1, - 'input_features': {}, - 'license': None, - 'load_vlm_weights': False, - 'max_action_dim': 32, - 'max_period': 4.0, - 'max_state_dim': 32, - 'min_period': 0.004, - 'n_action_steps': 1, - 'n_obs_steps': 1, - 'normalization_mapping': {'ACTION': , - 'STATE': , - 'VISUAL': }, - 'num_expert_layers': -1, - 'num_steps': 10, - 'num_vlm_layers': 16, - 'optimizer_betas': [0.9, 0.95], - 'optimizer_eps': 1e-08, - 'optimizer_grad_clip_norm': 10, - 'optimizer_lr': 0.0001, - 'optimizer_weight_decay': 1e-10, - 'output_features': {}, - 'pad_language_to': 'longest', - 'prefix_length': 0, - 'private': None, - 'push_to_hub': True, - 'repo_id': 'None', - 'resize_imgs_with_padding': [512, 512], - 'scheduler_decay_lr': 2.5e-06, - 'scheduler_decay_steps': 30000, - 'scheduler_warmup_steps': 1000, - 'self_attn_every_n_layers': 2, - 'tags': None, - 'tokenizer_max_length': 48, - 'train_expert_only': True, - 'train_state_proj': True, - 'type': 'smolvla', - 'use_amp': True, - 'use_cache': True, - 'use_delta_joint_actions_aloha': False, - 'vlm_model_name': 'HuggingFaceTB/SmolVLM2-500M-Instruct'}, - 'resume': False, - 'save_checkpoint': True, - 'save_freq': 20000, - 'scheduler': {'decay_lr': 2.5e-06, - 'num_decay_steps': 30000, - 'num_warmup_steps': 1000, - 'peak_lr': 0.0001, - 'type': 'cosine_decay_with_warmup'}, - 'seed': 1000, - 'steps': 100000, - 'use_policy_training_preset': True, - 'wandb': {'disable_artifact': False, - 'enable': False, - 'entity': None, - 'mode': None, - 'notes': None, - 'project': 'lerobot', - 'run_id': None}} -INFO 2025-09-09 15:53:49 ts/train.py:143 Logs will be saved locally. -INFO 2025-09-09 15:53:49 ts/train.py:153 Creating dataset -WARNING 2025-09-09 15:53:49 ts/utils.py:302 -The dataset you requested (physical-intelligence/libero) is in 2.0 format. -While current version of LeRobot is backward-compatible with it, the version of your dataset still uses global -stats instead of per-episode stats. Update your dataset stats to the new format using this command: -``` -python -m lerobot.datasets.v21.convert_dataset_v20_to_v21 --repo-id=physical-intelligence/libero -``` - -If you encounter a problem, contact LeRobot maintainers on [Discord](https://discord.com/invite/s3KuuzsPFb) -or open an [issue on GitHub](https://github.com/huggingface/lerobot/issues/new/choose). - -WARNING 2025-09-09 15:53:49 ts/utils.py:302 -The dataset you requested (physical-intelligence/libero) is in 2.0 format. -While current version of LeRobot is backward-compatible with it, the version of your dataset still uses global -stats instead of per-episode stats. Update your dataset stats to the new format using this command: -``` -python -m lerobot.datasets.v21.convert_dataset_v20_to_v21 --repo-id=physical-intelligence/libero -``` - -If you encounter a problem, contact LeRobot maintainers on [Discord](https://discord.com/invite/s3KuuzsPFb) -or open an [issue on GitHub](https://github.com/huggingface/lerobot/issues/new/choose). - -Resolving data files: 100%|████████████████████████████████████████████████████████| 1693/1693 [00:00<00:00, 34701.4 -4it/s] -Loading dataset shards: 100%|███████████████████████████████████████████████████████████| 70/70 [00:00<00:00, 5495.3 -7it/s] -INFO 2025-09-09 15:53:55 ts/train.py:163 Creating policy -Fetching 2 files: 100%|██████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 41943.0 -4it/s] -Fetching 2 files: 100%|███████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 5500.7 -3it/s] -Fetching 2 files: 100%|███████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 2361.6 -6it/s] -Fetching 2 files: 100%|███████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 5041.2 -3it/s] -Reducing the number of VLM layers to 16 ... -> /home/jade_choghari/lerobot/src/lerobot/policies/factory.py(173)make_policy() --> assert isinstance(policy, nn.Module) -(Pdb) features -{'image': PolicyFeature(type=, shape=(3, 256, 256)), 'wrist_image': PolicyFeature(type -=, shape=(3, 256, 256)), 'actions': PolicyFeature(type=, - shape=(7,))} -(Pdb) ds_meta.features -{'image': {'dtype': 'image', 'shape': (256, 256, 3), 'names': ['height', 'width', 'channel']}, 'wrist_image': {'dtyp -e': 'image', 'shape': (256, 256, 3), 'names': ['height', 'width', 'channel']}, 'state': {'dtype': 'float32', 'shape' -: (8,), 'names': ['state']}, 'actions': {'dtype': 'float32', 'shape': (7,), 'names': ['actions']}, 'timestamp': {'dt -ype': 'float32', 'shape': (1,), 'names': None}, 'frame_index': {'dtype': 'int64', 'shape': (1,), 'names': None}, 'ep -isode_index': {'dtype': 'int64', 'shape': (1,), 'names': None}, 'index': {'dtype': 'int64', 'shape': (1,), 'names': -None}, 'task_index': {'dtype': 'int64', 'shape': (1,), 'names': None}} -(Pdb) quit() - -Traceback (most recent call last): - File "/home/jade_choghari/lerobot/src/lerobot/scripts/train.py", line 343, in - main() - File "/home/jade_choghari/lerobot/src/lerobot/scripts/train.py", line 339, in main - train() - File "/home/jade_choghari/lerobot/src/lerobot/configs/parser.py", line 225, in wrapper_inner - response = fn(cfg, *args, **kwargs) - File "/home/jade_choghari/lerobot/src/lerobot/scripts/train.py", line 164, in train - policy = make_policy( - File "/home/jade_choghari/lerobot/src/lerobot/policies/factory.py", line 173, in make_policy - assert isinstance(policy, nn.Module) - File "/home/jade_choghari/lerobot/src/lerobot/policies/factory.py", line 173, in make_policy - assert isinstance(policy, nn.Module) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/bdb.py", line 90, in trace_dispatch - return self.dispatch_line(frame) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/bdb.py", line 115, in dispatch_line - if self.quitting: raise BdbQuit -bdb.BdbQuit -clear -(lerobot) jade_choghari@hf-dgx-01:~/lerobot$ clear -(lerobot) jade_choghari@hf-dgx-01:~/lerobot$ bash examples/8_train_smolvla_must.sh -Training dir: /raid/jade/logs/lerobot/lerobot_2_physical-intelligence_libero_smolvla_lr1e-4bs32steps100000 -/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/transformers/utils/hub.py:111: FutureWarnin -g: Using `TRANSFORMERS_CACHE` is deprecated and will be removed in v5 of Transformers. Use `HF_HOME` instead. - warnings.warn( -INFO 2025-09-09 15:56:35 ils/utils.py:48 Cuda backend detected, using cuda. -WARNING 2025-09-09 15:56:35 /policies.py:81 Device 'None' is not available. Switching to 'cuda'. -INFO 2025-09-09 15:56:35 ts/train.py:137 {'batch_size': 32, - 'dataset': {'episodes': None, - 'image_transforms': {'enable': False, - 'max_num_transforms': 3, - 'random_order': False, - 'tfs': {'brightness': {'kwargs': {'brightness': [0.8, - 1.2]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'contrast': {'kwargs': {'contrast': [0.8, - 1.2]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'hue': {'kwargs': {'hue': [-0.05, - 0.05]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'saturation': {'kwargs': {'saturation': [0.5, - 1.5]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'sharpness': {'kwargs': {'sharpness': [0.5, - 1.5]}, - 'type': 'SharpnessJitter', - 'weight': 1.0}}}, - 'repo_id': 'physical-intelligence/libero', - 'revision': None, - 'root': '/raid/jade/.cache/huggingface/datasets', - 'use_imagenet_stats': True, - 'video_backend': 'torchcodec'}, - 'env': {'camera_name': 'agentview_image,robot0_eye_in_hand_image', - 'episode_length': 520, - 'features': {'action': {'shape': [7], - 'type': }, - 'agent_pos': {'shape': [8], - 'type': }, - 'pixels/agentview_image': {'shape': [360, 360, 3], - 'type': }, - 'pixels/robot0_eye_in_hand_image': {'shape': [360, - 360, - 3], - 'type': }}, - 'features_map': {'action': 'action', - 'agent_pos': 'observation.state', - 'pixels/agentview_image': 'observation.images.image', - 'pixels/robot0_eye_in_hand_image': 'observation.images.image2'}, - 'fps': 30, - 'init_states': True, - 'max_parallel_tasks': 5, - 'multitask_eval': True, - 'obs_type': 'pixels_agent_pos', - 'render_mode': 'rgb_array', - 'task': 'libero_spatial', - 'type': 'libero'}, - 'eval': {'batch_size': 1, 'n_episodes': 1, 'use_async_envs': False}, - 'eval_freq': 0, - 'job_name': 'libero_smolvla', - 'log_freq': 200, - 'num_workers': 4, - 'optimizer': {'betas': [0.9, 0.95], - 'eps': 1e-08, - 'grad_clip_norm': 10, - 'lr': 0.0001, - 'type': 'adamw', - 'weight_decay': 1e-10}, - 'output_dir': '/raid/jade/logs/lerobot/lerobot_2_physical-intelligence_libero_smolvla_lr1e-4bs32steps100000', - 'policy': {'adapt_to_pi_aloha': False, - 'add_image_special_tokens': False, - 'attention_mode': 'cross_attn', - 'chunk_size': 50, - 'device': 'cuda', - 'empty_cameras': 0, - 'expert_width_multiplier': 0.5, - 'freeze_vision_encoder': True, - 'gradient_accumulation_steps': 1, - 'input_features': {}, - 'license': None, - 'load_vlm_weights': False, - 'max_action_dim': 32, - 'max_period': 4.0, - 'max_state_dim': 32, - 'min_period': 0.004, - 'n_action_steps': 1, - 'n_obs_steps': 1, - 'normalization_mapping': {'ACTION': , - 'STATE': , - 'VISUAL': }, - 'num_expert_layers': -1, - 'num_steps': 10, - 'num_vlm_layers': 16, - 'optimizer_betas': [0.9, 0.95], - 'optimizer_eps': 1e-08, - 'optimizer_grad_clip_norm': 10, - 'optimizer_lr': 0.0001, - 'optimizer_weight_decay': 1e-10, - 'output_features': {}, - 'pad_language_to': 'longest', - 'prefix_length': 0, - 'private': None, - 'push_to_hub': True, - 'repo_id': 'None', - 'resize_imgs_with_padding': [512, 512], - 'scheduler_decay_lr': 2.5e-06, - 'scheduler_decay_steps': 30000, - 'scheduler_warmup_steps': 1000, - 'self_attn_every_n_layers': 2, - 'tags': None, - 'tokenizer_max_length': 48, - 'train_expert_only': True, - 'train_state_proj': True, - 'type': 'smolvla', - 'use_amp': True, - 'use_cache': True, - 'use_delta_joint_actions_aloha': False, - 'vlm_model_name': 'HuggingFaceTB/SmolVLM2-500M-Instruct'}, - 'resume': False, - 'save_checkpoint': True, - 'save_freq': 20000, - 'scheduler': {'decay_lr': 2.5e-06, - 'num_decay_steps': 30000, - 'num_warmup_steps': 1000, - 'peak_lr': 0.0001, - 'type': 'cosine_decay_with_warmup'}, - 'seed': 1000, - 'steps': 100000, - 'use_policy_training_preset': True, - 'wandb': {'disable_artifact': False, - 'enable': False, - 'entity': None, - 'mode': None, - 'notes': None, - 'project': 'lerobot', - 'run_id': None}} -INFO 2025-09-09 15:56:35 ts/train.py:143 Logs will be saved locally. -INFO 2025-09-09 15:56:35 ts/train.py:153 Creating dataset -WARNING 2025-09-09 15:56:35 ts/utils.py:302 -The dataset you requested (physical-intelligence/libero) is in 2.0 format. -While current version of LeRobot is backward-compatible with it, the version of your dataset still uses global -stats instead of per-episode stats. Update your dataset stats to the new format using this command: -``` -python -m lerobot.datasets.v21.convert_dataset_v20_to_v21 --repo-id=physical-intelligence/libero -``` - -If you encounter a problem, contact LeRobot maintainers on [Discord](https://discord.com/invite/s3KuuzsPFb) -or open an [issue on GitHub](https://github.com/huggingface/lerobot/issues/new/choose). - -WARNING 2025-09-09 15:56:35 ts/utils.py:302 -The dataset you requested (physical-intelligence/libero) is in 2.0 format. -While current version of LeRobot is backward-compatible with it, the version of your dataset still uses global -stats instead of per-episode stats. Update your dataset stats to the new format using this command: -``` -python -m lerobot.datasets.v21.convert_dataset_v20_to_v21 --repo-id=physical-intelligence/libero -``` - -If you encounter a problem, contact LeRobot maintainers on [Discord](https://discord.com/invite/s3KuuzsPFb) -or open an [issue on GitHub](https://github.com/huggingface/lerobot/issues/new/choose). - -Resolving data files: 100%|████████████████████████████████████████████████████████| 1693/1693 [00:00<00:00, 78132.9 -5it/s] -Loading dataset shards: 100%|███████████████████████████████████████████████████████████| 70/70 [00:00<00:00, 4716.0 -3it/s] -INFO 2025-09-09 15:56:40 ts/train.py:163 Creating policy -Fetching 2 files: 100%|███████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 5259.3 -2it/s] -Fetching 2 files: 100%|███████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 3477.8 -6it/s] -Fetching 2 files: 100%|██████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 45343.8 -3it/s] -Fetching 2 files: 100%|███████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 5551.6 -9it/s] -Reducing the number of VLM layers to 16 ... -> /home/jade_choghari/lerobot/src/lerobot/policies/factory.py(173)make_policy() --> assert isinstance(policy, nn.Module) -(Pdb) features -{'image': PolicyFeature(type=, shape=(3, 256, 256)), 'wrist_image': PolicyFeature(type -=, shape=(3, 256, 256)), 'state': PolicyFeature(type=, sha -pe=(8,)), 'actions': PolicyFeature(type=, shape=(7,))} -(Pdb) quit() -Traceback (most recent call last): - File "/home/jade_choghari/lerobot/src/lerobot/scripts/train.py", line 343, in - File "/home/jade_choghari/lerobot/src/lerobot/scripts/train.py", line 339, in main - - File "/home/jade_choghari/lerobot/src/lerobot/configs/parser.py", line 225, in wrapper_inner - response = fn(cfg, *args, **kwargs) - File "/home/jade_choghari/lerobot/src/lerobot/scripts/train.py", line 164, in train - policy = make_policy( - File "/home/jade_choghari/lerobot/src/lerobot/policies/factory.py", line 173, in make_policy - # policy = torch.compile(policy, mode="reduce-overhead") - File "/home/jade_choghari/lerobot/src/lerobot/policies/factory.py", line 173, in make_policy - # policy = torch.compile(policy, mode="reduce-overhead") - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/bdb.py", line 90, in trace_dispatch - return self.dispatch_line(frame) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/bdb.py", line 115, in dispatch_line - if self.quitting: raise BdbQuit -bdb.BdbQuit -clear -(lerobot) jade_choghari@hf-dgx-01:~/lerobot$ clear -(lerobot) jade_choghari@hf-dgx-01:~/lerobot$ bash examples/8_train_smolvla_must.sh -Training dir: /raid/jade/logs/lerobot/lerobot_2_physical-intelligence_libero_smolvla_lr1e-4bs32steps100000 -/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/transformers/utils/hub.py:111: FutureWarnin -g: Using `TRANSFORMERS_CACHE` is deprecated and will be removed in v5 of Transformers. Use `HF_HOME` instead. - warnings.warn( -INFO 2025-09-09 15:58:35 ils/utils.py:48 Cuda backend detected, using cuda. -WARNING 2025-09-09 15:58:35 /policies.py:81 Device 'None' is not available. Switching to 'cuda'. -INFO 2025-09-09 15:58:35 ts/train.py:137 {'batch_size': 32, - 'dataset': {'episodes': None, - 'image_transforms': {'enable': False, - 'max_num_transforms': 3, - 'random_order': False, - 'tfs': {'brightness': {'kwargs': {'brightness': [0.8, - 1.2]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'contrast': {'kwargs': {'contrast': [0.8, - 1.2]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'hue': {'kwargs': {'hue': [-0.05, - 0.05]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'saturation': {'kwargs': {'saturation': [0.5, - 1.5]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'sharpness': {'kwargs': {'sharpness': [0.5, - 1.5]}, - 'type': 'SharpnessJitter', - 'weight': 1.0}}}, - 'repo_id': 'physical-intelligence/libero', - 'revision': None, - 'root': '/raid/jade/.cache/huggingface/datasets', - 'use_imagenet_stats': True, - 'video_backend': 'torchcodec'}, - 'env': {'camera_name': 'agentview_image,robot0_eye_in_hand_image', - 'episode_length': 520, - 'features': {'action': {'shape': [7], - 'type': }, - 'agent_pos': {'shape': [8], - 'type': }, - 'pixels/agentview_image': {'shape': [360, 360, 3], - 'type': }, - 'pixels/robot0_eye_in_hand_image': {'shape': [360, - 360, - 3], - 'type': }}, - 'features_map': {'action': 'action', - 'agent_pos': 'observation.state', - 'pixels/agentview_image': 'observation.images.image', - 'pixels/robot0_eye_in_hand_image': 'observation.images.image2'}, - 'fps': 30, - 'init_states': True, - 'max_parallel_tasks': 5, - 'multitask_eval': True, - 'obs_type': 'pixels_agent_pos', - 'render_mode': 'rgb_array', - 'task': 'libero_spatial', - 'type': 'libero'}, - 'eval': {'batch_size': 1, 'n_episodes': 1, 'use_async_envs': False}, - 'eval_freq': 0, - 'job_name': 'libero_smolvla', - 'log_freq': 200, - 'num_workers': 4, - 'optimizer': {'betas': [0.9, 0.95], - 'eps': 1e-08, - 'grad_clip_norm': 10, - 'lr': 0.0001, - 'type': 'adamw', - 'weight_decay': 1e-10}, - 'output_dir': '/raid/jade/logs/lerobot/lerobot_2_physical-intelligence_libero_smolvla_lr1e-4bs32steps100000', - 'policy': {'adapt_to_pi_aloha': False, - 'add_image_special_tokens': False, - 'attention_mode': 'cross_attn', - 'chunk_size': 50, - 'device': 'cuda', - 'empty_cameras': 0, - 'expert_width_multiplier': 0.5, - 'freeze_vision_encoder': True, - 'gradient_accumulation_steps': 1, - 'input_features': {}, - 'license': None, - 'load_vlm_weights': False, - 'max_action_dim': 32, - 'max_period': 4.0, - 'max_state_dim': 32, - 'min_period': 0.004, - 'n_action_steps': 1, - 'n_obs_steps': 1, - 'normalization_mapping': {'ACTION': , - 'STATE': , - 'VISUAL': }, - 'num_expert_layers': -1, - 'num_steps': 10, - 'num_vlm_layers': 16, - 'optimizer_betas': [0.9, 0.95], - 'optimizer_eps': 1e-08, - 'optimizer_grad_clip_norm': 10, - 'optimizer_lr': 0.0001, - 'optimizer_weight_decay': 1e-10, - 'output_features': {}, - 'pad_language_to': 'longest', - 'prefix_length': 0, - 'private': None, - 'push_to_hub': True, - 'repo_id': 'None', - 'resize_imgs_with_padding': [512, 512], - 'scheduler_decay_lr': 2.5e-06, - 'scheduler_decay_steps': 30000, - 'scheduler_warmup_steps': 1000, - 'self_attn_every_n_layers': 2, - 'tags': None, - 'tokenizer_max_length': 48, - 'train_expert_only': True, - 'train_state_proj': True, - 'type': 'smolvla', - 'use_amp': True, - 'use_cache': True, - 'use_delta_joint_actions_aloha': False, - 'vlm_model_name': 'HuggingFaceTB/SmolVLM2-500M-Instruct'}, - 'resume': False, - 'save_checkpoint': True, - 'save_freq': 20000, - 'scheduler': {'decay_lr': 2.5e-06, - 'num_decay_steps': 30000, - 'num_warmup_steps': 1000, - 'peak_lr': 0.0001, - 'type': 'cosine_decay_with_warmup'}, - 'seed': 1000, - 'steps': 100000, - 'use_policy_training_preset': True, - 'wandb': {'disable_artifact': False, - 'enable': False, - 'entity': None, - 'mode': None, - 'notes': None, - 'project': 'lerobot', - 'run_id': None}} -INFO 2025-09-09 15:58:35 ts/train.py:143 Logs will be saved locally. -INFO 2025-09-09 15:58:35 ts/train.py:153 Creating dataset -WARNING 2025-09-09 15:58:35 ts/utils.py:302 -The dataset you requested (physical-intelligence/libero) is in 2.0 format. -While current version of LeRobot is backward-compatible with it, the version of your dataset still uses global -stats instead of per-episode stats. Update your dataset stats to the new format using this command: -``` -python -m lerobot.datasets.v21.convert_dataset_v20_to_v21 --repo-id=physical-intelligence/libero -``` - -If you encounter a problem, contact LeRobot maintainers on [Discord](https://discord.com/invite/s3KuuzsPFb) -or open an [issue on GitHub](https://github.com/huggingface/lerobot/issues/new/choose). - -WARNING 2025-09-09 15:58:35 ts/utils.py:302 -The dataset you requested (physical-intelligence/libero) is in 2.0 format. -While current version of LeRobot is backward-compatible with it, the version of your dataset still uses global -stats instead of per-episode stats. Update your dataset stats to the new format using this command: -``` -python -m lerobot.datasets.v21.convert_dataset_v20_to_v21 --repo-id=physical-intelligence/libero -``` - -If you encounter a problem, contact LeRobot maintainers on [Discord](https://discord.com/invite/s3KuuzsPFb) -or open an [issue on GitHub](https://github.com/huggingface/lerobot/issues/new/choose). - -Resolving data files: 100%|████████████████████████████████████████████████████████| 1693/1693 [00:00<00:00, 27666.4 -6it/s] -Loading dataset shards: 100%|███████████████████████████████████████████████████████████| 70/70 [00:00<00:00, 5305.7 -0it/s] -INFO 2025-09-09 15:58:41 ts/train.py:163 Creating policy -Fetching 2 files: 100%|██████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 44384.1 -7it/s] -Fetching 2 files: 100%|███████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 3192.0 -1it/s] -Fetching 2 files: 100%|██████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 44620.2 -6it/s] -Fetching 2 files: 100%|██████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 42799.0 -2it/s] -Reducing the number of VLM layers to 16 ... -INFO 2025-09-09 15:59:13 ts/train.py:168 Creating optimizer and scheduler -INFO 2025-09-09 15:59:13 ts/train.py:180 Output dir: /raid/jade/logs/lerobot/lerobot_2_physical-intelligence_libero_ -smolvla_lr1e-4bs32steps100000 -INFO 2025-09-09 15:59:13 ts/train.py:182 cfg.env.task='libero_spatial' -INFO 2025-09-09 15:59:13 ts/train.py:183 cfg.steps=100000 (100K) -INFO 2025-09-09 15:59:13 ts/train.py:184 dataset.num_frames=273465 (273K) -INFO 2025-09-09 15:59:13 ts/train.py:185 dataset.num_episodes=1693 -INFO 2025-09-09 15:59:13 ts/train.py:186 num_learnable_params=49103712 (49M) -INFO 2025-09-09 15:59:13 ts/train.py:187 num_total_params=399268940 (399M) -INFO 2025-09-09 15:59:13 ts/train.py:225 Start offline training on a fixed dataset -Traceback (most recent call last): - File "/home/jade_choghari/lerobot/src/lerobot/scripts/train.py", line 342, in - main() - File "/home/jade_choghari/lerobot/src/lerobot/scripts/train.py", line 338, in main - train() - File "/home/jade_choghari/lerobot/src/lerobot/configs/parser.py", line 225, in wrapper_inner - response = fn(cfg, *args, **kwargs) - File "/home/jade_choghari/lerobot/src/lerobot/scripts/train.py", line 235, in train - train_tracker, output_dict = update_policy( - File "/home/jade_choghari/lerobot/src/lerobot/scripts/train.py", line 71, in update_policy - loss, output_dict = policy.forward(batch) - File "/home/jade_choghari/lerobot/src/lerobot/policies/smolvla/modeling_smolvla.py", line 458, in forward - actions = self.prepare_action(batch) - File "/home/jade_choghari/lerobot/src/lerobot/policies/smolvla/modeling_smolvla.py", line 580, in prepare_action - actions = pad_vector(batch[ACTION], self.config.max_action_dim) -KeyError: 'action' -Exception in thread Thread-3 (_pin_memory_loop): -Traceback (most recent call last): - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/threading.py", line 1016, in _bootstrap_inner - self.run() - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/threading.py", line 953, in run - self._target(*self._args, **self._kwargs) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/torch/utils/data/_utils/pin_memory. -py", line 61, in _pin_memory_loop - do_one_step() - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/torch/utils/data/_utils/pin_memory. -py", line 37, in do_one_step - r = in_queue.get(timeout=MP_STATUS_CHECK_INTERVAL) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/multiprocessing/queues.py", line 122, in get - return _ForkingPickler.loads(res) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/torch/multiprocessing/reductions.py -", line 541, in rebuild_storage_fd - fd = df.detach() - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/multiprocessing/resource_sharer.py", line 57, in -detach - with _resource_sharer.get_connection(self._id) as conn: - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/multiprocessing/resource_sharer.py", line 86, in -get_connection - c = Client(address, authkey=process.current_process().authkey) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/multiprocessing/connection.py", line 508, in Clie -nt - answer_challenge(c, authkey) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/multiprocessing/connection.py", line 752, in answ -er_challenge - message = connection.recv_bytes(256) # reject large message - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/multiprocessing/connection.py", line 216, in recv -_bytes - buf = self._recv_bytes(maxlength) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/multiprocessing/connection.py", line 414, in _rec -v_bytes - buf = self._recv(4) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/multiprocessing/connection.py", line 379, in _rec -v - chunk = read(handle, remaining) -ConnectionResetError: [Errno 104] Connection reset by peer -(lerobot) jade_choghari@hf-dgx-01:~/lerobot$ clear -(lerobot) jade_choghari@hf-dgx-01:~/lerobot$ bash examples/8_train_smolvla_must.sh -Training dir: /raid/jade/logs/lerobot/lerobot_2_physical-intelligence_libero_smolvla_lr1e-4bs32steps100000 -/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/transformers/utils/hub.py:111: FutureWarnin -g: Using `TRANSFORMERS_CACHE` is deprecated and will be removed in v5 of Transformers. Use `HF_HOME` instead. - warnings.warn( -INFO 2025-09-09 15:59:53 ils/utils.py:48 Cuda backend detected, using cuda. -WARNING 2025-09-09 15:59:53 /policies.py:81 Device 'None' is not available. Switching to 'cuda'. -INFO 2025-09-09 15:59:53 ts/train.py:137 {'batch_size': 32, - 'dataset': {'episodes': None, - 'image_transforms': {'enable': False, - 'max_num_transforms': 3, - 'random_order': False, - 'tfs': {'brightness': {'kwargs': {'brightness': [0.8, - 1.2]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'contrast': {'kwargs': {'contrast': [0.8, - 1.2]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'hue': {'kwargs': {'hue': [-0.05, - 0.05]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'saturation': {'kwargs': {'saturation': [0.5, - 1.5]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'sharpness': {'kwargs': {'sharpness': [0.5, - 1.5]}, - 'type': 'SharpnessJitter', - 'weight': 1.0}}}, - 'repo_id': 'physical-intelligence/libero', - 'revision': None, - 'root': '/raid/jade/.cache/huggingface/datasets', - 'use_imagenet_stats': True, - 'video_backend': 'torchcodec'}, - 'env': {'camera_name': 'agentview_image,robot0_eye_in_hand_image', - 'episode_length': 520, - 'features': {'action': {'shape': [7], - 'type': }, - 'agent_pos': {'shape': [8], - 'type': }, - 'pixels/agentview_image': {'shape': [360, 360, 3], - 'type': }, - 'pixels/robot0_eye_in_hand_image': {'shape': [360, - 360, - 3], - 'type': }}, - 'features_map': {'action': 'action', - 'agent_pos': 'observation.state', - 'pixels/agentview_image': 'observation.images.image', - 'pixels/robot0_eye_in_hand_image': 'observation.images.image2'}, - 'fps': 30, - 'init_states': True, - 'max_parallel_tasks': 5, - 'multitask_eval': True, - 'obs_type': 'pixels_agent_pos', - 'render_mode': 'rgb_array', - 'task': 'libero_spatial', - 'type': 'libero'}, - 'eval': {'batch_size': 1, 'n_episodes': 1, 'use_async_envs': False}, - 'eval_freq': 0, - 'job_name': 'libero_smolvla', - 'log_freq': 200, - 'num_workers': 4, - 'optimizer': {'betas': [0.9, 0.95], - 'eps': 1e-08, - 'grad_clip_norm': 10, - 'lr': 0.0001, - 'type': 'adamw', - 'weight_decay': 1e-10}, - 'output_dir': '/raid/jade/logs/lerobot/lerobot_2_physical-intelligence_libero_smolvla_lr1e-4bs32steps100000', - 'policy': {'adapt_to_pi_aloha': False, - 'add_image_special_tokens': False, - 'attention_mode': 'cross_attn', - 'chunk_size': 50, - 'device': 'cuda', - 'empty_cameras': 0, - 'expert_width_multiplier': 0.5, - 'freeze_vision_encoder': True, - 'gradient_accumulation_steps': 1, - 'input_features': {}, - 'license': None, - 'load_vlm_weights': False, - 'max_action_dim': 32, - 'max_period': 4.0, - 'max_state_dim': 32, - 'min_period': 0.004, - 'n_action_steps': 1, - 'n_obs_steps': 1, - 'normalization_mapping': {'ACTION': , - 'STATE': , - 'VISUAL': }, - 'num_expert_layers': -1, - 'num_steps': 10, - 'num_vlm_layers': 16, - 'optimizer_betas': [0.9, 0.95], - 'optimizer_eps': 1e-08, - 'optimizer_grad_clip_norm': 10, - 'optimizer_lr': 0.0001, - 'optimizer_weight_decay': 1e-10, - 'output_features': {}, - 'pad_language_to': 'longest', - 'prefix_length': 0, - 'private': None, - 'push_to_hub': True, - 'repo_id': 'None', - 'resize_imgs_with_padding': [512, 512], - 'scheduler_decay_lr': 2.5e-06, - 'scheduler_decay_steps': 30000, - 'scheduler_warmup_steps': 1000, - 'self_attn_every_n_layers': 2, - 'tags': None, - 'tokenizer_max_length': 48, - 'train_expert_only': True, - 'train_state_proj': True, - 'type': 'smolvla', - 'use_amp': True, - 'use_cache': True, - 'use_delta_joint_actions_aloha': False, - 'vlm_model_name': 'HuggingFaceTB/SmolVLM2-500M-Instruct'}, - 'resume': False, - 'save_checkpoint': True, - 'save_freq': 20000, - 'scheduler': {'decay_lr': 2.5e-06, - 'num_decay_steps': 30000, - 'num_warmup_steps': 1000, - 'peak_lr': 0.0001, - 'type': 'cosine_decay_with_warmup'}, - 'seed': 1000, - 'steps': 100000, - 'use_policy_training_preset': True, - 'wandb': {'disable_artifact': False, - 'enable': False, - 'entity': None, - 'mode': None, - 'notes': None, - 'project': 'lerobot', - 'run_id': None}} -INFO 2025-09-09 15:59:53 ts/train.py:143 Logs will be saved locally. -INFO 2025-09-09 15:59:53 ts/train.py:153 Creating dataset -WARNING 2025-09-09 15:59:53 ts/utils.py:302 -The dataset you requested (physical-intelligence/libero) is in 2.0 format. -While current version of LeRobot is backward-compatible with it, the version of your dataset still uses global -stats instead of per-episode stats. Update your dataset stats to the new format using this command: -``` -python -m lerobot.datasets.v21.convert_dataset_v20_to_v21 --repo-id=physical-intelligence/libero -``` - -If you encounter a problem, contact LeRobot maintainers on [Discord](https://discord.com/invite/s3KuuzsPFb) -or open an [issue on GitHub](https://github.com/huggingface/lerobot/issues/new/choose). - -WARNING 2025-09-09 15:59:53 ts/utils.py:302 -The dataset you requested (physical-intelligence/libero) is in 2.0 format. -While current version of LeRobot is backward-compatible with it, the version of your dataset still uses global -stats instead of per-episode stats. Update your dataset stats to the new format using this command: -``` -python -m lerobot.datasets.v21.convert_dataset_v20_to_v21 --repo-id=physical-intelligence/libero -``` - -If you encounter a problem, contact LeRobot maintainers on [Discord](https://discord.com/invite/s3KuuzsPFb) -or open an [issue on GitHub](https://github.com/huggingface/lerobot/issues/new/choose). - -Resolving data files: 100%|████████████████████████████████████████████████████████| 1693/1693 [00:00<00:00, 72147.3 -3it/s] -Loading dataset shards: 100%|███████████████████████████████████████████████████████████| 70/70 [00:00<00:00, 5076.7 -1it/s] -INFO 2025-09-09 15:59:58 ts/train.py:163 Creating policy -Fetching 2 files: 100%|███████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 6096.3 -7it/s] -Fetching 2 files: 100%|███████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 4348.6 -8it/s] -Fetching 2 files: 100%|██████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 46091.2 -5it/s] -Fetching 2 files: 100%|███████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 3225.1 -5it/s] -Reducing the number of VLM layers to 16 ... -INFO 2025-09-09 16:00:31 ts/train.py:168 Creating optimizer and scheduler -INFO 2025-09-09 16:00:31 ts/train.py:180 Output dir: /raid/jade/logs/lerobot/lerobot_2_physical-intelligence_libero_ -smolvla_lr1e-4bs32steps100000 -INFO 2025-09-09 16:00:31 ts/train.py:182 cfg.env.task='libero_spatial' -INFO 2025-09-09 16:00:31 ts/train.py:183 cfg.steps=100000 (100K) -INFO 2025-09-09 16:00:31 ts/train.py:184 dataset.num_frames=273465 (273K) -INFO 2025-09-09 16:00:31 ts/train.py:185 dataset.num_episodes=1693 -INFO 2025-09-09 16:00:31 ts/train.py:186 num_learnable_params=49103712 (49M) -INFO 2025-09-09 16:00:31 ts/train.py:187 num_total_params=399268940 (399M) -INFO 2025-09-09 16:00:31 ts/train.py:225 Start offline training on a fixed dataset -Traceback (most recent call last): - File "/home/jade_choghari/lerobot/src/lerobot/scripts/train.py", line 342, in - main() - File "/home/jade_choghari/lerobot/src/lerobot/scripts/train.py", line 338, in main - train() - File "/home/jade_choghari/lerobot/src/lerobot/configs/parser.py", line 225, in wrapper_inner - response = fn(cfg, *args, **kwargs) - File "/home/jade_choghari/lerobot/src/lerobot/scripts/train.py", line 235, in train - train_tracker, output_dict = update_policy( - File "/home/jade_choghari/lerobot/src/lerobot/scripts/train.py", line 71, in update_policy - loss, output_dict = policy.forward(batch) - File "/home/jade_choghari/lerobot/src/lerobot/policies/smolvla/modeling_smolvla.py", line 461, in forward - losses = self.model.forward(images, img_masks, lang_tokens, lang_masks, state, actions, noise, time) - File "/home/jade_choghari/lerobot/src/lerobot/policies/smolvla/modeling_smolvla.py", line 850, in forward - att_2d_masks = make_att_2d_masks(pad_masks, att_masks) - File "/home/jade_choghari/lerobot/src/lerobot/policies/smolvla/modeling_smolvla.py", line 226, in make_att_2d_mask -s - att_2d_masks = att_2d_masks & pad_2d_masks -RuntimeError: The size of tensor a (199) must match the size of tensor b (181) at non-singleton dimension 2 -(lerobot) jade_choghari@hf-dgx-01:~/lerobot$ clear -(lerobot) jade_choghari@hf-dgx-01:~/lerobot$ bash examples/8_train_smolvla_must.sh -Training dir: /raid/jade/logs/lerobot/lerobot_2_physical-intelligence_libero_smolvla_lr1e-4bs32steps100000 -/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/transformers/utils/hub.py:111: FutureWarnin -g: Using `TRANSFORMERS_CACHE` is deprecated and will be removed in v5 of Transformers. Use `HF_HOME` instead. - warnings.warn( -INFO 2025-09-09 16:10:03 ils/utils.py:48 Cuda backend detected, using cuda. -WARNING 2025-09-09 16:10:03 /policies.py:81 Device 'None' is not available. Switching to 'cuda'. -INFO 2025-09-09 16:10:03 ts/train.py:137 {'batch_size': 32, - 'dataset': {'episodes': None, - 'image_transforms': {'enable': False, - 'max_num_transforms': 3, - 'random_order': False, - 'tfs': {'brightness': {'kwargs': {'brightness': [0.8, - 1.2]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'contrast': {'kwargs': {'contrast': [0.8, - 1.2]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'hue': {'kwargs': {'hue': [-0.05, - 0.05]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'saturation': {'kwargs': {'saturation': [0.5, - 1.5]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'sharpness': {'kwargs': {'sharpness': [0.5, - 1.5]}, - 'type': 'SharpnessJitter', - 'weight': 1.0}}}, - 'repo_id': 'physical-intelligence/libero', - 'revision': None, - 'root': '/raid/jade/.cache/huggingface/datasets', - 'use_imagenet_stats': True, - 'video_backend': 'torchcodec'}, - 'env': {'camera_name': 'agentview_image,robot0_eye_in_hand_image', - 'episode_length': 520, - 'features': {'action': {'shape': [7], - 'type': }, - 'agent_pos': {'shape': [8], - 'type': }, - 'pixels/agentview_image': {'shape': [360, 360, 3], - 'type': }, - 'pixels/robot0_eye_in_hand_image': {'shape': [360, - 360, - 3], - 'type': }}, - 'features_map': {'action': 'action', - 'agent_pos': 'observation.state', - 'pixels/agentview_image': 'observation.images.image', - 'pixels/robot0_eye_in_hand_image': 'observation.images.image2'}, - 'fps': 30, - 'init_states': True, - 'max_parallel_tasks': 5, - 'multitask_eval': True, - 'obs_type': 'pixels_agent_pos', - 'render_mode': 'rgb_array', - 'task': 'libero_spatial', - 'type': 'libero'}, - 'eval': {'batch_size': 1, 'n_episodes': 1, 'use_async_envs': False}, - 'eval_freq': 0, - 'job_name': 'libero_smolvla', - 'log_freq': 200, - 'num_workers': 4, - 'optimizer': {'betas': [0.9, 0.95], - 'eps': 1e-08, - 'grad_clip_norm': 10, - 'lr': 0.0001, - 'type': 'adamw', - 'weight_decay': 1e-10}, - 'output_dir': '/raid/jade/logs/lerobot/lerobot_2_physical-intelligence_libero_smolvla_lr1e-4bs32steps100000', - 'policy': {'adapt_to_pi_aloha': False, - 'add_image_special_tokens': False, - 'attention_mode': 'cross_attn', - 'chunk_size': 50, - 'device': 'cuda', - 'empty_cameras': 0, - 'expert_width_multiplier': 0.5, - 'freeze_vision_encoder': True, - 'gradient_accumulation_steps': 1, - 'input_features': {}, - 'license': None, - 'load_vlm_weights': False, - 'max_action_dim': 32, - 'max_period': 4.0, - 'max_state_dim': 32, - 'min_period': 0.004, - 'n_action_steps': 1, - 'n_obs_steps': 1, - 'normalization_mapping': {'ACTION': , - 'STATE': , - 'VISUAL': }, - 'num_expert_layers': -1, - 'num_steps': 10, - 'num_vlm_layers': 16, - 'optimizer_betas': [0.9, 0.95], - 'optimizer_eps': 1e-08, - 'optimizer_grad_clip_norm': 10, - 'optimizer_lr': 0.0001, - 'optimizer_weight_decay': 1e-10, - 'output_features': {}, - 'pad_language_to': 'longest', - 'prefix_length': 0, - 'private': None, - 'push_to_hub': True, - 'repo_id': 'None', - 'resize_imgs_with_padding': [512, 512], - 'scheduler_decay_lr': 2.5e-06, - 'scheduler_decay_steps': 30000, - 'scheduler_warmup_steps': 1000, - 'self_attn_every_n_layers': 2, - 'tags': None, - 'tokenizer_max_length': 48, - 'train_expert_only': True, - 'train_state_proj': True, - 'type': 'smolvla', - 'use_amp': True, - 'use_cache': True, - 'use_delta_joint_actions_aloha': False, - 'vlm_model_name': 'HuggingFaceTB/SmolVLM2-500M-Instruct'}, - 'resume': False, - 'save_checkpoint': True, - 'save_freq': 20000, - 'scheduler': {'decay_lr': 2.5e-06, - 'num_decay_steps': 30000, - 'num_warmup_steps': 1000, - 'peak_lr': 0.0001, - 'type': 'cosine_decay_with_warmup'}, - 'seed': 1000, - 'steps': 100000, - 'use_policy_training_preset': True, - 'wandb': {'disable_artifact': False, - 'enable': False, - 'entity': None, - 'mode': None, - 'notes': None, - 'project': 'lerobot', - 'run_id': None}} -INFO 2025-09-09 16:10:03 ts/train.py:143 Logs will be saved locally. -INFO 2025-09-09 16:10:03 ts/train.py:153 Creating dataset -WARNING 2025-09-09 16:10:03 ts/utils.py:302 -The dataset you requested (physical-intelligence/libero) is in 2.0 format. -While current version of LeRobot is backward-compatible with it, the version of your dataset still uses global -stats instead of per-episode stats. Update your dataset stats to the new format using this command: -``` -python -m lerobot.datasets.v21.convert_dataset_v20_to_v21 --repo-id=physical-intelligence/libero -``` - -If you encounter a problem, contact LeRobot maintainers on [Discord](https://discord.com/invite/s3KuuzsPFb) -or open an [issue on GitHub](https://github.com/huggingface/lerobot/issues/new/choose). - -WARNING 2025-09-09 16:10:03 ts/utils.py:302 -The dataset you requested (physical-intelligence/libero) is in 2.0 format. -While current version of LeRobot is backward-compatible with it, the version of your dataset still uses global -stats instead of per-episode stats. Update your dataset stats to the new format using this command: -``` -python -m lerobot.datasets.v21.convert_dataset_v20_to_v21 --repo-id=physical-intelligence/libero -``` - -If you encounter a problem, contact LeRobot maintainers on [Discord](https://discord.com/invite/s3KuuzsPFb) -or open an [issue on GitHub](https://github.com/huggingface/lerobot/issues/new/choose). - -Resolving data files: 100%|█████████████████████████████████| 1693/1693 [00:00<00:00, 54574.89it/s] -Loading dataset shards: 100%|████████████████████████████████████| 70/70 [00:00<00:00, 7567.63it/s] -INFO 2025-09-09 16:10:09 ts/train.py:163 Creating policy -Fetching 2 files: 100%|███████████████████████████████████████████| 2/2 [00:00<00:00, 40721.40it/s] -Fetching 2 files: 100%|████████████████████████████████████████████| 2/2 [00:00<00:00, 7516.67it/s] -Fetching 2 files: 100%|████████████████████████████████████████████| 2/2 [00:00<00:00, 3158.36it/s] -Fetching 2 files: 100%|████████████████████████████████████████████| 2/2 [00:00<00:00, 6775.94it/s] -Reducing the number of VLM layers to 16 ... -INFO 2025-09-09 16:10:41 ts/train.py:168 Creating optimizer and scheduler -INFO 2025-09-09 16:10:41 ts/train.py:180 Output dir: /raid/jade/logs/lerobot/lerobot_2_physical-intelligence_libero_ -smolvla_lr1e-4bs32steps100000 -INFO 2025-09-09 16:10:41 ts/train.py:182 cfg.env.task='libero_spatial' -INFO 2025-09-09 16:10:41 ts/train.py:183 cfg.steps=100000 (100K) -INFO 2025-09-09 16:10:41 ts/train.py:184 dataset.num_frames=273465 (273K) -INFO 2025-09-09 16:10:41 ts/train.py:185 dataset.num_episodes=1693 -INFO 2025-09-09 16:10:41 ts/train.py:186 num_learnable_params=49103712 (49M) -INFO 2025-09-09 16:10:41 ts/train.py:187 num_total_params=399268940 (399M) -INFO 2025-09-09 16:10:41 ts/train.py:225 Start offline training on a fixed dataset -Traceback (most recent call last): - File "/home/jade_choghari/lerobot/src/lerobot/scripts/train.py", line 342, in - main() - File "/home/jade_choghari/lerobot/src/lerobot/scripts/train.py", line 338, in main - train() - File "/home/jade_choghari/lerobot/src/lerobot/configs/parser.py", line 225, in wrapper_inner - response = fn(cfg, *args, **kwargs) - File "/home/jade_choghari/lerobot/src/lerobot/scripts/train.py", line 235, in train - train_tracker, output_dict = update_policy( - File "/home/jade_choghari/lerobot/src/lerobot/scripts/train.py", line 76, in update_policy - grad_scaler.unscale_(optimizer) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/torch/amp/grad_scaler.py", line 342 -, in unscale_ - optimizer_state["found_inf_per_device"] = self._unscale_grads_( - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/torch/amp/grad_scaler.py", line 283 -, in _unscale_grads_ - torch._amp_foreach_non_finite_check_and_unscale_( -RuntimeError: "_amp_foreach_non_finite_check_and_unscale_cuda" not implemented for 'BFloat16' -(lerobot) jade_choghari@hf-dgx-01:~/lerobot$ clear -(lerobot) jade_choghari@hf-dgx-01:~/lerobot$ bash examples/8_train_smolvla_must.sh -Training dir: /raid/jade/logs/lerobot/lerobot_2_physical-intelligence_libero_smolvla_lr1e-4bs32steps100000 -/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/transformers/utils/hub.py:111: FutureWarnin -g: Using `TRANSFORMERS_CACHE` is deprecated and will be removed in v5 of Transformers. Use `HF_HOME` instead. - warnings.warn( -INFO 2025-09-09 16:12:28 ils/utils.py:48 Cuda backend detected, using cuda. -WARNING 2025-09-09 16:12:28 /policies.py:81 Device 'None' is not available. Switching to 'cuda'. -INFO 2025-09-09 16:12:28 ts/train.py:137 {'batch_size': 32, - 'dataset': {'episodes': None, - 'image_transforms': {'enable': False, - 'max_num_transforms': 3, - 'random_order': False, - 'tfs': {'brightness': {'kwargs': {'brightness': [0.8, - 1.2]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'contrast': {'kwargs': {'contrast': [0.8, - 1.2]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'hue': {'kwargs': {'hue': [-0.05, - 0.05]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'saturation': {'kwargs': {'saturation': [0.5, - 1.5]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'sharpness': {'kwargs': {'sharpness': [0.5, - 1.5]}, - 'type': 'SharpnessJitter', - 'weight': 1.0}}}, - 'repo_id': 'physical-intelligence/libero', - 'revision': None, - 'root': '/raid/jade/.cache/huggingface/datasets', - 'use_imagenet_stats': True, - 'video_backend': 'torchcodec'}, - 'env': {'camera_name': 'agentview_image,robot0_eye_in_hand_image', - 'episode_length': 520, - 'features': {'action': {'shape': [7], - 'type': }, - 'agent_pos': {'shape': [8], - 'type': }, - 'pixels/agentview_image': {'shape': [360, 360, 3], - 'type': }, - 'pixels/robot0_eye_in_hand_image': {'shape': [360, - 360, - 3], - 'type': }}, - 'features_map': {'action': 'action', - 'agent_pos': 'observation.state', - 'pixels/agentview_image': 'observation.images.image', - 'pixels/robot0_eye_in_hand_image': 'observation.images.image2'}, - 'fps': 30, - 'init_states': True, - 'max_parallel_tasks': 5, - 'multitask_eval': True, - 'obs_type': 'pixels_agent_pos', - 'render_mode': 'rgb_array', - 'task': 'libero_spatial', - 'type': 'libero'}, - 'eval': {'batch_size': 1, 'n_episodes': 1, 'use_async_envs': False}, - 'eval_freq': 0, - 'job_name': 'libero_smolvla', - 'log_freq': 200, - 'num_workers': 4, - 'optimizer': {'betas': [0.9, 0.95], - 'eps': 1e-08, - 'grad_clip_norm': 10, - 'lr': 0.0001, - 'type': 'adamw', - 'weight_decay': 1e-10}, - 'output_dir': '/raid/jade/logs/lerobot/lerobot_2_physical-intelligence_libero_smolvla_lr1e-4bs32steps100000', - 'policy': {'adapt_to_pi_aloha': False, - 'add_image_special_tokens': False, - 'attention_mode': 'cross_attn', - 'chunk_size': 50, - 'device': 'cuda', - 'empty_cameras': 0, - 'expert_width_multiplier': 0.5, - 'freeze_vision_encoder': True, - 'gradient_accumulation_steps': 1, - 'input_features': {}, - 'license': None, - 'load_vlm_weights': False, - 'max_action_dim': 32, - 'max_period': 4.0, - 'max_state_dim': 32, - 'min_period': 0.004, - 'n_action_steps': 1, - 'n_obs_steps': 1, - 'normalization_mapping': {'ACTION': , - 'STATE': , - 'VISUAL': }, - 'num_expert_layers': -1, - 'num_steps': 10, - 'num_vlm_layers': 16, - 'optimizer_betas': [0.9, 0.95], - 'optimizer_eps': 1e-08, - 'optimizer_grad_clip_norm': 10, - 'optimizer_lr': 0.0001, - 'optimizer_weight_decay': 1e-10, - 'output_features': {}, - 'pad_language_to': 'longest', - 'prefix_length': 0, - 'private': None, - 'push_to_hub': True, - 'repo_id': 'None', - 'resize_imgs_with_padding': [512, 512], - 'scheduler_decay_lr': 2.5e-06, - 'scheduler_decay_steps': 30000, - 'scheduler_warmup_steps': 1000, - 'self_attn_every_n_layers': 2, - 'tags': None, - 'tokenizer_max_length': 48, - 'train_expert_only': True, - 'train_state_proj': True, - 'type': 'smolvla', - 'use_amp': True, - 'use_cache': True, - 'use_delta_joint_actions_aloha': False, - 'vlm_model_name': 'HuggingFaceTB/SmolVLM2-500M-Instruct'}, - 'resume': False, - 'save_checkpoint': True, - 'save_freq': 20000, - 'scheduler': {'decay_lr': 2.5e-06, - 'num_decay_steps': 30000, - 'num_warmup_steps': 1000, - 'peak_lr': 0.0001, - 'type': 'cosine_decay_with_warmup'}, - 'seed': 1000, - 'steps': 100000, - 'use_policy_training_preset': True, - 'wandb': {'disable_artifact': False, - 'enable': False, - 'entity': None, - 'mode': None, - 'notes': None, - 'project': 'lerobot', - 'run_id': None}} -INFO 2025-09-09 16:12:28 ts/train.py:143 Logs will be saved locally. -INFO 2025-09-09 16:12:28 ts/train.py:153 Creating dataset -WARNING 2025-09-09 16:12:28 ts/utils.py:302 -The dataset you requested (physical-intelligence/libero) is in 2.0 format. -While current version of LeRobot is backward-compatible with it, the version of your dataset still uses global -stats instead of per-episode stats. Update your dataset stats to the new format using this command: -``` -python -m lerobot.datasets.v21.convert_dataset_v20_to_v21 --repo-id=physical-intelligence/libero -``` - -If you encounter a problem, contact LeRobot maintainers on [Discord](https://discord.com/invite/s3KuuzsPFb) -or open an [issue on GitHub](https://github.com/huggingface/lerobot/issues/new/choose). - -WARNING 2025-09-09 16:12:28 ts/utils.py:302 -The dataset you requested (physical-intelligence/libero) is in 2.0 format. -While current version of LeRobot is backward-compatible with it, the version of your dataset still uses global -stats instead of per-episode stats. Update your dataset stats to the new format using this command: -``` -python -m lerobot.datasets.v21.convert_dataset_v20_to_v21 --repo-id=physical-intelligence/libero -``` - -If you encounter a problem, contact LeRobot maintainers on [Discord](https://discord.com/invite/s3KuuzsPFb) -or open an [issue on GitHub](https://github.com/huggingface/lerobot/issues/new/choose). - -Resolving data files: 100%|█████████████████████████████████| 1693/1693 [00:00<00:00, 87666.13it/s] -Loading dataset shards: 100%|████████████████████████████████████| 70/70 [00:00<00:00, 4223.20it/s] -INFO 2025-09-09 16:12:34 ts/train.py:163 Creating policy -Fetching 2 files: 100%|███████████████████████████████████████████| 2/2 [00:00<00:00, 43690.67it/s] -Fetching 2 files: 100%|████████████████████████████████████████████| 2/2 [00:00<00:00, 4871.43it/s] -Fetching 2 files: 100%|████████████████████████████████████████████| 2/2 [00:00<00:00, 6512.89it/s] -Fetching 2 files: 100%|███████████████████████████████████████████| 2/2 [00:00<00:00, 43018.50it/s] -Reducing the number of VLM layers to 16 ... -INFO 2025-09-09 16:13:06 ts/train.py:168 Creating optimizer and scheduler -INFO 2025-09-09 16:13:06 ts/train.py:180 Output dir: /raid/jade/logs/lerobot/lerobot_2_physical-intelligence_libero_ -smolvla_lr1e-4bs32steps100000 -INFO 2025-09-09 16:13:06 ts/train.py:182 cfg.env.task='libero_spatial' -INFO 2025-09-09 16:13:06 ts/train.py:183 cfg.steps=100000 (100K) -INFO 2025-09-09 16:13:06 ts/train.py:184 dataset.num_frames=273465 (273K) -INFO 2025-09-09 16:13:06 ts/train.py:185 dataset.num_episodes=1693 -INFO 2025-09-09 16:13:06 ts/train.py:186 num_learnable_params=49103712 (49M) -INFO 2025-09-09 16:13:06 ts/train.py:187 num_total_params=399268940 (399M) -INFO 2025-09-09 16:13:06 ts/train.py:225 Start offline training on a fixed dataset -Traceback (most recent call last): - File "/home/jade_choghari/lerobot/src/lerobot/scripts/train.py", line 342, in - main() - File "/home/jade_choghari/lerobot/src/lerobot/scripts/train.py", line 338, in main - train() - File "/home/jade_choghari/lerobot/src/lerobot/configs/parser.py", line 225, in wrapper_inner - response = fn(cfg, *args, **kwargs) - File "/home/jade_choghari/lerobot/src/lerobot/scripts/train.py", line 235, in train - train_tracker, output_dict = update_policy( - File "/home/jade_choghari/lerobot/src/lerobot/scripts/train.py", line 76, in update_policy - grad_scaler.unscale_(optimizer) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/torch/amp/grad_scaler.py", line 342 -, in unscale_ - optimizer_state["found_inf_per_device"] = self._unscale_grads_( - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/torch/amp/grad_scaler.py", line 283 -, in _unscale_grads_ - torch._amp_foreach_non_finite_check_and_unscale_( -RuntimeError: "_amp_foreach_non_finite_check_and_unscale_cuda" not implemented for 'BFloat16' -(lerobot) jade_choghari@hf-dgx-01:~/lerobot$ bash examples/8_train_smolvla_must.sh -Training dir: /raid/jade/logs/lerobot/lerobot_2_physical-intelligence_libero_smolvla_lr1e-4bs32steps100000 -/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/transformers/utils/hub.py:111: FutureWarnin -g: Using `TRANSFORMERS_CACHE` is deprecated and will be removed in v5 of Transformers. Use `HF_HOME` instead. - warnings.warn( -INFO 2025-09-09 16:13:51 ils/utils.py:48 Cuda backend detected, using cuda. -WARNING 2025-09-09 16:13:51 /policies.py:81 Device 'None' is not available. Switching to 'cuda'. -INFO 2025-09-09 16:13:51 ts/train.py:137 {'batch_size': 32, - 'dataset': {'episodes': None, - 'image_transforms': {'enable': False, - 'max_num_transforms': 3, - 'random_order': False, - 'tfs': {'brightness': {'kwargs': {'brightness': [0.8, - 1.2]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'contrast': {'kwargs': {'contrast': [0.8, - 1.2]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'hue': {'kwargs': {'hue': [-0.05, - 0.05]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'saturation': {'kwargs': {'saturation': [0.5, - 1.5]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'sharpness': {'kwargs': {'sharpness': [0.5, - 1.5]}, - 'type': 'SharpnessJitter', - 'weight': 1.0}}}, - 'repo_id': 'physical-intelligence/libero', - 'revision': None, - 'root': '/raid/jade/.cache/huggingface/datasets', - 'use_imagenet_stats': True, - 'video_backend': 'torchcodec'}, - 'env': {'camera_name': 'agentview_image,robot0_eye_in_hand_image', - 'episode_length': 520, - 'features': {'action': {'shape': [7], - 'type': }, - 'agent_pos': {'shape': [8], - 'type': }, - 'pixels/agentview_image': {'shape': [360, 360, 3], - 'type': }, - 'pixels/robot0_eye_in_hand_image': {'shape': [360, - 360, - 3], - 'type': }}, - 'features_map': {'action': 'action', - 'agent_pos': 'observation.state', - 'pixels/agentview_image': 'observation.images.image', - 'pixels/robot0_eye_in_hand_image': 'observation.images.image2'}, - 'fps': 30, - 'init_states': True, - 'max_parallel_tasks': 5, - 'multitask_eval': True, - 'obs_type': 'pixels_agent_pos', - 'render_mode': 'rgb_array', - 'task': 'libero_spatial', - 'type': 'libero'}, - 'eval': {'batch_size': 1, 'n_episodes': 1, 'use_async_envs': False}, - 'eval_freq': 0, - 'job_name': 'libero_smolvla', - 'log_freq': 200, - 'num_workers': 4, - 'optimizer': {'betas': [0.9, 0.95], - 'eps': 1e-08, - 'grad_clip_norm': 10, - 'lr': 0.0001, - 'type': 'adamw', - 'weight_decay': 1e-10}, - 'output_dir': '/raid/jade/logs/lerobot/lerobot_2_physical-intelligence_libero_smolvla_lr1e-4bs32steps100000', - 'policy': {'adapt_to_pi_aloha': False, - 'add_image_special_tokens': False, - 'attention_mode': 'cross_attn', - 'chunk_size': 50, - 'device': 'cuda', - 'empty_cameras': 0, - 'expert_width_multiplier': 0.5, - 'freeze_vision_encoder': True, - 'gradient_accumulation_steps': 1, - 'input_features': {}, - 'license': None, - 'load_vlm_weights': False, - 'max_action_dim': 32, - 'max_period': 4.0, - 'max_state_dim': 32, - 'min_period': 0.004, - 'n_action_steps': 1, - 'n_obs_steps': 1, - 'normalization_mapping': {'ACTION': , - 'STATE': , - 'VISUAL': }, - 'num_expert_layers': -1, - 'num_steps': 10, - 'num_vlm_layers': 16, - 'optimizer_betas': [0.9, 0.95], - 'optimizer_eps': 1e-08, - 'optimizer_grad_clip_norm': 10, - 'optimizer_lr': 0.0001, - 'optimizer_weight_decay': 1e-10, - 'output_features': {}, - 'pad_language_to': 'longest', - 'prefix_length': 0, - 'private': None, - 'push_to_hub': True, - 'repo_id': 'None', - 'resize_imgs_with_padding': [512, 512], - 'scheduler_decay_lr': 2.5e-06, - 'scheduler_decay_steps': 30000, - 'scheduler_warmup_steps': 1000, - 'self_attn_every_n_layers': 2, - 'tags': None, - 'tokenizer_max_length': 48, - 'train_expert_only': True, - 'train_state_proj': True, - 'type': 'smolvla', - 'use_amp': False, - 'use_cache': True, - 'use_delta_joint_actions_aloha': False, - 'vlm_model_name': 'HuggingFaceTB/SmolVLM2-500M-Instruct'}, - 'resume': False, - 'save_checkpoint': True, - 'save_freq': 20000, - 'scheduler': {'decay_lr': 2.5e-06, - 'num_decay_steps': 30000, - 'num_warmup_steps': 1000, - 'peak_lr': 0.0001, - 'type': 'cosine_decay_with_warmup'}, - 'seed': 1000, - 'steps': 100000, - 'use_policy_training_preset': True, - 'wandb': {'disable_artifact': False, - 'enable': False, - 'entity': None, - 'mode': None, - 'notes': None, - 'project': 'lerobot', - 'run_id': None}} -INFO 2025-09-09 16:13:51 ts/train.py:143 Logs will be saved locally. -INFO 2025-09-09 16:13:51 ts/train.py:153 Creating dataset -WARNING 2025-09-09 16:13:51 ts/utils.py:302 -The dataset you requested (physical-intelligence/libero) is in 2.0 format. -While current version of LeRobot is backward-compatible with it, the version of your dataset still uses global -stats instead of per-episode stats. Update your dataset stats to the new format using this command: -``` -python -m lerobot.datasets.v21.convert_dataset_v20_to_v21 --repo-id=physical-intelligence/libero -``` - -If you encounter a problem, contact LeRobot maintainers on [Discord](https://discord.com/invite/s3KuuzsPFb) -or open an [issue on GitHub](https://github.com/huggingface/lerobot/issues/new/choose). - -WARNING 2025-09-09 16:13:51 ts/utils.py:302 -The dataset you requested (physical-intelligence/libero) is in 2.0 format. -While current version of LeRobot is backward-compatible with it, the version of your dataset still uses global -stats instead of per-episode stats. Update your dataset stats to the new format using this command: -``` -python -m lerobot.datasets.v21.convert_dataset_v20_to_v21 --repo-id=physical-intelligence/libero -``` - -If you encounter a problem, contact LeRobot maintainers on [Discord](https://discord.com/invite/s3KuuzsPFb) -or open an [issue on GitHub](https://github.com/huggingface/lerobot/issues/new/choose). - -Resolving data files: 100%|█████████████████████████████████| 1693/1693 [00:00<00:00, 82981.28it/s] -Loading dataset shards: 100%|████████████████████████████████████| 70/70 [00:00<00:00, 4687.94it/s] -INFO 2025-09-09 16:13:57 ts/train.py:163 Creating policy -Fetching 2 files: 100%|███████████████████████████████████████████| 2/2 [00:00<00:00, 21345.06it/s] -Fetching 2 files: 100%|████████████████████████████████████████████| 2/2 [00:00<00:00, 4226.00it/s] -Fetching 2 files: 100%|████████████████████████████████████████████| 2/2 [00:00<00:00, 2966.27it/s] -Fetching 2 files: 100%|████████████████████████████████████████████| 2/2 [00:00<00:00, 6497.76it/s] -Reducing the number of VLM layers to 16 ... -INFO 2025-09-09 16:14:30 ts/train.py:168 Creating optimizer and scheduler -INFO 2025-09-09 16:14:30 ts/train.py:180 Output dir: /raid/jade/logs/lerobot/lerobot_2_physical-intelligence_libero_ -smolvla_lr1e-4bs32steps100000 -INFO 2025-09-09 16:14:30 ts/train.py:182 cfg.env.task='libero_spatial' -INFO 2025-09-09 16:14:30 ts/train.py:183 cfg.steps=100000 (100K) -INFO 2025-09-09 16:14:30 ts/train.py:184 dataset.num_frames=273465 (273K) -INFO 2025-09-09 16:14:30 ts/train.py:185 dataset.num_episodes=1693 -INFO 2025-09-09 16:14:30 ts/train.py:186 num_learnable_params=49103712 (49M) -INFO 2025-09-09 16:14:30 ts/train.py:187 num_total_params=399268940 (399M) -INFO 2025-09-09 16:14:30 ts/train.py:225 Start offline training on a fixed dataset -INFO 2025-09-09 16:16:20 ts/train.py:255 step:200 smpl:6K ep:40 epch:0.02 loss:1.244 grdn:2.492 lr:1.0e-05 updt_s:0. -536 data_s:0.007 -INFO 2025-09-09 16:17:56 ts/train.py:255 step:400 smpl:13K ep:79 epch:0.05 loss:0.685 grdn:4.262 lr:3.0e-05 updt_s:0 -.481 data_s:0.000 -INFO 2025-09-09 16:19:33 ts/train.py:255 step:600 smpl:19K ep:119 epch:0.07 loss:0.364 grdn:4.849 lr:5.0e-05 updt_s: -0.482 data_s:0.000 -INFO 2025-09-09 16:21:10 ts/train.py:255 step:800 smpl:26K ep:158 epch:0.09 loss:0.239 grdn:4.024 lr:7.0e-05 updt_s: -0.481 data_s:0.000 -INFO 2025-09-09 16:22:46 ts/train.py:255 step:1K smpl:32K ep:198 epch:0.12 loss:0.197 grdn:3.267 lr:9.0e-05 updt_s:0 -.478 data_s:0.000 -INFO 2025-09-09 16:24:22 ts/train.py:255 step:1K smpl:38K ep:238 epch:0.14 loss:0.173 grdn:2.319 lr:1.0e-04 updt_s:0 -.481 data_s:0.000 -INFO 2025-09-09 16:25:59 ts/train.py:255 step:1K smpl:45K ep:277 epch:0.16 loss:0.153 grdn:1.741 lr:1.0e-04 updt_s:0 -.483 data_s:0.000 -INFO 2025-09-09 16:27:36 ts/train.py:255 step:2K smpl:51K ep:317 epch:0.19 loss:0.135 grdn:1.354 lr:9.9e-05 updt_s:0 -.483 data_s:0.000 -INFO 2025-09-09 16:29:14 ts/train.py:255 step:2K smpl:58K ep:357 epch:0.21 loss:0.126 grdn:1.177 lr:9.9e-05 updt_s:0 -.484 data_s:0.000 - diff --git a/src/lerobot/envs/libero copy.py b/src/lerobot/envs/libero copy.py deleted file mode 100644 index 83ccd2fb9..000000000 --- a/src/lerobot/envs/libero copy.py +++ /dev/null @@ -1,326 +0,0 @@ -import math -import os -from collections import defaultdict -from collections.abc import Callable -from itertools import chain -from typing import Any - -import gymnasium as gym -import numpy as np -import torch -from gymnasium import spaces -from libero.libero import benchmark, get_libero_path -from libero.libero.envs import OffScreenRenderEnv - - -def create_libero_envs( - task: str, - n_envs: int, - gym_kwargs: dict[str, Any] = None, - camera_name: str = "agentview_image,robot0_eye_in_hand_image", - init_states: bool = True, - env_cls: Callable = None, - multitask_eval: bool = True, -) -> dict[str, dict[str, Any]]: - """ - Here n_envs is per task and equal to the number of rollouts. - Returns: - dict[str, dict[str, list[LiberoEnv]]]: keys are task_suite and values are list of LiberoEnv envs. - """ - print("num envs", n_envs) - print("multitask_eval", multitask_eval) - print("gym_kwargs", gym_kwargs) - if gym_kwargs is None: - gym_kwargs = {} - - if not multitask_eval: - benchmark_dict = benchmark.get_benchmark_dict() - task_suite = benchmark_dict[task]() # can also choose libero_spatial, libero_object, libero_10 etc. - tasks_id = list(range(len(task_suite.tasks))) - episode_indices = [0 for i in range(len(tasks_id))] - if len(tasks_id) == 1: - tasks_id = [tasks_id[0] for _ in range(n_envs)] - episode_indices = list(range(n_envs)) - elif len(tasks_id) < n_envs and n_envs % len(tasks_id) == 0: - n_repeat = n_envs // len(tasks_id) - print("n_repeat", n_repeat) - episode_indices = [] - for _ in range(len(tasks_id)): - episode_indices.extend(list(range(n_repeat))) - tasks_id = list(chain.from_iterable([[item] * n_repeat for item in tasks_id])) - elif n_envs < len(tasks_id): - tasks_id = tasks_id[:n_envs] - episode_indices = list(range(n_envs))[:n_envs] - print(f"WARNING: n_envs < len(tasks_id), evaluating only on {tasks_id}") - print(f"Creating Libero envs with task ids {tasks_id} from suite {task}") - assert n_envs == len(tasks_id), ( - f"len(n_envs) and tasks_id should be the same, got {n_envs} and {len(tasks_id)}" - ) - return env_cls( - [ - lambda i=i: LiberoEnv( - task_suite=task_suite, - task_id=tasks_id[i], - task_suite_name=task, - camera_name=camera_name, - init_states=init_states, - episode_index=episode_indices[i], - **gym_kwargs, - ) - for i in range(n_envs) - ] - ) - else: - envs = defaultdict(dict) - benchmark_dict = benchmark.get_benchmark_dict() - task = task.split(",") - for _task in task: - task_suite = benchmark_dict[ - _task - ]() # can also choose libero_spatial, libero_object, libero_10 etc. - tasks_ids = list(range(len(task_suite.tasks))) - for tasks_id in tasks_ids: - episode_indices = list(range(n_envs)) - print( - f"Creating Libero envs with task ids {tasks_id} from suite {_task}, episode_indices: {episode_indices}" - ) - envs_list = [ - ( - lambda i=i, - task_suite=task_suite, - tasks_id=tasks_id, - _task=_task, - episode_indices=episode_indices: LiberoEnv( - task_suite=task_suite, - task_id=tasks_id, - task_suite_name=_task, - camera_name=camera_name, - init_states=init_states, - episode_index=episode_indices[i], - **gym_kwargs, - ) - ) - for i in range(n_envs) - ] - envs[_task][tasks_id] = env_cls(envs_list) - return envs - - -def quat2axisangle(quat): - """ - Copied from robosuite: https://github.com/ARISE-Initiative/robosuite/blob/eafb81f54ffc104f905ee48a16bb15f059176ad3/robosuite/utils/transform_utils.py#L490C1-L512C55 - - Converts quaternion to axis-angle format. - Returns a unit vector direction scaled by its angle in radians. - - Args: - quat (np.array): (x,y,z,w) vec4 float angles - - Returns: - np.array: (ax,ay,az) axis-angle exponential coordinates - """ - # clip quaternion - if quat[3] > 1.0: - quat[3] = 1.0 - elif quat[3] < -1.0: - quat[3] = -1.0 - - den = np.sqrt(1.0 - quat[3] * quat[3]) - if math.isclose(den, 0.0): - # This is (close to) a zero degree rotation, immediately return - return np.zeros(3) - - return (quat[:3] * 2.0 * math.acos(quat[3])) / den - - -def get_task_init_states(task_suite, i): - init_states_path = os.path.join( - get_libero_path("init_states"), - task_suite.tasks[i].problem_folder, - task_suite.tasks[i].init_states_file, - ) - init_states = torch.load(init_states_path, weights_only=False) # nosec B614 - return init_states - - -def get_libero_dummy_action(): - """Get dummy/no-op action, used to roll out the simulation while the robot does nothing.""" - return [0, 0, 0, 0, 0, 0, -1] - - -OBS_STATE_DIM = 8 -ACTION_DIM = 7 - - -class LiberoEnv(gym.Env): - metadata = {"render_modes": ["rgb_array"], "render_fps": 80} - - def __init__( - self, - task_suite, - task_id, - task_suite_name, - camera_name="agentview_image,robot0_eye_in_hand_image", - obs_type="pixels", - render_mode="rgb_array", - observation_width=256, - observation_height=256, - visualization_width=640, - visualization_height=480, - init_states=True, - episode_index=0, - ): - super().__init__() - self.task_id = task_id - self.obs_type = obs_type - self.render_mode = render_mode - self.observation_width = observation_width - self.observation_height = observation_height - self.visualization_width = visualization_width - self.visualization_height = visualization_height - self.init_states = init_states - self.camera_name = camera_name.split( - "," - ) # agentview_image (main) or robot0_eye_in_hand_image (wrist) - - # Map raw camera names to "image1" and "image2". - # The preprocessing step `preprocess_observation` will then prefix these with `.images.*`, - # following the LeRobot convention (e.g., `observation.images.image`, `observation.images.image2`). - # This ensures the policy consistently receives observations in the - # expected format regardless of the original camera naming. - self.camera_name_mapping = { - "agentview_image": "image", - "robot0_eye_in_hand_image": "image2", - } - - self.num_steps_wait = ( - 10 # Do nothing for the first few timesteps to wait for the simulator drops objects - ) - self.episode_index = episode_index - - self._env = self._make_envs_task(task_suite, self.task_id) - if task_suite_name == "libero_spatial": - max_steps = 220 # longest training demo has 193 steps - elif task_suite_name == "libero_object": - max_steps = 280 # longest training demo has 254 steps - elif task_suite_name == "libero_goal": - max_steps = 300 # longest training demo has 270 steps - elif task_suite_name == "libero_10": - max_steps = 520 # longest training demo has 505 steps - elif task_suite_name == "libero_90": - max_steps = 400 # longest training demo has 373 steps - self._max_episode_steps = max_steps - - images = {} - for cam in self.camera_name: - images[self.camera_name_mapping[cam]] = spaces.Box( - low=0, - high=255, - shape=(self.observation_height, self.observation_width, 3), - dtype=np.uint8, - ) - - if self.obs_type == "state": - raise NotImplementedError() - elif self.obs_type == "pixels": - self.observation_space = spaces.Dict( - { - "pixels": spaces.Dict(images), - } - ) - elif self.obs_type == "pixels_agent_pos": - self.observation_space = spaces.Dict( - { - "pixels": spaces.Dict(images), - "agent_pos": spaces.Box( - low=-1000.0, - high=1000.0, - shape=(OBS_STATE_DIM,), - dtype=np.float64, - ), - } - ) - - self.action_space = spaces.Box(low=-1, high=1, shape=(ACTION_DIM,), dtype=np.float32) - - def render(self): - raw_obs = self._env.env._get_observations() - image = self._format_raw_obs(raw_obs)["pixels"]["image"] - return image - - def _make_envs_task(self, task_suite, task_id: int = 0): - task = task_suite.get_task(task_id) - self.task = task.name - self.task_description = task.language - task_bddl_file = os.path.join(get_libero_path("bddl_files"), task.problem_folder, task.bddl_file) - - env_args = { - "bddl_file_name": task_bddl_file, - "camera_heights": self.observation_height, - "camera_widths": self.observation_width, - } - env = OffScreenRenderEnv(**env_args) - env.reset() - if self.init_states: - init_states = get_task_init_states( - task_suite, task_id - ) # for benchmarking purpose, we fix the a set of initial states FIXME(mshukor): should be in the reset()? - init_state_id = self.episode_index # episode index - env.set_init_state(init_states[init_state_id]) - - return env - - def _format_raw_obs(self, raw_obs): - images = {} - for camera_name in self.camera_name: - image = raw_obs[camera_name] - image = image[::-1, ::-1] # rotate 180 degrees - images[self.camera_name_mapping[camera_name]] = image - state = np.concatenate( - ( - raw_obs["robot0_eef_pos"], - quat2axisangle(raw_obs["robot0_eef_quat"]), - raw_obs["robot0_gripper_qpos"], - ) - ) - agent_pos = state - if self.obs_type == "state": - raise NotImplementedError() - elif self.obs_type == "pixels": - obs = {"pixels": images.copy()} - elif self.obs_type == "pixels_agent_pos": - obs = { - "pixels": images.copy(), - "agent_pos": agent_pos, - } - return obs - - def reset(self, seed=None, **kwargs): - super().reset(seed=seed) - - self._env.seed(seed) - raw_obs = self._env.reset() - # Do nothing for the first few timesteps to wait for the simulator drops objects - for _ in range(self.num_steps_wait): - raw_obs, _, _, _ = self._env.step(get_libero_dummy_action()) - observation = self._format_raw_obs(raw_obs) - info = {"is_success": False} - return observation, info - - def step(self, action): - assert action.ndim == 1 - raw_obs, reward, done, info = self._env.step(action) - - is_success = self._env.check_success() - terminated = done or is_success - info["is_success"] = done # is_success - - observation = self._format_raw_obs(raw_obs) - if done: - self.reset() - print(self.task, self.task_id, done, is_success) - truncated = False - return observation, reward, terminated, truncated, info - - def close(self): - self._env.close() diff --git a/src/lerobot/envs/libero2.py b/src/lerobot/envs/libero2.py deleted file mode 100644 index 1e794072c..000000000 --- a/src/lerobot/envs/libero2.py +++ /dev/null @@ -1,308 +0,0 @@ -import math -import os -from collections import defaultdict -from itertools import chain -from typing import Any, Callable - -import gymnasium as gym -import numpy as np -import torch -from gymnasium import spaces -from libero.libero import benchmark, get_libero_path -from libero.libero.envs import OffScreenRenderEnv - - -OBS_IMAGE = "observation.image" -OBS_IMAGE_2 = "observation.image2" -def create_libero_envs( - task: str, - n_envs: int, - gym_kwargs: dict[str, Any] = None, - camera_name: str = "agentview_image,robot0_eye_in_hand_image", - init_states: bool = True, - env_cls: Callable = None, - multitask_eval: bool = True, -) -> dict[str, dict[str, Any]]: - """ - Here n_envs is per task and equal to the number of rollouts. - Returns: - dict[str, dict[str, list[LiberoEnv]]]: keys are task_suite and values are list of LiberoEnv envs. - """ - if gym_kwargs is None: - gym_kwargs = {} - - if not multitask_eval: - benchmark_dict = benchmark.get_benchmark_dict() - task_suite = benchmark_dict[task]() # can also choose libero_spatial, libero_object, libero_10 etc. - tasks_id = list(range(len(task_suite.tasks))) - episode_indices = [0 for i in range(len(tasks_id))] - if len(tasks_id) == 1: - tasks_id = [tasks_id[0] for _ in range(n_envs)] - episode_indices = list(range(n_envs)) - elif len(tasks_id) < n_envs and n_envs % len(tasks_id) == 0: - n_repeat = n_envs // len(tasks_id) - episode_indices = [] - for i in range(len(tasks_id)): - episode_indices.extend(list(range(n_repeat))) - tasks_id = list(chain.from_iterable([[item] * n_repeat for item in tasks_id])) - elif n_envs < len(tasks_id): - tasks_id = tasks_id[:n_envs] - episode_indices = list(range(n_envs))[:n_envs] - print(f"WARNING: n_envs < len(tasks_id), evaluating only on {tasks_id}") - print(f"Creating Libero envs with task ids {tasks_id} from suite {task}") - assert n_envs == len( - tasks_id - ), f"len(n_envs) and tasks_id should be the same, got {n_envs} and {len(tasks_id)}" - return env_cls( - [ - lambda i=i: LiberoEnv( - task_suite=task_suite, - task_id=tasks_id[i], - task_suite_name=task, - camera_name=camera_name, - init_states=init_states, - episode_index=episode_indices[i], - **gym_kwargs, - ) - for i in range(n_envs) - ] - ) - else: - envs = defaultdict(dict) - benchmark_dict = benchmark.get_benchmark_dict() - task = task.split(",") - for _task in task: - task_suite = benchmark_dict[ - _task - ]() # can also choose libero_spatial, libero_object, libero_10 etc. - tasks_ids = list(range(len(task_suite.tasks))) - # tasks_ids = [0] # FIXME(mshukor): debug - for tasks_id in tasks_ids: - episode_indices = list(range(n_envs)) - print( - f"Creating Libero envs with task ids {tasks_id} from suite {_task}, episode_indices: {episode_indices}" - ) - envs_list = [ - lambda i=i: LiberoEnv( - task_suite=task_suite, - task_id=tasks_id, - task_suite_name=_task, - camera_name=camera_name, - init_states=init_states, - episode_index=episode_indices[i], - **gym_kwargs, - ) - for i in range(n_envs) - ] - envs[_task][tasks_id] = env_cls(envs_list) - return envs - - -def quat2axisangle(quat): - """ - Copied from robosuite: https://github.com/ARISE-Initiative/robosuite/blob/eafb81f54ffc104f905ee48a16bb15f059176ad3/robosuite/utils/transform_utils.py#L490C1-L512C55 - - Converts quaternion to axis-angle format. - Returns a unit vector direction scaled by its angle in radians. - - Args: - quat (np.array): (x,y,z,w) vec4 float angles - - Returns: - np.array: (ax,ay,az) axis-angle exponential coordinates - """ - # clip quaternion - if quat[3] > 1.0: - quat[3] = 1.0 - elif quat[3] < -1.0: - quat[3] = -1.0 - - den = np.sqrt(1.0 - quat[3] * quat[3]) - if math.isclose(den, 0.0): - # This is (close to) a zero degree rotation, immediately return - return np.zeros(3) - - return (quat[:3] * 2.0 * math.acos(quat[3])) / den - - -def get_task_init_states(task_suite, i): - init_states_path = os.path.join( - get_libero_path("init_states"), - task_suite.tasks[i].problem_folder, - task_suite.tasks[i].init_states_file, - ) - init_states = torch.load(init_states_path, weights_only=False) - return init_states - - -def get_libero_dummy_action(): - """Get dummy/no-op action, used to roll out the simulation while the robot does nothing.""" - return [0, 0, 0, 0, 0, 0, -1] - - -class LiberoEnv(gym.Env): - metadata = {"render_modes": ["rgb_array"], "render_fps": 80} - - def __init__( - self, - task_suite, - task_id, - task_suite_name, - camera_name="agentview_image,robot0_eye_in_hand_image", - obs_type="pixels", - render_mode="rgb_array", - observation_width=256, - observation_height=256, - visualization_width=640, - visualization_height=480, - init_states=True, - episode_index=0, - ): - super().__init__() - self.task_id = task_id - self.obs_type = obs_type - self.render_mode = render_mode - self.observation_width = observation_width - self.observation_height = observation_height - self.visualization_width = visualization_width - self.visualization_height = visualization_height - self.init_states = init_states - self.camera_name = camera_name.split( - "," - ) # agentview_image (main) or robot0_eye_in_hand_image (wrist) - self.camera_name_mapping = { - "agentview_image": OBS_IMAGE, - "robot0_eye_in_hand_image": OBS_IMAGE_2, - } - self.num_steps_wait = ( - 10 # Do nothing for the first few timesteps to wait for the simulator drops objects - ) - self.episode_index = episode_index - - self._env = self._make_envs_task(task_suite, self.task_id) - if task_suite_name == "libero_spatial": - max_steps = 220 # longest training demo has 193 steps - elif task_suite_name == "libero_object": - max_steps = 280 # longest training demo has 254 steps - elif task_suite_name == "libero_goal": - max_steps = 300 # longest training demo has 270 steps - elif task_suite_name == "libero_10": - max_steps = 520 # longest training demo has 505 steps - elif task_suite_name == "libero_90": - max_steps = 400 # longest training demo has 373 steps - self._max_episode_steps = max_steps - - images = {} - for cam in self.camera_name: - images[self.camera_name_mapping[cam]] = spaces.Box( - low=0, - high=255, - shape=(self.observation_height, self.observation_width, 3), - dtype=np.uint8, - ) - - if self.obs_type == "state": - raise NotImplementedError() - elif self.obs_type == "pixels": - self.observation_space = spaces.Dict( - { - "pixels": spaces.Dict(images), - } - ) - elif self.obs_type == "pixels_agent_pos": - self.observation_space = spaces.Dict( - { - "pixels": spaces.Dict(images), - "agent_pos": spaces.Box( - low=-1000.0, - high=1000.0, - shape=(8,), - dtype=np.float64, - ), - } - ) - - self.action_space = spaces.Box(low=-1, high=1, shape=(7,), dtype=np.float32) - - def render(self): - raw_obs = self._env.env._get_observations() - image = self._format_raw_obs(raw_obs)["pixels"][OBS_IMAGE] - return image - - def _make_envs_task(self, task_suite, task_id: int = 0): - task = task_suite.get_task(task_id) - self.task = task.name - self.task_description = task.language - task_bddl_file = os.path.join(get_libero_path("bddl_files"), task.problem_folder, task.bddl_file) - - env_args = { - "bddl_file_name": task_bddl_file, - "camera_heights": self.observation_height, - "camera_widths": self.observation_width, - } - env = OffScreenRenderEnv(**env_args) - env.reset() - if self.init_states: - init_states = get_task_init_states( - task_suite, task_id - ) # for benchmarking purpose, we fix the a set of initial states FIXME(mshukor): should be in the reset()? - init_state_id = self.episode_index # episode index - env.set_init_state(init_states[init_state_id]) - - return env - - def _format_raw_obs(self, raw_obs): - images = {} - for camera_name in self.camera_name: - image = raw_obs[camera_name] - image = image[::-1, ::-1] # rotate 180 degrees - images[self.camera_name_mapping[camera_name]] = image - # images = image if len(images) == 1 else images - state = np.concatenate( - ( - raw_obs["robot0_eef_pos"], - quat2axisangle(raw_obs["robot0_eef_quat"]), - raw_obs["robot0_gripper_qpos"], - ) - ) - agent_pos = state - if self.obs_type == "state": - raise NotImplementedError() - elif self.obs_type == "pixels": - obs = {"pixels": images.copy()} - elif self.obs_type == "pixels_agent_pos": - obs = { - "pixels": images.copy(), - "agent_pos": agent_pos, - } - return obs - - def reset(self, seed=None, **kwargs): - super().reset(seed=seed) - - self._env.seed(seed) - raw_obs = self._env.reset() - # Do nothing for the first few timesteps to wait for the simulator drops objects - for _ in range(self.num_steps_wait): - raw_obs, _, _, _ = self._env.step(get_libero_dummy_action()) - observation = self._format_raw_obs(raw_obs) - info = {"is_success": False} - return observation, info - - def step(self, action): - assert action.ndim == 1 - raw_obs, reward, done, info = self._env.step(action) - - is_success = self._env.check_success() - terminated = done or is_success - info["is_success"] = done # is_success - - observation = self._format_raw_obs(raw_obs) - if done: - self.reset() - print(self.task, self.task_id, done, is_success) - truncated = False - return observation, reward, terminated, truncated, info - - def close(self): - self._env.close() diff --git a/tmux_log.txt b/tmux_log.txt deleted file mode 100644 index 4936578bc..000000000 --- a/tmux_log.txt +++ /dev/null @@ -1,2008 +0,0 @@ - 'resume': False, - 'save_checkpoint': True, - 'save_freq': 20000, - 'scheduler': {'decay_lr': 2.5e-06, - 'num_decay_steps': 30000, - 'num_warmup_steps': 1000, - 'peak_lr': 0.0001, - 'type': 'cosine_decay_with_warmup'}, - 'seed': 1000, - 'steps': 100000, - 'use_policy_training_preset': True, - 'wandb': {'disable_artifact': False, - 'enable': False, - 'entity': None, - 'mode': None, - 'notes': None, - 'project': 'lerobot', - 'run_id': None}} -INFO 2025-09-08 13:23:15 ils/utils.py:48 Cuda backend detected, using cuda. -WARNING 2025-09-08 13:23:15 /policies.py:81 Device 'None' is not available. Switching to 'cuda'. -INFO 2025-09-08 13:23:15 ccelerate.py:99 {'batch_size': 32, - 'dataset': {'episodes': None, - 'image_transforms': {'enable': False, - 'max_num_transforms': 3, - 'random_order': False, - 'tfs': {'brightness': {'kwargs': {'brightness': [0.8, - 1.2]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'contrast': {'kwargs': {'contrast': [0.8, - 1.2]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'hue': {'kwargs': {'hue': [-0.05, - 0.05]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'saturation': {'kwargs': {'saturation': [0.5, - 1.5]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'sharpness': {'kwargs': {'sharpness': [0.5, - 1.5]}, - 'type': 'SharpnessJitter', - 'weight': 1.0}}}, - 'repo_id': 'HuggingFaceVLA/libero', - 'revision': None, - 'root': '/raid/jade/.cache/huggingface/datasets', - 'use_imagenet_stats': True, - 'video_backend': 'torchcodec'}, - 'env': {'camera_name': 'agentview_image,robot0_eye_in_hand_image', - 'episode_length': 520, - 'features': {'action': {'shape': [7], - 'type': }, - 'agent_pos': {'shape': [8], - 'type': }, - 'pixels/agentview_image': {'shape': [360, 360, 3], - 'type': }, - 'pixels/robot0_eye_in_hand_image': {'shape': [360, - 360, - 3], - 'type': }}, - 'features_map': {'action': 'action', - 'agent_pos': 'observation.state', - 'pixels/agentview_image': 'observation.images.image', - 'pixels/robot0_eye_in_hand_image': 'observation.images.image2'}, - 'fps': 30, - 'init_states': True, - 'max_parallel_tasks': 5, - 'multitask_eval': True, - 'obs_type': 'pixels_agent_pos', - 'render_mode': 'rgb_array', - 'task': 'libero_spatial', - 'type': 'libero'}, - 'eval': {'batch_size': 1, 'n_episodes': 1, 'use_async_envs': False}, - 'eval_freq': 0, - 'job_name': 'libero_smolvla', - 'log_freq': 200, - 'num_workers': 4, - 'optimizer': {'betas': [0.9, 0.95], - 'eps': 1e-08, - 'grad_clip_norm': 10, - 'lr': 0.0001, - 'type': 'adamw', - 'weight_decay': 1e-10}, - 'output_dir': '/raid/jade/logs/lerobot/lerobot_2_HuggingFaceVLA_libero_smolvla_lr1e-4bs32steps100000', - 'policy': {'adapt_to_pi_aloha': False, - 'add_image_special_tokens': False, - 'attention_mode': 'cross_attn', - 'chunk_size': 50, - 'device': 'cuda', - 'empty_cameras': 0, - 'expert_width_multiplier': 0.75, - 'freeze_vision_encoder': True, - 'gradient_accumulation_steps': 1, - 'input_features': {}, - 'license': None, - 'load_vlm_weights': False, - 'max_action_dim': 32, - 'max_period': 4.0, - 'max_state_dim': 32, - 'min_period': 0.004, - 'n_action_steps': 1, - 'n_obs_steps': 1, - 'normalization_mapping': {'ACTION': , - 'STATE': , - 'VISUAL': }, - 'num_expert_layers': -1, - 'num_steps': 10, - 'num_vlm_layers': 16, - 'optimizer_betas': [0.9, 0.95], - 'optimizer_eps': 1e-08, - 'optimizer_grad_clip_norm': 10, - 'optimizer_lr': 0.0001, - 'optimizer_weight_decay': 1e-10, - 'output_features': {}, - 'pad_language_to': 'longest', - 'prefix_length': -1, - 'private': None, - 'push_to_hub': True, - 'repo_id': 'None', - 'resize_imgs_with_padding': [512, 512], - 'scheduler_decay_lr': 2.5e-06, - 'scheduler_decay_steps': 30000, - 'scheduler_warmup_steps': 1000, - 'self_attn_every_n_layers': 2, - 'tags': None, - 'tokenizer_max_length': 48, - 'train_expert_only': True, - 'train_state_proj': True, - 'type': 'smolvla', - 'use_amp': True, - 'use_cache': True, - 'use_delta_joint_actions_aloha': False, - 'vlm_model_name': 'HuggingFaceTB/SmolVLM2-500M-Instruct'}, - 'resume': False, - 'save_checkpoint': True, - 'save_freq': 20000, - 'scheduler': {'decay_lr': 2.5e-06, - 'num_decay_steps': 30000, - 'num_warmup_steps': 1000, - 'peak_lr': 0.0001, - 'type': 'cosine_decay_with_warmup'}, - 'seed': 1000, - 'steps': 100000, - 'use_policy_training_preset': True, - 'wandb': {'disable_artifact': False, - 'enable': False, - 'entity': None, - 'mode': None, - 'notes': None, - 'project': 'lerobot', - 'run_id': None}} -WARNING 2025-09-08 13:23:15 ls/other.py:512 Detected kernel version 5.4.0, which is below the recommended minimum of - 5.5.0; this can cause the process to hang. It is recommended to upgrade the kernel to the minimum version or higher -. -WARNING 2025-09-08 13:23:15 ls/other.py:512 Detected kernel version 5.4.0, which is below the recommended minimum of - 5.5.0; this can cause the process to hang. It is recommended to upgrade the kernel to the minimum version or higher -. -INFO 2025-09-08 13:23:15 celerate.py:149 Creating dataset -Resolving data files: 100%|█████████████████████████████████| 1693/1693 [00:00<00:00, 35414.48it/s] -Loading dataset shards: 100%|████████████████████████████████████| 69/69 [00:00<00:00, 5660.00it/s] -Resolving data files: 100%|█████████████████████████████████| 1693/1693 [00:00<00:00, 43760.67it/s] -Loading dataset shards: 100%|████████████████████████████████████| 69/69 [00:00<00:00, 5629.72it/s] -c -INFO 2025-09-08 13:23:22 celerate.py:160 Creating policy -/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/transformers/utils/hub.py:111: FutureWarnin -g: Using `TRANSFORMERS_CACHE` is deprecated and will be removed in v5 of Transformers. Use `HF_HOME` instead. - warnings.warn( -c -/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/torch/distributed/distributed_c10d.py:4631: - UserWarning: No device id is provided via `init_process_group` or `barrier `. Using the current device set by the u -ser. - warnings.warn( # warn only once -[rank1]:[W908 13:23:22.785516795 ProcessGroupNCCL.cpp:4718] [PG ID 0 PG GUID 0 Rank 1] using GPU 1 as device used b -y this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You - can pecify device_id in init_process_group() to force use of a particular device. -Reducing the number of VLM layers to 16 ... -/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/torch/distributed/distributed_c10d.py:4631: - UserWarning: No device id is provided via `init_process_group` or `barrier `. Using the current device set by the u -ser. - warnings.warn( # warn only once -[rank0]:[W908 13:23:43.028071493 ProcessGroupNCCL.cpp:4718] [PG ID 0 PG GUID 0 Rank 0] using GPU 0 as device used b -y this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You - can pecify device_id in init_process_group() to force use of a particular device. -INFO 2025-09-08 13:23:43 celerate.py:171 Creating optimizer and scheduler -/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/transformers/utils/hub.py:111: FutureWarnin -g: Using `TRANSFORMERS_CACHE` is deprecated and will be removed in v5 of Transformers. Use `HF_HOME` instead. - warnings.warn( -Reducing the number of VLM layers to 16 ... -INFO 2025-09-08 13:24:04 celerate.py:211 Output dir: /raid/jade/logs/lerobot/lerobot_2_HuggingFaceVLA_libero_smolvla -_lr1e-4bs32steps100000 -INFO 2025-09-08 13:24:04 celerate.py:213 cfg.env.task='libero_spatial' -INFO 2025-09-08 13:24:04 celerate.py:214 cfg.steps=100000 (100K) -INFO 2025-09-08 13:24:04 celerate.py:215 dataset.num_frames=273465 (273K) -INFO 2025-09-08 13:24:04 celerate.py:216 dataset.num_episodes=1693 -INFO 2025-09-08 13:24:04 celerate.py:217 num_learnable_params=99880992 (100M) -INFO 2025-09-08 13:24:04 celerate.py:218 num_total_params=450046220 (450M) -INFO 2025-09-08 13:24:04 celerate.py:219 Number of processes: 2 -INFO 2025-09-08 13:24:04 celerate.py:220 Device: cuda:0 -INFO 2025-09-08 13:24:04 celerate.py:221 Mixed precision: bf16 -INFO 2025-09-08 13:24:04 celerate.py:243 Start offline training on a fixed dataset - -bach: dict_keys(['observation.images.image', 'observation.images.image2', 'observation.state', 'action', 'timestamp -', 'frame_index', 'episode_index', 'index', 'task_index', 'observation.images.image_is_pad', 'observation.images.ima -ge2_is_pad', 'observation.state_is_pad', 'action_is_pad', 'task']) -> /home/jade_choghari/lerobot/src/lerobot/scripts/train_accelerate.py(263)train() --> train_tracker, output_dict = update_policy( -(Pdb) -bach: dict_keys(['observation.images.image', 'observation.images.image2', 'observation.state', 'action', 'timestamp -', 'frame_index', 'episode_index', 'index', 'task_index', 'observation.images.image_is_pad', 'observation.images.ima -ge2_is_pad', 'observation.state_is_pad', 'action_is_pad', 'task']) -> /home/jade_choghari/lerobot/src/lerobot/scripts/train_accelerate.py(263)train() --> train_tracker, output_dict = update_policy( -(Pdb) batch.keys()[rank0]:[W908 13:24:43.868440913 reducer.cpp:1430] Warning: find_unused_parameters=True was specif -ied in DDP constructor, but did not find any unused parameters in the forward pass. This flag results in an extra tr -aversal of the autograd graph every iteration, which can adversely affect performance. If your model indeed never h -as any unused parameters in the forward pass, consider turning this flag off. Note that this warning may be a false -positive if your model has flow control causing later iterations to have unused parameters. (function operator()) -policy.config.input_features -*** SyntaxError: invalid syntax -(Pdb) policy.config.input_features -*** AttributeError: 'DistributedDataParallel' object has no attribute 'config' -(Pdb) policy -DistributedDataParallel( - (module): SmolVLAPolicy( - (normalize_inputs): Normalize( - (buffer_observation_state): ParameterDict( - (mean): Parameter containing: [torch.cuda.FloatTensor of size 8 (cuda:1)] - (std): Parameter containing: [torch.cuda.FloatTensor of size 8 (cuda:1)] - ) - ) - (normalize_targets): Normalize( - (buffer_action): ParameterDict( - (mean): Parameter containing: [torch.cuda.FloatTensor of size 7 (cuda:1)] - (std): Parameter containing: [torch.cuda.FloatTensor of size 7 (cuda:1)] - ) - ) - (unnormalize_outputs): Unnormalize( - (buffer_action): ParameterDict( - (mean): Parameter containing: [torch.cuda.FloatTensor of size 7 (cuda:1)] - (std): Parameter containing: [torch.cuda.FloatTensor of size 7 (cuda:1)] - ) - ) - (model): VLAFlowMatching( - (vlm_with_expert): SmolVLMWithExpertModel( - (vlm): SmolVLMForConditionalGeneration( - (model): SmolVLMModel( - (vision_model): SmolVLMVisionTransformer( - (embeddings): SmolVLMVisionEmbeddings( - (patch_embedding): Conv2d(3, 768, kernel_size=(16, 16), stride=(16, 16), padding=valid) - (position_embedding): Embedding(1024, 768) - ) - (encoder): SmolVLMEncoder( - (layers): ModuleList( - (0-11): 12 x SmolVLMEncoderLayer( - (self_attn): SmolVLMVisionAttention( - (k_proj): Linear(in_features=768, out_features=768, bias=True) - (v_proj): Linear(in_features=768, out_features=768, bias=True) - (q_proj): Linear(in_features=768, out_features=768, bias=True) - (out_proj): Linear(in_features=768, out_features=768, bias=True) - ) - (layer_norm1): LayerNorm((768,), eps=1e-06, elementwise_affine=True) - (mlp): SmolVLMVisionMLP( - (activation_fn): PytorchGELUTanh() - (fc1): Linear(in_features=768, out_features=3072, bias=True) - (fc2): Linear(in_features=3072, out_features=768, bias=True) - ) - (layer_norm2): LayerNorm((768,), eps=1e-06, elementwise_affine=True) - ) - ) - ) - (post_layernorm): LayerNorm((768,), eps=1e-06, elementwise_affine=True) - ) - (connector): SmolVLMConnector( - (modality_projection): SmolVLMSimpleMLP( - (proj): Linear(in_features=12288, out_features=960, bias=False) - ) - ) - (text_model): LlamaModel( - (embed_tokens): Embedding(49280, 960, padding_idx=2) - (layers): ModuleList( - (0-15): 16 x LlamaDecoderLayer( - (self_attn): LlamaAttention( - (q_proj): Linear(in_features=960, out_features=960, bias=False) - (k_proj): Linear(in_features=960, out_features=320, bias=False) - (v_proj): Linear(in_features=960, out_features=320, bias=False) - (o_proj): Linear(in_features=960, out_features=960, bias=False) - ) - (mlp): LlamaMLP( - (gate_proj): Linear(in_features=960, out_features=2560, bias=False) - (up_proj): Linear(in_features=960, out_features=2560, bias=False) - (down_proj): Linear(in_features=2560, out_features=960, bias=False) - (act_fn): SiLU() - ) - (input_layernorm): LlamaRMSNorm((960,), eps=1e-05) - (post_attention_layernorm): LlamaRMSNorm((960,), eps=1e-05) - ) - ) - (norm): LlamaRMSNorm((960,), eps=1e-05) - (rotary_emb): LlamaRotaryEmbedding() - ) - ) - (lm_head): Linear(in_features=960, out_features=49280, bias=False) - ) - (lm_expert): LlamaModel( - (embed_tokens): None - (layers): ModuleList( - (0): LlamaDecoderLayer( - (self_attn): LlamaAttention( - (q_proj): Linear(in_features=720, out_features=960, bias=False) - (k_proj): Linear(in_features=720, out_features=320, bias=False) - (v_proj): Linear(in_features=720, out_features=320, bias=False) - (o_proj): Linear(in_features=960, out_features=720, bias=False) - ) - (mlp): LlamaMLP( - (gate_proj): Linear(in_features=720, out_features=2048, bias=False) - (up_proj): Linear(in_features=720, out_features=2048, bias=False) - (down_proj): Linear(in_features=2048, out_features=720, bias=False) - (act_fn): SiLU() - ) - (input_layernorm): LlamaRMSNorm((720,), eps=1e-05) - (post_attention_layernorm): LlamaRMSNorm((720,), eps=1e-05) - ) - (1): LlamaDecoderLayer( - (self_attn): LlamaAttention( - (q_proj): Linear(in_features=720, out_features=960, bias=False) - (k_proj): Linear(in_features=320, out_features=320, bias=False) - (v_proj): Linear(in_features=320, out_features=320, bias=False) - (o_proj): Linear(in_features=960, out_features=720, bias=False) - ) - (mlp): LlamaMLP( - (gate_proj): Linear(in_features=720, out_features=2048, bias=False) - (up_proj): Linear(in_features=720, out_features=2048, bias=False) - (down_proj): Linear(in_features=2048, out_features=720, bias=False) - (act_fn): SiLU() - ) - (input_layernorm): LlamaRMSNorm((720,), eps=1e-05) - (post_attention_layernorm): LlamaRMSNorm((720,), eps=1e-05) - ) - (2): LlamaDecoderLayer( - (self_attn): LlamaAttention( - (q_proj): Linear(in_features=720, out_features=960, bias=False) - (k_proj): Linear(in_features=720, out_features=320, bias=False) - (v_proj): Linear(in_features=720, out_features=320, bias=False) - (o_proj): Linear(in_features=960, out_features=720, bias=False) - ) - (mlp): LlamaMLP( - (gate_proj): Linear(in_features=720, out_features=2048, bias=False) - (up_proj): Linear(in_features=720, out_features=2048, bias=False) - (down_proj): Linear(in_features=2048, out_features=720, bias=False) - (act_fn): SiLU() - ) - (input_layernorm): LlamaRMSNorm((720,), eps=1e-05) - (post_attention_layernorm): LlamaRMSNorm((720,), eps=1e-05) - ) - (3): LlamaDecoderLayer( - (self_attn): LlamaAttention( - (q_proj): Linear(in_features=720, out_features=960, bias=False) - (k_proj): Linear(in_features=320, out_features=320, bias=False) - (v_proj): Linear(in_features=320, out_features=320, bias=False) - (o_proj): Linear(in_features=960, out_features=720, bias=False) - ) - (mlp): LlamaMLP( - (gate_proj): Linear(in_features=720, out_features=2048, bias=False) - (up_proj): Linear(in_features=720, out_features=2048, bias=False) - (down_proj): Linear(in_features=2048, out_features=720, bias=False) - (act_fn): SiLU() - ) - (input_layernorm): LlamaRMSNorm((720,), eps=1e-05) - (post_attention_layernorm): LlamaRMSNorm((720,), eps=1e-05) - ) - (4): LlamaDecoderLayer( - (self_attn): LlamaAttention( - (q_proj): Linear(in_features=720, out_features=960, bias=False) - (k_proj): Linear(in_features=720, out_features=320, bias=False) - (v_proj): Linear(in_features=720, out_features=320, bias=False) - (o_proj): Linear(in_features=960, out_features=720, bias=False) - ) - (mlp): LlamaMLP( - (gate_proj): Linear(in_features=720, out_features=2048, bias=False) - (up_proj): Linear(in_features=720, out_features=2048, bias=False) - (down_proj): Linear(in_features=2048, out_features=720, bias=False) - (act_fn): SiLU() - ) - (input_layernorm): LlamaRMSNorm((720,), eps=1e-05) - (post_attention_layernorm): LlamaRMSNorm((720,), eps=1e-05) - ) - (5): LlamaDecoderLayer( - (self_attn): LlamaAttention( - (q_proj): Linear(in_features=720, out_features=960, bias=False) - (k_proj): Linear(in_features=320, out_features=320, bias=False) - (v_proj): Linear(in_features=320, out_features=320, bias=False) - (o_proj): Linear(in_features=960, out_features=720, bias=False) - ) - (mlp): LlamaMLP( - (gate_proj): Linear(in_features=720, out_features=2048, bias=False) - (up_proj): Linear(in_features=720, out_features=2048, bias=False) - (down_proj): Linear(in_features=2048, out_features=720, bias=False) - (act_fn): SiLU() - ) - (input_layernorm): LlamaRMSNorm((720,), eps=1e-05) - (post_attention_layernorm): LlamaRMSNorm((720,), eps=1e-05) - ) - (6): LlamaDecoderLayer( - (self_attn): LlamaAttention( - (q_proj): Linear(in_features=720, out_features=960, bias=False) - (k_proj): Linear(in_features=720, out_features=320, bias=False) - (v_proj): Linear(in_features=720, out_features=320, bias=False) - (o_proj): Linear(in_features=960, out_features=720, bias=False) - ) - (mlp): LlamaMLP( - (gate_proj): Linear(in_features=720, out_features=2048, bias=False) - (up_proj): Linear(in_features=720, out_features=2048, bias=False) - (down_proj): Linear(in_features=2048, out_features=720, bias=False) - (act_fn): SiLU() - ) - (input_layernorm): LlamaRMSNorm((720,), eps=1e-05) - (post_attention_layernorm): LlamaRMSNorm((720,), eps=1e-05) - ) - (7): LlamaDecoderLayer( - (self_attn): LlamaAttention( - (q_proj): Linear(in_features=720, out_features=960, bias=False) - (k_proj): Linear(in_features=320, out_features=320, bias=False) - (v_proj): Linear(in_features=320, out_features=320, bias=False) - (o_proj): Linear(in_features=960, out_features=720, bias=False) - ) - (mlp): LlamaMLP( - (gate_proj): Linear(in_features=720, out_features=2048, bias=False) - (up_proj): Linear(in_features=720, out_features=2048, bias=False) - (down_proj): Linear(in_features=2048, out_features=720, bias=False) - (act_fn): SiLU() - ) - (input_layernorm): LlamaRMSNorm((720,), eps=1e-05) - (post_attention_layernorm): LlamaRMSNorm((720,), eps=1e-05) - ) - (8): LlamaDecoderLayer( - (self_attn): LlamaAttention( - (q_proj): Linear(in_features=720, out_features=960, bias=False) - (k_proj): Linear(in_features=720, out_features=320, bias=False) - (v_proj): Linear(in_features=720, out_features=320, bias=False) - (o_proj): Linear(in_features=960, out_features=720, bias=False) - ) - (mlp): LlamaMLP( - (gate_proj): Linear(in_features=720, out_features=2048, bias=False) - (up_proj): Linear(in_features=720, out_features=2048, bias=False) - (down_proj): Linear(in_features=2048, out_features=720, bias=False) - (act_fn): SiLU() - ) - (input_layernorm): LlamaRMSNorm((720,), eps=1e-05) - (post_attention_layernorm): LlamaRMSNorm((720,), eps=1e-05) - ) - (9): LlamaDecoderLayer( - (self_attn): LlamaAttention( - (q_proj): Linear(in_features=720, out_features=960, bias=False) - (k_proj): Linear(in_features=320, out_features=320, bias=False) - (v_proj): Linear(in_features=320, out_features=320, bias=False) - (o_proj): Linear(in_features=960, out_features=720, bias=False) - ) - (mlp): LlamaMLP( - (gate_proj): Linear(in_features=720, out_features=2048, bias=False) - (up_proj): Linear(in_features=720, out_features=2048, bias=False) - (down_proj): Linear(in_features=2048, out_features=720, bias=False) - (act_fn): SiLU() - ) - (input_layernorm): LlamaRMSNorm((720,), eps=1e-05) - (post_attention_layernorm): LlamaRMSNorm((720,), eps=1e-05) - ) - (10): LlamaDecoderLayer( - (self_attn): LlamaAttention( - (q_proj): Linear(in_features=720, out_features=960, bias=False) - (k_proj): Linear(in_features=720, out_features=320, bias=False) - (v_proj): Linear(in_features=720, out_features=320, bias=False) - (o_proj): Linear(in_features=960, out_features=720, bias=False) - ) - (mlp): LlamaMLP( - (gate_proj): Linear(in_features=720, out_features=2048, bias=False) - (up_proj): Linear(in_features=720, out_features=2048, bias=False) - (down_proj): Linear(in_features=2048, out_features=720, bias=False) - (act_fn): SiLU() - ) - (input_layernorm): LlamaRMSNorm((720,), eps=1e-05) - (post_attention_layernorm): LlamaRMSNorm((720,), eps=1e-05) - ) - (11): LlamaDecoderLayer( - (self_attn): LlamaAttention( - (q_proj): Linear(in_features=720, out_features=960, bias=False) - (k_proj): Linear(in_features=320, out_features=320, bias=False) - (v_proj): Linear(in_features=320, out_features=320, bias=False) - (o_proj): Linear(in_features=960, out_features=720, bias=False) - ) - (mlp): LlamaMLP( - (gate_proj): Linear(in_features=720, out_features=2048, bias=False) - (up_proj): Linear(in_features=720, out_features=2048, bias=False) - (down_proj): Linear(in_features=2048, out_features=720, bias=False) - (act_fn): SiLU() - ) - (input_layernorm): LlamaRMSNorm((720,), eps=1e-05) - (post_attention_layernorm): LlamaRMSNorm((720,), eps=1e-05) - ) - (12): LlamaDecoderLayer( - (self_attn): LlamaAttention( - (q_proj): Linear(in_features=720, out_features=960, bias=False) - (k_proj): Linear(in_features=720, out_features=320, bias=False) - (v_proj): Linear(in_features=720, out_features=320, bias=False) - (o_proj): Linear(in_features=960, out_features=720, bias=False) - ) - (mlp): LlamaMLP( - (gate_proj): Linear(in_features=720, out_features=2048, bias=False) - (up_proj): Linear(in_features=720, out_features=2048, bias=False) - (down_proj): Linear(in_features=2048, out_features=720, bias=False) - (act_fn): SiLU() - ) - (input_layernorm): LlamaRMSNorm((720,), eps=1e-05) - (post_attention_layernorm): LlamaRMSNorm((720,), eps=1e-05) - ) - (13): LlamaDecoderLayer( - (self_attn): LlamaAttention( - (q_proj): Linear(in_features=720, out_features=960, bias=False) - (k_proj): Linear(in_features=320, out_features=320, bias=False) - (v_proj): Linear(in_features=320, out_features=320, bias=False) - (o_proj): Linear(in_features=960, out_features=720, bias=False) - ) - (mlp): LlamaMLP( - (gate_proj): Linear(in_features=720, out_features=2048, bias=False) - (up_proj): Linear(in_features=720, out_features=2048, bias=False) - (down_proj): Linear(in_features=2048, out_features=720, bias=False) - (act_fn): SiLU() - ) - (input_layernorm): LlamaRMSNorm((720,), eps=1e-05) - (post_attention_layernorm): LlamaRMSNorm((720,), eps=1e-05) - ) - (14): LlamaDecoderLayer( - (self_attn): LlamaAttention( - (q_proj): Linear(in_features=720, out_features=960, bias=False) - (k_proj): Linear(in_features=720, out_features=320, bias=False) - (v_proj): Linear(in_features=720, out_features=320, bias=False) - (o_proj): Linear(in_features=960, out_features=720, bias=False) - ) - (mlp): LlamaMLP( - (gate_proj): Linear(in_features=720, out_features=2048, bias=False) - (up_proj): Linear(in_features=720, out_features=2048, bias=False) - (down_proj): Linear(in_features=2048, out_features=720, bias=False) - (act_fn): SiLU() - ) - (input_layernorm): LlamaRMSNorm((720,), eps=1e-05) - (post_attention_layernorm): LlamaRMSNorm((720,), eps=1e-05) - ) - (15): LlamaDecoderLayer( - (self_attn): LlamaAttention( - (q_proj): Linear(in_features=720, out_features=960, bias=False) - (k_proj): Linear(in_features=320, out_features=320, bias=False) - (v_proj): Linear(in_features=320, out_features=320, bias=False) - (o_proj): Linear(in_features=960, out_features=720, bias=False) - ) - (mlp): LlamaMLP( - (gate_proj): Linear(in_features=720, out_features=2048, bias=False) - (up_proj): Linear(in_features=720, out_features=2048, bias=False) - (down_proj): Linear(in_features=2048, out_features=720, bias=False) - (act_fn): SiLU() - ) - (input_layernorm): LlamaRMSNorm((720,), eps=1e-05) - (post_attention_layernorm): LlamaRMSNorm((720,), eps=1e-05) - ) - ) - (norm): LlamaRMSNorm((720,), eps=1e-05) - (rotary_emb): LlamaRotaryEmbedding() - ) - ) - (state_proj): Linear(in_features=32, out_features=960, bias=True) - (action_in_proj): Linear(in_features=32, out_features=720, bias=True) - (action_out_proj): Linear(in_features=720, out_features=32, bias=True) - (action_time_mlp_in): Linear(in_features=1440, out_features=720, bias=True) - (action_time_mlp_out): Linear(in_features=720, out_features=720, bias=True) - ) - ) -) -(Pdb) policy.config -*** AttributeError: 'DistributedDataParallel' object has no attribute 'config' -(Pdb) policy.input_features -*** AttributeError: 'DistributedDataParallel' object has no attribute 'input_features' -(Pdb) quit() -[rank1]: Traceback (most recent call last): -[rank1]: File "/home/jade_choghari/lerobot/src/lerobot/scripts/train_accelerate.py", line 368, in -[rank1]: train() -[rank1]: File "/home/jade_choghari/lerobot/src/lerobot/configs/parser.py", line 225, in wrapper_inner -[rank1]: response = fn(cfg, *args, **kwargs) -[rank1]: File "/home/jade_choghari/lerobot/src/lerobot/scripts/train_accelerate.py", line 263, in train -[rank1]: train_tracker, output_dict = update_policy( -[rank1]: File "/home/jade_choghari/lerobot/src/lerobot/scripts/train_accelerate.py", line 263, in train -[rank1]: train_tracker, output_dict = update_policy( -[rank1]: File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/bdb.py", line 90, in trace_dispatch -[rank1]: return self.dispatch_line(frame) -[rank1]: File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/bdb.py", line 115, in dispatch_line -[rank1]: if self.quitting: raise BdbQuit -[rank1]: bdb.BdbQuit -W0908 13:25:34.274000 776579 site-packages/torch/distributed/elastic/multiprocessing/api.py:900] Sending process 776 -663 closing signal SIGTERM -E0908 13:25:34.589000 776579 site-packages/torch/distributed/elastic/multiprocessing/api.py:874] failed (exitcode: 1 -) local_rank: 1 (pid: 776664) of binary: /home/jade_choghari/miniconda3/envs/lerobot/bin/python -Traceback (most recent call last): - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/runpy.py", line 196, in _run_module_as_main - return _run_code(code, main_globals, None, - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/runpy.py", line 86, in _run_code - exec(code, run_globals) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/accelerate/commands/launch.py", lin -e 1245, in - main() - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/accelerate/commands/launch.py", lin -e 1241, in main - launch_command(args) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/accelerate/commands/launch.py", lin -e 1226, in launch_command - multi_gpu_launcher(args) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/accelerate/commands/launch.py", lin -e 853, in multi_gpu_launcher - distrib_run.run(args) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/torch/distributed/run.py", line 883 -, in run - elastic_launch( - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/torch/distributed/launcher/api.py", - line 139, in __call__ - return launch_agent(self._config, self._entrypoint, list(args)) - File "/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/torch/distributed/launcher/api.py", - line 270, in launch_agent - raise ChildFailedError( -torch.distributed.elastic.multiprocessing.errors.ChildFailedError: -============================================================ -src/lerobot/scripts/train_accelerate.py FAILED ------------------------------------------------------------- -Failures: - ------------------------------------------------------------- -Root Cause (first observed failure): -[0]: - time : 2025-09-08_13:25:34 - host : hf-dgx-01 - rank : 1 (local_rank: 1) - exitcode : 1 (pid: 776664) - error_file: - traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html -============================================================ -(lerobot) jade_choghari@hf-dgx-01:~/lerobot$ clear -(lerobot) jade_choghari@hf-dgx-01:~/lerobot$ clear -(lerobot) jade_choghari@hf-dgx-01:~/lerobot$ bash examples/7_train_acc.sh -Training dir: /raid/jade/logs/lerobot/lerobot_2_HuggingFaceVLA_libero_smolvla_lr1e-4bs32steps100000 -The following values were not passed to `accelerate launch` and had defaults used instead: - More than one GPU was found, enabling multi-GPU training. - If this was unintended please pass in `--num_processes=1`. - `--dynamo_backend` was set to a value of `'no'` -To avoid this warning pass in values for each of the problematic parameters or run `accelerate config`. -/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/accelerate/utils/launch.py:238: UserWarning -: Port `29522` is already in use. Accelerate will attempt to launch in a standalone-like mode by finding an open por -t automatically for this session. If this current attempt fails, or for more control in future runs, please specify -a different port (e.g., `--main_process_port `) or use `--main_process_port 0` for automatic selec -tion in your launch command or Accelerate config file. - warnings.warn( -INFO 2025-09-08 13:33:47 ils/utils.py:48 Cuda backend detected, using cuda. -WARNING 2025-09-08 13:33:47 /policies.py:81 Device 'None' is not available. Switching to 'cuda'. -INFO 2025-09-08 13:33:47 ccelerate.py:99 {'batch_size': 32, - 'dataset': {'episodes': None, - 'image_transforms': {'enable': False, - 'max_num_transforms': 3, - 'random_order': False, - 'tfs': {'brightness': {'kwargs': {'brightness': [0.8, - 1.2]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'contrast': {'kwargs': {'contrast': [0.8, - 1.2]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'hue': {'kwargs': {'hue': [-0.05, - 0.05]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'saturation': {'kwargs': {'saturation': [0.5, - 1.5]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'sharpness': {'kwargs': {'sharpness': [0.5, - 1.5]}, - 'type': 'SharpnessJitter', - 'weight': 1.0}}}, - 'repo_id': 'HuggingFaceVLA/libero', - 'revision': None, - 'root': '/raid/jade/.cache/huggingface/datasets', - 'use_imagenet_stats': True, - 'video_backend': 'torchcodec'}, - 'env': {'camera_name': 'agentview_image,robot0_eye_in_hand_image', - 'episode_length': 520, - 'features': {'action': {'shape': [7], - 'type': }, - 'agent_pos': {'shape': [8], - 'type': }, - 'pixels/agentview_image': {'shape': [360, 360, 3], - 'type': }, - 'pixels/robot0_eye_in_hand_image': {'shape': [360, - 360, - 3], - 'type': }}, - 'features_map': {'action': 'action', - 'agent_pos': 'observation.state', - 'pixels/agentview_image': 'observation.images.image', - 'pixels/robot0_eye_in_hand_image': 'observation.images.image2'}, - 'fps': 30, - 'init_states': True, - 'max_parallel_tasks': 5, - 'multitask_eval': True, - 'obs_type': 'pixels_agent_pos', - 'render_mode': 'rgb_array', - 'task': 'libero_spatial', - 'type': 'libero'}, - 'eval': {'batch_size': 1, 'n_episodes': 1, 'use_async_envs': False}, - 'eval_freq': 0, - 'job_name': 'libero_smolvla', - 'log_freq': 200, - 'num_workers': 4, - 'optimizer': {'betas': [0.9, 0.95], - 'eps': 1e-08, - 'grad_clip_norm': 10, - 'lr': 0.0001, - 'type': 'adamw', - 'weight_decay': 1e-10}, - 'output_dir': '/raid/jade/logs/lerobot/lerobot_2_HuggingFaceVLA_libero_smolvla_lr1e-4bs32steps100000', - 'policy': {'adapt_to_pi_aloha': False, - 'add_image_special_tokens': False, - 'attention_mode': 'cross_attn', - 'chunk_size': 50, - 'device': 'cuda', - 'empty_cameras': 0, - 'expert_width_multiplier': 0.75, - 'freeze_vision_encoder': True, - 'gradient_accumulation_steps': 1, - 'input_features': {}, - 'license': None, - 'load_vlm_weights': False, - 'max_action_dim': 32, - 'max_period': 4.0, - 'max_state_dim': 32, - 'min_period': 0.004, - 'n_action_steps': 1, - 'n_obs_steps': 1, - 'normalization_mapping': {'ACTION': , - 'STATE': , - 'VISUAL': }, - 'num_expert_layers': -1, - 'num_steps': 10, - 'num_vlm_layers': 16, - 'optimizer_betas': [0.9, 0.95], - 'optimizer_eps': 1e-08, - 'optimizer_grad_clip_norm': 10, - 'optimizer_lr': 0.0001, - 'optimizer_weight_decay': 1e-10, - 'output_features': {}, - 'pad_language_to': 'longest', - 'prefix_length': -1, - 'private': None, - 'push_to_hub': True, - 'repo_id': 'None', - 'resize_imgs_with_padding': [512, 512], - 'scheduler_decay_lr': 2.5e-06, - 'scheduler_decay_steps': 30000, - 'scheduler_warmup_steps': 1000, - 'self_attn_every_n_layers': 2, - 'tags': None, - 'tokenizer_max_length': 48, - 'train_expert_only': True, - 'train_state_proj': True, - 'type': 'smolvla', - 'use_amp': True, - 'use_cache': True, - 'use_delta_joint_actions_aloha': False, - 'vlm_model_name': 'HuggingFaceTB/SmolVLM2-500M-Instruct'}, - 'resume': False, - 'save_checkpoint': True, - 'save_freq': 20000, - 'scheduler': {'decay_lr': 2.5e-06, - 'num_decay_steps': 30000, - 'num_warmup_steps': 1000, - 'peak_lr': 0.0001, - 'type': 'cosine_decay_with_warmup'}, - 'seed': 1000, - 'steps': 100000, - 'use_policy_training_preset': True, - 'wandb': {'disable_artifact': False, - 'enable': False, - 'entity': None, - 'mode': None, - 'notes': None, - 'project': 'lerobot', - 'run_id': None}} -INFO 2025-09-08 13:33:47 ils/utils.py:48 Cuda backend detected, using cuda. -WARNING 2025-09-08 13:33:47 /policies.py:81 Device 'None' is not available. Switching to 'cuda'. -INFO 2025-09-08 13:33:47 ccelerate.py:99 {'batch_size': 32, - 'dataset': {'episodes': None, - 'image_transforms': {'enable': False, - 'max_num_transforms': 3, - 'random_order': False, - 'tfs': {'brightness': {'kwargs': {'brightness': [0.8, - 1.2]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'contrast': {'kwargs': {'contrast': [0.8, - 1.2]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'hue': {'kwargs': {'hue': [-0.05, - 0.05]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'saturation': {'kwargs': {'saturation': [0.5, - 1.5]}, - 'type': 'ColorJitter', - 'weight': 1.0}, - 'sharpness': {'kwargs': {'sharpness': [0.5, - 1.5]}, - 'type': 'SharpnessJitter', - 'weight': 1.0}}}, - 'repo_id': 'HuggingFaceVLA/libero', - 'revision': None, - 'root': '/raid/jade/.cache/huggingface/datasets', - 'use_imagenet_stats': True, - 'video_backend': 'torchcodec'}, - 'env': {'camera_name': 'agentview_image,robot0_eye_in_hand_image', - 'episode_length': 520, - 'features': {'action': {'shape': [7], - 'type': }, - 'agent_pos': {'shape': [8], - 'type': }, - 'pixels/agentview_image': {'shape': [360, 360, 3], - 'type': }, - 'pixels/robot0_eye_in_hand_image': {'shape': [360, - 360, - 3], - 'type': }}, - 'features_map': {'action': 'action', - 'agent_pos': 'observation.state', - 'pixels/agentview_image': 'observation.images.image', - 'pixels/robot0_eye_in_hand_image': 'observation.images.image2'}, - 'fps': 30, - 'init_states': True, - 'max_parallel_tasks': 5, - 'multitask_eval': True, - 'obs_type': 'pixels_agent_pos', - 'render_mode': 'rgb_array', - 'task': 'libero_spatial', - 'type': 'libero'}, - 'eval': {'batch_size': 1, 'n_episodes': 1, 'use_async_envs': False}, - 'eval_freq': 0, - 'job_name': 'libero_smolvla', - 'log_freq': 200, - 'num_workers': 4, - 'optimizer': {'betas': [0.9, 0.95], - 'eps': 1e-08, - 'grad_clip_norm': 10, - 'lr': 0.0001, - 'type': 'adamw', - 'weight_decay': 1e-10}, - 'output_dir': '/raid/jade/logs/lerobot/lerobot_2_HuggingFaceVLA_libero_smolvla_lr1e-4bs32steps100000', - 'policy': {'adapt_to_pi_aloha': False, - 'add_image_special_tokens': False, - 'attention_mode': 'cross_attn', - 'chunk_size': 50, - 'device': 'cuda', - 'empty_cameras': 0, - 'expert_width_multiplier': 0.75, - 'freeze_vision_encoder': True, - 'gradient_accumulation_steps': 1, - 'input_features': {}, - 'license': None, - 'load_vlm_weights': False, - 'max_action_dim': 32, - 'max_period': 4.0, - 'max_state_dim': 32, - 'min_period': 0.004, - 'n_action_steps': 1, - 'n_obs_steps': 1, - 'normalization_mapping': {'ACTION': , - 'STATE': , - 'VISUAL': }, - 'num_expert_layers': -1, - 'num_steps': 10, - 'num_vlm_layers': 16, - 'optimizer_betas': [0.9, 0.95], - 'optimizer_eps': 1e-08, - 'optimizer_grad_clip_norm': 10, - 'optimizer_lr': 0.0001, - 'optimizer_weight_decay': 1e-10, - 'output_features': {}, - 'pad_language_to': 'longest', - 'prefix_length': -1, - 'private': None, - 'push_to_hub': True, - 'repo_id': 'None', - 'resize_imgs_with_padding': [512, 512], - 'scheduler_decay_lr': 2.5e-06, - 'scheduler_decay_steps': 30000, - 'scheduler_warmup_steps': 1000, - 'self_attn_every_n_layers': 2, - 'tags': None, - 'tokenizer_max_length': 48, - 'train_expert_only': True, - 'train_state_proj': True, - 'type': 'smolvla', - 'use_amp': True, - 'use_cache': True, - 'use_delta_joint_actions_aloha': False, - 'vlm_model_name': 'HuggingFaceTB/SmolVLM2-500M-Instruct'}, - 'resume': False, - 'save_checkpoint': True, - 'save_freq': 20000, - 'scheduler': {'decay_lr': 2.5e-06, - 'num_decay_steps': 30000, - 'num_warmup_steps': 1000, - 'peak_lr': 0.0001, - 'type': 'cosine_decay_with_warmup'}, - 'seed': 1000, - 'steps': 100000, - 'use_policy_training_preset': True, - 'wandb': {'disable_artifact': False, - 'enable': False, - 'entity': None, - 'mode': None, - 'notes': None, - 'project': 'lerobot', - 'run_id': None}} -WARNING 2025-09-08 13:33:47 ls/other.py:512 Detected kernel version 5.4.0, which is below the recommended minimum of - 5.5.0; this can cause the process to hang. It is recommended to upgrade the kernel to the minimum version or higher -. -WARNING 2025-09-08 13:33:47 ls/other.py:512 Detected kernel version 5.4.0, which is below the recommended minimum of - 5.5.0; this can cause the process to hang. It is recommended to upgrade the kernel to the minimum version or higher -. -INFO 2025-09-08 13:33:47 celerate.py:149 Creating dataset -Resolving data files: 100%|████████████████████████████████| 1693/1693 [00:00<00:00, 103295.66it/s] -Loading dataset shards: 100%|████████████████████████████████████| 69/69 [00:00<00:00, 5229.81it/s] -Resolving data files: 100%|████████████████████████████████| 1693/1693 [00:00<00:00, 360601.09it/s] -Loading dataset shards: 100%|████████████████████████████████████| 69/69 [00:00<00:00, 4881.54it/s] -c -INFO 2025-09-08 13:33:53 celerate.py:160 Creating policy -/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/transformers/utils/hub.py:111: FutureWarnin -g: Using `TRANSFORMERS_CACHE` is deprecated and will be removed in v5 of Transformers. Use `HF_HOME` instead. - warnings.warn( -c -/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/torch/distributed/distributed_c10d.py:4631: - UserWarning: No device id is provided via `init_process_group` or `barrier `. Using the current device set by the u -ser. - warnings.warn( # warn only once -[rank1]:[W908 13:33:54.613597516 ProcessGroupNCCL.cpp:4718] [PG ID 0 PG GUID 0 Rank 1] using GPU 1 as device used b -y this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You - can pecify device_id in init_process_group() to force use of a particular device. -Reducing the number of VLM layers to 16 ... -/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/torch/distributed/distributed_c10d.py:4631: - UserWarning: No device id is provided via `init_process_group` or `barrier `. Using the current device set by the u -ser. - warnings.warn( # warn only once -[rank0]:[W908 13:34:15.806448425 ProcessGroupNCCL.cpp:4718] [PG ID 0 PG GUID 0 Rank 0] using GPU 0 as device used b -y this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You - can pecify device_id in init_process_group() to force use of a particular device. -INFO 2025-09-08 13:34:15 celerate.py:171 Creating optimizer and scheduler -/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/transformers/utils/hub.py:111: FutureWarnin -g: Using `TRANSFORMERS_CACHE` is deprecated and will be removed in v5 of Transformers. Use `HF_HOME` instead. - warnings.warn( -Reducing the number of VLM layers to 16 ... -INFO 2025-09-08 13:34:36 celerate.py:211 Output dir: /raid/jade/logs/lerobot/lerobot_2_HuggingFaceVLA_libero_smolvla -_lr1e-4bs32steps100000 -INFO 2025-09-08 13:34:36 celerate.py:213 cfg.env.task='libero_spatial' -INFO 2025-09-08 13:34:36 celerate.py:214 cfg.steps=100000 (100K) -INFO 2025-09-08 13:34:36 celerate.py:215 dataset.num_frames=273465 (273K) -INFO 2025-09-08 13:34:36 celerate.py:216 dataset.num_episodes=1693 -INFO 2025-09-08 13:34:36 celerate.py:217 num_learnable_params=99880992 (100M) -INFO 2025-09-08 13:34:36 celerate.py:218 num_total_params=450046220 (450M) -INFO 2025-09-08 13:34:36 celerate.py:219 Number of processes: 2 -INFO 2025-09-08 13:34:36 celerate.py:220 Device: cuda:0 -INFO 2025-09-08 13:34:36 celerate.py:221 Mixed precision: bf16 -INFO 2025-09-08 13:34:36 celerate.py:243 Start offline training on a fixed dataset -[rank1]:[W908 13:34:39.454560620 reducer.cpp:1430] Warning: find_unused_parameters=True was specified in DDP constru -ctor, but did not find any unused parameters in the forward pass. This flag results in an extra traversal of the aut -ograd graph every iteration, which can adversely affect performance. If your model indeed never has any unused para -meters in the forward pass, consider turning this flag off. Note that this warning may be a false positive if your m -odel has flow control causing later iterations to have unused parameters. (function operator()) -[rank0]:[W908 13:34:40.502702504 reducer.cpp:1430] Warning: find_unused_parameters=True was specified in DDP constru -ctor, but did not find any unused parameters in the forward pass. This flag results in an extra traversal of the aut -ograd graph every iteration, which can adversely affect performance. If your model indeed never has any unused para -meters in the forward pass, consider turning this flag off. Note that this warning may be a false positive if your m -odel has flow control causing later iterations to have unused parameters. (function operator()) -INFO 2025-09-08 13:36:23 celerate.py:281 step:200 smpl:13K ep:79 epch:0.05 loss:0.963 grdn:2.699 lr:2.0e-05 updt_s:0 -.506 data_s:0.027 -INFO 2025-09-08 13:38:09 celerate.py:281 step:400 smpl:26K ep:158 epch:0.09 loss:0.389 grdn:3.127 lr:6.0e-05 updt_s: -0.525 data_s:0.003 -INFO 2025-09-08 13:39:53 celerate.py:281 step:600 smpl:38K ep:238 epch:0.14 loss:0.261 grdn:2.618 lr:9.5e-05 updt_s: -0.517 data_s:0.003 -INFO 2025-09-08 13:41:37 celerate.py:281 step:800 smpl:51K ep:317 epch:0.19 loss:0.231 grdn:1.684 lr:9.9e-05 updt_s: -0.516 data_s:0.003 -INFO 2025-09-08 13:43:21 celerate.py:281 step:1K smpl:64K ep:396 epch:0.23 loss:0.211 grdn:1.258 lr:9.9e-05 updt_s:0 -.514 data_s:0.003 -INFO 2025-09-08 13:45:05 celerate.py:281 step:1K smpl:77K ep:475 epch:0.28 loss:0.198 grdn:1.032 lr:9.9e-05 updt_s:0 -.517 data_s:0.003 -INFO 2025-09-08 13:46:49 celerate.py:281 step:1K smpl:90K ep:555 epch:0.33 loss:0.182 grdn:0.880 lr:9.8e-05 updt_s:0 -.515 data_s:0.003 -INFO 2025-09-08 13:48:33 celerate.py:281 step:2K smpl:102K ep:634 epch:0.37 loss:0.167 grdn:0.744 lr:9.8e-05 updt_s: -0.514 data_s:0.003 -INFO 2025-09-08 13:50:17 celerate.py:281 step:2K smpl:115K ep:713 epch:0.42 loss:0.157 grdn:0.680 lr:9.7e-05 updt_s: -0.514 data_s:0.003 -INFO 2025-09-08 13:52:01 celerate.py:281 step:2K smpl:128K ep:792 epch:0.47 loss:0.147 grdn:0.612 lr:9.6e-05 updt_s: -0.517 data_s:0.003 -INFO 2025-09-08 13:53:44 celerate.py:281 step:2K smpl:141K ep:872 epch:0.51 loss:0.142 grdn:0.576 lr:9.5e-05 updt_s: -0.510 data_s:0.003 -INFO 2025-09-08 13:55:27 celerate.py:281 step:2K smpl:154K ep:951 epch:0.56 loss:0.136 grdn:0.523 lr:9.4e-05 updt_s: -0.514 data_s:0.003 -INFO 2025-09-08 13:57:11 celerate.py:281 step:3K smpl:166K ep:1K epch:0.61 loss:0.132 grdn:0.509 lr:9.3e-05 updt_s:0 -.516 data_s:0.003 -INFO 2025-09-08 13:58:57 celerate.py:281 step:3K smpl:179K ep:1K epch:0.66 loss:0.126 grdn:0.492 lr:9.2e-05 updt_s:0 -.525 data_s:0.003 -INFO 2025-09-08 14:00:43 celerate.py:281 step:3K smpl:192K ep:1K epch:0.70 loss:0.124 grdn:0.467 lr:9.1e-05 updt_s:0 -.525 data_s:0.003 -INFO 2025-09-08 14:02:26 celerate.py:281 step:3K smpl:205K ep:1K epch:0.75 loss:0.119 grdn:0.438 lr:9.0e-05 updt_s:0 -.508 data_s:0.003 -INFO 2025-09-08 14:04:27 celerate.py:281 step:3K smpl:218K ep:1K epch:0.80 loss:0.118 grdn:0.426 lr:8.9e-05 updt_s:0 -.564 data_s:0.039 -INFO 2025-09-08 14:06:10 celerate.py:281 step:4K smpl:230K ep:1K epch:0.84 loss:0.116 grdn:0.422 lr:8.7e-05 updt_s:0 -.511 data_s:0.004 -INFO 2025-09-08 14:07:55 celerate.py:281 step:4K smpl:243K ep:2K epch:0.89 loss:0.113 grdn:0.395 lr:8.6e-05 updt_s:0 -.517 data_s:0.003 -INFO 2025-09-08 14:09:38 celerate.py:281 step:4K smpl:256K ep:2K epch:0.94 loss:0.111 grdn:0.401 lr:8.5e-05 updt_s:0 -.511 data_s:0.003 -INFO 2025-09-08 14:11:21 celerate.py:281 step:4K smpl:269K ep:2K epch:0.98 loss:0.110 grdn:0.380 lr:8.3e-05 updt_s:0 -.511 data_s:0.003 -INFO 2025-09-08 14:13:08 celerate.py:281 step:4K smpl:282K ep:2K epch:1.03 loss:0.109 grdn:0.381 lr:8.2e-05 updt_s:0 -.413 data_s:0.119 -INFO 2025-09-08 14:14:52 celerate.py:281 step:5K smpl:294K ep:2K epch:1.08 loss:0.107 grdn:0.387 lr:8.0e-05 updt_s:0 -.373 data_s:0.146 -INFO 2025-09-08 14:16:36 celerate.py:281 step:5K smpl:307K ep:2K epch:1.12 loss:0.107 grdn:0.366 lr:7.8e-05 updt_s:0 -.446 data_s:0.072 -INFO 2025-09-08 14:18:19 celerate.py:281 step:5K smpl:320K ep:2K epch:1.17 loss:0.105 grdn:0.347 lr:7.6e-05 updt_s:0 -.468 data_s:0.045 -INFO 2025-09-08 14:20:01 celerate.py:281 step:5K smpl:333K ep:2K epch:1.22 loss:0.103 grdn:0.350 lr:7.5e-05 updt_s:0 -.510 data_s:0.003 -INFO 2025-09-08 14:21:46 celerate.py:281 step:5K smpl:346K ep:2K epch:1.26 loss:0.101 grdn:0.336 lr:7.3e-05 updt_s:0 -.512 data_s:0.011 -INFO 2025-09-08 14:23:30 celerate.py:281 step:6K smpl:358K ep:2K epch:1.31 loss:0.102 grdn:0.345 lr:7.1e-05 updt_s:0 -.515 data_s:0.003 -INFO 2025-09-08 14:25:15 celerate.py:281 step:6K smpl:371K ep:2K epch:1.36 loss:0.100 grdn:0.333 lr:6.9e-05 updt_s:0 -.521 data_s:0.003 -INFO 2025-09-08 14:26:59 celerate.py:281 step:6K smpl:384K ep:2K epch:1.40 loss:0.100 grdn:0.328 lr:6.7e-05 updt_s:0 -.516 data_s:0.003 -INFO 2025-09-08 14:28:43 celerate.py:281 step:6K smpl:397K ep:2K epch:1.45 loss:0.099 grdn:0.319 lr:6.5e-05 updt_s:0 -.512 data_s:0.003 -INFO 2025-09-08 14:30:26 celerate.py:281 step:6K smpl:410K ep:3K epch:1.50 loss:0.098 grdn:0.313 lr:6.3e-05 updt_s:0 -.515 data_s:0.003 -INFO 2025-09-08 14:32:11 celerate.py:281 step:7K smpl:422K ep:3K epch:1.54 loss:0.097 grdn:0.319 lr:6.1e-05 updt_s:0 -.519 data_s:0.004 -INFO 2025-09-08 14:33:55 celerate.py:281 step:7K smpl:435K ep:3K epch:1.59 loss:0.097 grdn:0.312 lr:5.9e-05 updt_s:0 -.506 data_s:0.010 -INFO 2025-09-08 14:35:39 celerate.py:281 step:7K smpl:448K ep:3K epch:1.64 loss:0.097 grdn:0.307 lr:5.7e-05 updt_s:0 -.516 data_s:0.003 -INFO 2025-09-08 14:37:23 celerate.py:281 step:7K smpl:461K ep:3K epch:1.69 loss:0.095 grdn:0.294 lr:5.5e-05 updt_s:0 -.518 data_s:0.003 -INFO 2025-09-08 14:39:07 celerate.py:281 step:7K smpl:474K ep:3K epch:1.73 loss:0.095 grdn:0.299 lr:5.3e-05 updt_s:0 -.507 data_s:0.007 -INFO 2025-09-08 14:40:52 celerate.py:281 step:8K smpl:486K ep:3K epch:1.78 loss:0.094 grdn:0.283 lr:5.1e-05 updt_s:0 -.523 data_s:0.003 -INFO 2025-09-08 14:42:36 celerate.py:281 step:8K smpl:499K ep:3K epch:1.83 loss:0.093 grdn:0.284 lr:4.9e-05 updt_s:0 -.517 data_s:0.003 -INFO 2025-09-08 14:44:22 celerate.py:281 step:8K smpl:512K ep:3K epch:1.87 loss:0.092 grdn:0.284 lr:4.7e-05 updt_s:0 -.465 data_s:0.060 -INFO 2025-09-08 14:46:06 celerate.py:281 step:8K smpl:525K ep:3K epch:1.92 loss:0.093 grdn:0.292 lr:4.5e-05 updt_s:0 -.456 data_s:0.066 -INFO 2025-09-08 14:47:49 celerate.py:281 step:8K smpl:538K ep:3K epch:1.97 loss:0.093 grdn:0.290 lr:4.3e-05 updt_s:0 -.510 data_s:0.003 -INFO 2025-09-08 14:49:37 celerate.py:281 step:9K smpl:550K ep:3K epch:2.01 loss:0.092 grdn:0.283 lr:4.1e-05 updt_s:0 -.419 data_s:0.117 -INFO 2025-09-08 14:51:20 celerate.py:281 step:9K smpl:563K ep:3K epch:2.06 loss:0.092 grdn:0.275 lr:3.9e-05 updt_s:0 -.463 data_s:0.053 -INFO 2025-09-08 14:53:05 celerate.py:281 step:9K smpl:576K ep:4K epch:2.11 loss:0.090 grdn:0.272 lr:3.7e-05 updt_s:0 -.517 data_s:0.003 -INFO 2025-09-08 14:54:49 celerate.py:281 step:9K smpl:589K ep:4K epch:2.15 loss:0.090 grdn:0.268 lr:3.5e-05 updt_s:0 -.506 data_s:0.013 -INFO 2025-09-08 14:56:32 celerate.py:281 step:9K smpl:602K ep:4K epch:2.20 loss:0.090 grdn:0.271 lr:3.3e-05 updt_s:0 -.513 data_s:0.003 -INFO 2025-09-08 14:58:17 celerate.py:281 step:10K smpl:614K ep:4K epch:2.25 loss:0.090 grdn:0.268 lr:3.1e-05 updt_s: -0.520 data_s:0.003 -INFO 2025-09-08 15:00:02 celerate.py:281 step:10K smpl:627K ep:4K epch:2.29 loss:0.089 grdn:0.261 lr:3.0e-05 updt_s: -0.519 data_s:0.003 -INFO 2025-09-08 15:01:48 celerate.py:281 step:10K smpl:640K ep:4K epch:2.34 loss:0.090 grdn:0.271 lr:2.8e-05 updt_s: -0.526 data_s:0.003 -INFO 2025-09-08 15:03:33 celerate.py:281 step:10K smpl:653K ep:4K epch:2.39 loss:0.089 grdn:0.262 lr:2.6e-05 updt_s: -0.521 data_s:0.003 -INFO 2025-09-08 15:05:18 celerate.py:281 step:10K smpl:666K ep:4K epch:2.43 loss:0.090 grdn:0.264 lr:2.4e-05 updt_s: -0.519 data_s:0.003 -INFO 2025-09-08 15:07:32 celerate.py:281 step:11K smpl:678K ep:4K epch:2.48 loss:0.089 grdn:0.255 lr:2.3e-05 updt_s: -0.663 data_s:0.004 -INFO 2025-09-08 15:09:21 celerate.py:281 step:11K smpl:691K ep:4K epch:2.53 loss:0.090 grdn:0.263 lr:2.1e-05 updt_s: -0.514 data_s:0.030 -INFO 2025-09-08 15:11:06 celerate.py:281 step:11K smpl:704K ep:4K epch:2.57 loss:0.088 grdn:0.254 lr:1.9e-05 updt_s: -0.517 data_s:0.006 -INFO 2025-09-08 15:12:51 celerate.py:281 step:11K smpl:717K ep:4K epch:2.62 loss:0.088 grdn:0.252 lr:1.8e-05 updt_s: -0.517 data_s:0.005 -INFO 2025-09-08 15:14:38 celerate.py:281 step:11K smpl:730K ep:5K epch:2.67 loss:0.088 grdn:0.251 lr:1.6e-05 updt_s: -0.532 data_s:0.003 -INFO 2025-09-08 15:16:23 celerate.py:281 step:12K smpl:742K ep:5K epch:2.71 loss:0.088 grdn:0.253 lr:1.5e-05 updt_s: -0.520 data_s:0.003 -INFO 2025-09-08 15:18:08 celerate.py:281 step:12K smpl:755K ep:5K epch:2.76 loss:0.087 grdn:0.244 lr:1.4e-05 updt_s: -0.521 data_s:0.003 -INFO 2025-09-08 15:19:54 celerate.py:281 step:12K smpl:768K ep:5K epch:2.81 loss:0.088 grdn:0.247 lr:1.2e-05 updt_s: -0.524 data_s:0.003 -INFO 2025-09-08 15:21:39 celerate.py:281 step:12K smpl:781K ep:5K epch:2.86 loss:0.087 grdn:0.242 lr:1.1e-05 updt_s: -0.520 data_s:0.003 -INFO 2025-09-08 15:23:32 celerate.py:281 step:12K smpl:794K ep:5K epch:2.90 loss:0.088 grdn:0.243 lr:1.0e-05 updt_s: -0.560 data_s:0.003 -INFO 2025-09-08 15:25:48 celerate.py:281 step:13K smpl:806K ep:5K epch:2.95 loss:0.087 grdn:0.240 lr:9.0e-06 updt_s: -0.674 data_s:0.005 -INFO 2025-09-08 15:28:02 celerate.py:281 step:13K smpl:819K ep:5K epch:3.00 loss:0.088 grdn:0.245 lr:8.0e-06 updt_s: -0.662 data_s:0.004 -INFO 2025-09-08 15:31:06 celerate.py:281 step:13K smpl:832K ep:5K epch:3.04 loss:0.086 grdn:0.236 lr:7.1e-06 updt_s: -0.688 data_s:0.231 -INFO 2025-09-08 15:32:52 celerate.py:281 step:13K smpl:845K ep:5K epch:3.09 loss:0.087 grdn:0.231 lr:6.3e-06 updt_s: -0.521 data_s:0.003 -INFO 2025-09-08 15:35:46 celerate.py:281 step:13K smpl:858K ep:5K epch:3.14 loss:0.088 grdn:0.235 lr:5.6e-06 updt_s: -0.637 data_s:0.232 -INFO 2025-09-08 15:37:34 celerate.py:281 step:14K smpl:870K ep:5K epch:3.18 loss:0.087 grdn:0.238 lr:4.9e-06 updt_s: -0.514 data_s:0.025 -INFO 2025-09-08 15:39:18 celerate.py:281 step:14K smpl:883K ep:5K epch:3.23 loss:0.087 grdn:0.226 lr:4.3e-06 updt_s: -0.515 data_s:0.003 -INFO 2025-09-08 15:41:02 celerate.py:281 step:14K smpl:896K ep:6K epch:3.28 loss:0.087 grdn:0.230 lr:3.8e-06 updt_s: -0.515 data_s:0.003 -INFO 2025-09-08 15:42:45 celerate.py:281 step:14K smpl:909K ep:6K epch:3.32 loss:0.086 grdn:0.229 lr:3.4e-06 updt_s: -0.507 data_s:0.008 -INFO 2025-09-08 15:44:29 celerate.py:281 step:14K smpl:922K ep:6K epch:3.37 loss:0.087 grdn:0.229 lr:3.0e-06 updt_s: -0.515 data_s:0.003 -INFO 2025-09-08 15:46:12 celerate.py:281 step:15K smpl:934K ep:6K epch:3.42 loss:0.087 grdn:0.228 lr:2.8e-06 updt_s: -0.502 data_s:0.011 -INFO 2025-09-08 15:47:56 celerate.py:281 step:15K smpl:947K ep:6K epch:3.46 loss:0.086 grdn:0.232 lr:2.6e-06 updt_s: -0.515 data_s:0.004 -INFO 2025-09-08 15:49:39 celerate.py:281 step:15K smpl:960K ep:6K epch:3.51 loss:0.087 grdn:0.234 lr:2.5e-06 updt_s: -0.365 data_s:0.147 -INFO 2025-09-08 15:51:22 celerate.py:281 step:15K smpl:973K ep:6K epch:3.56 loss:0.086 grdn:0.230 lr:2.5e-06 updt_s: -0.333 data_s:0.179 -INFO 2025-09-08 15:53:07 celerate.py:281 step:15K smpl:986K ep:6K epch:3.60 loss:0.087 grdn:0.229 lr:2.5e-06 updt_s: -0.357 data_s:0.164 -INFO 2025-09-08 15:54:50 celerate.py:281 step:16K smpl:998K ep:6K epch:3.65 loss:0.087 grdn:0.230 lr:2.5e-06 updt_s: -0.365 data_s:0.151 -INFO 2025-09-08 15:56:35 celerate.py:281 step:16K smpl:1M ep:6K epch:3.70 loss:0.086 grdn:0.228 lr:2.5e-06 updt_s:0. -450 data_s:0.071 -INFO 2025-09-08 15:58:19 celerate.py:281 step:16K smpl:1M ep:6K epch:3.74 loss:0.087 grdn:0.232 lr:2.5e-06 updt_s:0. -495 data_s:0.023 -INFO 2025-09-08 16:00:04 celerate.py:281 step:16K smpl:1M ep:6K epch:3.79 loss:0.087 grdn:0.227 lr:2.5e-06 updt_s:0. -393 data_s:0.131 -INFO 2025-09-08 16:01:48 celerate.py:281 step:16K smpl:1M ep:6K epch:3.84 loss:0.086 grdn:0.230 lr:2.5e-06 updt_s:0. -515 data_s:0.003 -INFO 2025-09-08 16:03:32 celerate.py:281 step:17K smpl:1M ep:7K epch:3.88 loss:0.087 grdn:0.231 lr:2.5e-06 updt_s:0. -518 data_s:0.003 -INFO 2025-09-08 16:05:17 celerate.py:281 step:17K smpl:1M ep:7K epch:3.93 loss:0.087 grdn:0.231 lr:2.5e-06 updt_s:0. -397 data_s:0.125 -INFO 2025-09-08 16:12:41 celerate.py:281 step:17K smpl:1M ep:7K epch:3.98 loss:0.086 grdn:0.230 lr:2.5e-06 updt_s:1. -496 data_s:0.719 -INFO 2025-09-08 16:14:28 celerate.py:281 step:17K smpl:1M ep:7K epch:4.03 loss:0.087 grdn:0.228 lr:2.5e-06 updt_s:0. -413 data_s:0.124 -INFO 2025-09-08 16:16:13 celerate.py:281 step:17K smpl:1M ep:7K epch:4.07 loss:0.087 grdn:0.234 lr:2.5e-06 updt_s:0. -469 data_s:0.050 -INFO 2025-09-08 16:17:57 celerate.py:281 step:18K smpl:1M ep:7K epch:4.12 loss:0.087 grdn:0.228 lr:2.5e-06 updt_s:0. -375 data_s:0.145 -INFO 2025-09-08 16:19:42 celerate.py:281 step:18K smpl:1M ep:7K epch:4.17 loss:0.086 grdn:0.230 lr:2.5e-06 updt_s:0. -333 data_s:0.189 -INFO 2025-09-08 16:21:26 celerate.py:281 step:18K smpl:1M ep:7K epch:4.21 loss:0.087 grdn:0.231 lr:2.5e-06 updt_s:0. -334 data_s:0.184 -INFO 2025-09-08 16:23:09 celerate.py:281 step:18K smpl:1M ep:7K epch:4.26 loss:0.086 grdn:0.235 lr:2.5e-06 updt_s:0. -331 data_s:0.185 -INFO 2025-09-08 16:24:53 celerate.py:281 step:18K smpl:1M ep:7K epch:4.31 loss:0.088 grdn:0.236 lr:2.5e-06 updt_s:0. -333 data_s:0.182 -INFO 2025-09-08 16:26:38 celerate.py:281 step:19K smpl:1M ep:7K epch:4.35 loss:0.086 grdn:0.230 lr:2.5e-06 updt_s:0. -337 data_s:0.188 -INFO 2025-09-08 16:28:22 celerate.py:281 step:19K smpl:1M ep:7K epch:4.40 loss:0.086 grdn:0.231 lr:2.5e-06 updt_s:0. -420 data_s:0.099 -INFO 2025-09-08 16:30:06 celerate.py:281 step:19K smpl:1M ep:8K epch:4.45 loss:0.086 grdn:0.232 lr:2.5e-06 updt_s:0. -444 data_s:0.075 -INFO 2025-09-08 16:31:49 celerate.py:281 step:19K smpl:1M ep:8K epch:4.49 loss:0.087 grdn:0.235 lr:2.5e-06 updt_s:0. -475 data_s:0.036 -INFO 2025-09-08 16:33:33 celerate.py:281 step:19K smpl:1M ep:8K epch:4.54 loss:0.087 grdn:0.236 lr:2.5e-06 updt_s:0. -379 data_s:0.139 -INFO 2025-09-08 16:35:17 celerate.py:281 step:20K smpl:1M ep:8K epch:4.59 loss:0.087 grdn:0.237 lr:2.5e-06 updt_s:0. -348 data_s:0.171 -INFO 2025-09-08 16:37:01 celerate.py:281 step:20K smpl:1M ep:8K epch:4.63 loss:0.087 grdn:0.231 lr:2.5e-06 updt_s:0. -332 data_s:0.185 -/home/jade_choghari/miniconda3/envs/lerobot/lib/python3.10/site-packages/torch/distributed/distributed_c10d.py:4631: - UserWarning: No device id is provided via `init_process_group` or `barrier `. Using the current device set by the u -ser. - warnings.warn( # warn only once -INFO 2025-09-08 16:38:46 celerate.py:281 step:20K smpl:1M ep:8K epch:4.68 loss:0.086 grdn:0.228 lr:2.5e-06 updt_s:0. -486 data_s:0.037 -INFO 2025-09-08 16:38:46 celerate.py:295 Checkpoint policy after step 20000 -INFO 2025-09-08 16:40:30 celerate.py:281 step:20K smpl:1M ep:8K epch:4.73 loss:0.086 grdn:0.232 lr:2.5e-06 updt_s:0. -509 data_s:0.003 -INFO 2025-09-08 16:42:16 celerate.py:281 step:20K smpl:1M ep:8K epch:4.77 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s:0. -527 data_s:0.003 -INFO 2025-09-08 16:44:01 celerate.py:281 step:21K smpl:1M ep:8K epch:4.82 loss:0.086 grdn:0.238 lr:2.5e-06 updt_s:0. -519 data_s:0.003 -INFO 2025-09-08 16:45:45 celerate.py:281 step:21K smpl:1M ep:8K epch:4.87 loss:0.087 grdn:0.235 lr:2.5e-06 updt_s:0. -504 data_s:0.013 -INFO 2025-09-08 16:47:29 celerate.py:281 step:21K smpl:1M ep:8K epch:4.91 loss:0.087 grdn:0.233 lr:2.5e-06 updt_s:0. -509 data_s:0.011 -INFO 2025-09-08 16:49:19 celerate.py:281 step:21K smpl:1M ep:8K epch:4.96 loss:0.086 grdn:0.234 lr:2.5e-06 updt_s:0. -544 data_s:0.003 -INFO 2025-09-08 16:51:04 celerate.py:281 step:21K smpl:1M ep:8K epch:5.01 loss:0.086 grdn:0.225 lr:2.5e-06 updt_s:0. -488 data_s:0.039 -INFO 2025-09-08 16:52:51 celerate.py:281 step:22K smpl:1M ep:9K epch:5.06 loss:0.086 grdn:0.232 lr:2.5e-06 updt_s:0. -430 data_s:0.099 -INFO 2025-09-08 16:54:36 celerate.py:281 step:22K smpl:1M ep:9K epch:5.10 loss:0.087 grdn:0.236 lr:2.5e-06 updt_s:0. -521 data_s:0.003 -INFO 2025-09-08 16:56:23 celerate.py:281 step:22K smpl:1M ep:9K epch:5.15 loss:0.086 grdn:0.233 lr:2.5e-06 updt_s:0. -521 data_s:0.014 -INFO 2025-09-08 16:58:09 celerate.py:281 step:22K smpl:1M ep:9K epch:5.20 loss:0.087 grdn:0.234 lr:2.5e-06 updt_s:0. -525 data_s:0.003 -INFO 2025-09-08 17:00:04 celerate.py:281 step:22K smpl:1M ep:9K epch:5.24 loss:0.086 grdn:0.233 lr:2.5e-06 updt_s:0. -568 data_s:0.003 -INFO 2025-09-08 17:02:00 celerate.py:281 step:23K smpl:1M ep:9K epch:5.29 loss:0.087 grdn:0.238 lr:2.5e-06 updt_s:0. -575 data_s:0.003 -INFO 2025-09-08 17:03:49 celerate.py:281 step:23K smpl:1M ep:9K epch:5.34 loss:0.087 grdn:0.233 lr:2.5e-06 updt_s:0. -513 data_s:0.030 -INFO 2025-09-08 17:05:39 celerate.py:281 step:23K smpl:1M ep:9K epch:5.38 loss:0.085 grdn:0.227 lr:2.5e-06 updt_s:0. -523 data_s:0.027 -INFO 2025-09-08 17:07:26 celerate.py:281 step:23K smpl:1M ep:9K epch:5.43 loss:0.086 grdn:0.233 lr:2.5e-06 updt_s:0. -529 data_s:0.003 -INFO 2025-09-08 17:09:12 celerate.py:281 step:23K smpl:1M ep:9K epch:5.48 loss:0.086 grdn:0.232 lr:2.5e-06 updt_s:0. -526 data_s:0.003 -INFO 2025-09-08 17:10:55 celerate.py:281 step:24K smpl:2M ep:9K epch:5.52 loss:0.087 grdn:0.230 lr:2.5e-06 updt_s:0. -443 data_s:0.072 -INFO 2025-09-08 17:12:40 celerate.py:281 step:24K smpl:2M ep:9K epch:5.57 loss:0.087 grdn:0.229 lr:2.5e-06 updt_s:0. -518 data_s:0.004 -INFO 2025-09-08 17:14:25 celerate.py:281 step:24K smpl:2M ep:10K epch:5.62 loss:0.087 grdn:0.232 lr:2.5e-06 updt_s:0 -.521 data_s:0.003 -INFO 2025-09-08 17:16:11 celerate.py:281 step:24K smpl:2M ep:10K epch:5.66 loss:0.086 grdn:0.230 lr:2.5e-06 updt_s:0 -.523 data_s:0.003 -INFO 2025-09-08 17:17:55 celerate.py:281 step:24K smpl:2M ep:10K epch:5.71 loss:0.086 grdn:0.228 lr:2.5e-06 updt_s:0 -.515 data_s:0.005 -INFO 2025-09-08 17:19:39 celerate.py:281 step:25K smpl:2M ep:10K epch:5.76 loss:0.087 grdn:0.229 lr:2.5e-06 updt_s:0 -.415 data_s:0.106 -INFO 2025-09-08 17:21:24 celerate.py:281 step:25K smpl:2M ep:10K epch:5.80 loss:0.086 grdn:0.234 lr:2.5e-06 updt_s:0 -.507 data_s:0.016 -INFO 2025-09-08 17:23:08 celerate.py:281 step:25K smpl:2M ep:10K epch:5.85 loss:0.085 grdn:0.229 lr:2.5e-06 updt_s:0 -.514 data_s:0.003 -INFO 2025-09-08 17:24:54 celerate.py:281 step:25K smpl:2M ep:10K epch:5.90 loss:0.087 grdn:0.227 lr:2.5e-06 updt_s:0 -.518 data_s:0.008 -INFO 2025-09-08 17:26:41 celerate.py:281 step:25K smpl:2M ep:10K epch:5.94 loss:0.087 grdn:0.237 lr:2.5e-06 updt_s:0 -.529 data_s:0.003 -INFO 2025-09-08 17:28:24 celerate.py:281 step:26K smpl:2M ep:10K epch:5.99 loss:0.087 grdn:0.232 lr:2.5e-06 updt_s:0 -.513 data_s:0.003 -INFO 2025-09-08 17:30:11 celerate.py:281 step:26K smpl:2M ep:10K epch:6.04 loss:0.087 grdn:0.233 lr:2.5e-06 updt_s:0 -.370 data_s:0.164 -INFO 2025-09-08 17:31:55 celerate.py:281 step:26K smpl:2M ep:10K epch:6.08 loss:0.087 grdn:0.231 lr:2.5e-06 updt_s:0 -.385 data_s:0.132 -INFO 2025-09-08 17:33:39 celerate.py:281 step:26K smpl:2M ep:10K epch:6.13 loss:0.087 grdn:0.237 lr:2.5e-06 updt_s:0 -.450 data_s:0.069 -INFO 2025-09-08 17:35:24 celerate.py:281 step:26K smpl:2M ep:10K epch:6.18 loss:0.087 grdn:0.238 lr:2.5e-06 updt_s:0 -.468 data_s:0.052 -INFO 2025-09-08 17:37:07 celerate.py:281 step:27K smpl:2M ep:11K epch:6.23 loss:0.087 grdn:0.233 lr:2.5e-06 updt_s:0 -.514 data_s:0.004 -INFO 2025-09-08 17:38:52 celerate.py:281 step:27K smpl:2M ep:11K epch:6.27 loss:0.087 grdn:0.236 lr:2.5e-06 updt_s:0 -.519 data_s:0.003 -INFO 2025-09-08 17:40:40 celerate.py:281 step:27K smpl:2M ep:11K epch:6.32 loss:0.086 grdn:0.234 lr:2.5e-06 updt_s:0 -.534 data_s:0.003 -INFO 2025-09-08 17:42:57 celerate.py:281 step:27K smpl:2M ep:11K epch:6.37 loss:0.087 grdn:0.231 lr:2.5e-06 updt_s:0 -.678 data_s:0.007 -INFO 2025-09-08 17:46:13 celerate.py:281 step:27K smpl:2M ep:11K epch:6.41 loss:0.086 grdn:0.233 lr:2.5e-06 updt_s:0 -.968 data_s:0.009 -INFO 2025-09-08 17:49:20 celerate.py:281 step:28K smpl:2M ep:11K epch:6.46 loss:0.087 grdn:0.231 lr:2.5e-06 updt_s:0 -.895 data_s:0.037 -INFO 2025-09-08 17:51:22 celerate.py:281 step:28K smpl:2M ep:11K epch:6.51 loss:0.087 grdn:0.235 lr:2.5e-06 updt_s:0 -.604 data_s:0.003 -INFO 2025-09-08 17:53:07 celerate.py:281 step:28K smpl:2M ep:11K epch:6.55 loss:0.086 grdn:0.236 lr:2.5e-06 updt_s:0 -.521 data_s:0.003 -INFO 2025-09-08 17:54:51 celerate.py:281 step:28K smpl:2M ep:11K epch:6.60 loss:0.087 grdn:0.234 lr:2.5e-06 updt_s:0 -.516 data_s:0.003 -INFO 2025-09-08 17:56:36 celerate.py:281 step:28K smpl:2M ep:11K epch:6.65 loss:0.086 grdn:0.232 lr:2.5e-06 updt_s:0 -.519 data_s:0.003 -INFO 2025-09-08 17:58:21 celerate.py:281 step:29K smpl:2M ep:11K epch:6.69 loss:0.085 grdn:0.228 lr:2.5e-06 updt_s:0 -.521 data_s:0.003 -INFO 2025-09-08 18:00:06 celerate.py:281 step:29K smpl:2M ep:11K epch:6.74 loss:0.087 grdn:0.237 lr:2.5e-06 updt_s:0 -.513 data_s:0.011 -INFO 2025-09-08 18:01:50 celerate.py:281 step:29K smpl:2M ep:11K epch:6.79 loss:0.086 grdn:0.232 lr:2.5e-06 updt_s:0 -.476 data_s:0.041 -INFO 2025-09-08 18:03:34 celerate.py:281 step:29K smpl:2M ep:12K epch:6.83 loss:0.086 grdn:0.236 lr:2.5e-06 updt_s:0 -.506 data_s:0.012 -INFO 2025-09-08 18:05:21 celerate.py:281 step:29K smpl:2M ep:12K epch:6.88 loss:0.086 grdn:0.229 lr:2.5e-06 updt_s:0 -.455 data_s:0.075 -INFO 2025-09-08 18:07:04 celerate.py:281 step:30K smpl:2M ep:12K epch:6.93 loss:0.086 grdn:0.233 lr:2.5e-06 updt_s:0 -.514 data_s:0.003 -INFO 2025-09-08 18:08:47 celerate.py:281 step:30K smpl:2M ep:12K epch:6.97 loss:0.086 grdn:0.236 lr:2.5e-06 updt_s:0 -.509 data_s:0.003 -INFO 2025-09-08 18:10:33 celerate.py:281 step:30K smpl:2M ep:12K epch:7.02 loss:0.087 grdn:0.237 lr:2.5e-06 updt_s:0 -.422 data_s:0.105 -INFO 2025-09-08 18:12:19 celerate.py:281 step:30K smpl:2M ep:12K epch:7.07 loss:0.087 grdn:0.236 lr:2.5e-06 updt_s:0 -.347 data_s:0.182 -INFO 2025-09-08 18:14:05 celerate.py:281 step:30K smpl:2M ep:12K epch:7.11 loss:0.086 grdn:0.231 lr:2.5e-06 updt_s:0 -.473 data_s:0.053 -INFO 2025-09-08 18:15:52 celerate.py:281 step:31K smpl:2M ep:12K epch:7.16 loss:0.086 grdn:0.238 lr:2.5e-06 updt_s:0 -.531 data_s:0.005 -INFO 2025-09-08 18:17:37 celerate.py:281 step:31K smpl:2M ep:12K epch:7.21 loss:0.086 grdn:0.235 lr:2.5e-06 updt_s:0 -.520 data_s:0.003 -INFO 2025-09-08 18:19:22 celerate.py:281 step:31K smpl:2M ep:12K epch:7.26 loss:0.086 grdn:0.231 lr:2.5e-06 updt_s:0 -.500 data_s:0.020 -INFO 2025-09-08 18:21:06 celerate.py:281 step:31K smpl:2M ep:12K epch:7.30 loss:0.087 grdn:0.243 lr:2.5e-06 updt_s:0 -.511 data_s:0.009 -INFO 2025-09-08 18:22:50 celerate.py:281 step:31K smpl:2M ep:12K epch:7.35 loss:0.087 grdn:0.227 lr:2.5e-06 updt_s:0 -.518 data_s:0.003 -INFO 2025-09-08 18:24:33 celerate.py:281 step:32K smpl:2M ep:13K epch:7.40 loss:0.086 grdn:0.235 lr:2.5e-06 updt_s:0 -.507 data_s:0.007 -INFO 2025-09-08 18:26:16 celerate.py:281 step:32K smpl:2M ep:13K epch:7.44 loss:0.087 grdn:0.238 lr:2.5e-06 updt_s:0 -.463 data_s:0.047 -INFO 2025-09-08 18:27:59 celerate.py:281 step:32K smpl:2M ep:13K epch:7.49 loss:0.087 grdn:0.240 lr:2.5e-06 updt_s:0 -.509 data_s:0.007 -INFO 2025-09-08 18:29:43 celerate.py:281 step:32K smpl:2M ep:13K epch:7.54 loss:0.087 grdn:0.234 lr:2.5e-06 updt_s:0 -.514 data_s:0.003 -INFO 2025-09-08 18:31:26 celerate.py:281 step:32K smpl:2M ep:13K epch:7.58 loss:0.086 grdn:0.234 lr:2.5e-06 updt_s:0 -.511 data_s:0.003 -INFO 2025-09-08 18:33:11 celerate.py:281 step:33K smpl:2M ep:13K epch:7.63 loss:0.086 grdn:0.233 lr:2.5e-06 updt_s:0 -.518 data_s:0.003 -INFO 2025-09-08 18:34:57 celerate.py:281 step:33K smpl:2M ep:13K epch:7.68 loss:0.087 grdn:0.236 lr:2.5e-06 updt_s:0 -.512 data_s:0.016 -INFO 2025-09-08 18:36:41 celerate.py:281 step:33K smpl:2M ep:13K epch:7.72 loss:0.086 grdn:0.229 lr:2.5e-06 updt_s:0 -.517 data_s:0.003 -INFO 2025-09-08 18:38:25 celerate.py:281 step:33K smpl:2M ep:13K epch:7.77 loss:0.087 grdn:0.236 lr:2.5e-06 updt_s:0 -.516 data_s:0.003 -INFO 2025-09-08 18:40:07 celerate.py:281 step:33K smpl:2M ep:13K epch:7.82 loss:0.086 grdn:0.235 lr:2.5e-06 updt_s:0 -.506 data_s:0.003 -INFO 2025-09-08 18:41:51 celerate.py:281 step:34K smpl:2M ep:13K epch:7.86 loss:0.087 grdn:0.234 lr:2.5e-06 updt_s:0 -.509 data_s:0.006 -INFO 2025-09-08 18:43:36 celerate.py:281 step:34K smpl:2M ep:13K epch:7.91 loss:0.086 grdn:0.239 lr:2.5e-06 updt_s:0 -.521 data_s:0.003 -INFO 2025-09-08 18:45:19 celerate.py:281 step:34K smpl:2M ep:13K epch:7.96 loss:0.086 grdn:0.233 lr:2.5e-06 updt_s:0 -.511 data_s:0.003 -INFO 2025-09-08 18:47:05 celerate.py:281 step:34K smpl:2M ep:14K epch:8.00 loss:0.086 grdn:0.232 lr:2.5e-06 updt_s:0 -.495 data_s:0.035 -INFO 2025-09-08 18:48:51 celerate.py:281 step:34K smpl:2M ep:14K epch:8.05 loss:0.087 grdn:0.243 lr:2.5e-06 updt_s:0 -.413 data_s:0.112 -INFO 2025-09-08 18:50:34 celerate.py:281 step:35K smpl:2M ep:14K epch:8.10 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s:0 -.515 data_s:0.003 -INFO 2025-09-08 18:52:19 celerate.py:281 step:35K smpl:2M ep:14K epch:8.14 loss:0.087 grdn:0.237 lr:2.5e-06 updt_s:0 -.520 data_s:0.003 -INFO 2025-09-08 18:54:03 celerate.py:281 step:35K smpl:2M ep:14K epch:8.19 loss:0.087 grdn:0.231 lr:2.5e-06 updt_s:0 -.515 data_s:0.003 -INFO 2025-09-08 18:55:48 celerate.py:281 step:35K smpl:2M ep:14K epch:8.24 loss:0.086 grdn:0.231 lr:2.5e-06 updt_s:0 -.420 data_s:0.101 -INFO 2025-09-08 18:57:33 celerate.py:281 step:35K smpl:2M ep:14K epch:8.28 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s:0 -.506 data_s:0.022 -INFO 2025-09-08 18:59:19 celerate.py:281 step:36K smpl:2M ep:14K epch:8.33 loss:0.087 grdn:0.235 lr:2.5e-06 updt_s:0 -.525 data_s:0.003 -INFO 2025-09-08 19:01:03 celerate.py:281 step:36K smpl:2M ep:14K epch:8.38 loss:0.086 grdn:0.228 lr:2.5e-06 updt_s:0 -.516 data_s:0.003 -INFO 2025-09-08 19:02:48 celerate.py:281 step:36K smpl:2M ep:14K epch:8.43 loss:0.086 grdn:0.232 lr:2.5e-06 updt_s:0 -.516 data_s:0.003 -INFO 2025-09-08 19:04:32 celerate.py:281 step:36K smpl:2M ep:14K epch:8.47 loss:0.087 grdn:0.235 lr:2.5e-06 updt_s:0 -.505 data_s:0.013 -INFO 2025-09-08 19:06:15 celerate.py:281 step:36K smpl:2M ep:14K epch:8.52 loss:0.086 grdn:0.232 lr:2.5e-06 updt_s:0 -.384 data_s:0.130 -INFO 2025-09-08 19:07:58 celerate.py:281 step:37K smpl:2M ep:15K epch:8.57 loss:0.087 grdn:0.235 lr:2.5e-06 updt_s:0 -.430 data_s:0.084 -INFO 2025-09-08 19:09:41 celerate.py:281 step:37K smpl:2M ep:15K epch:8.61 loss:0.087 grdn:0.234 lr:2.5e-06 updt_s:0 -.351 data_s:0.162 -INFO 2025-09-08 19:11:25 celerate.py:281 step:37K smpl:2M ep:15K epch:8.66 loss:0.086 grdn:0.239 lr:2.5e-06 updt_s:0 -.337 data_s:0.181 -INFO 2025-09-08 19:13:08 celerate.py:281 step:37K smpl:2M ep:15K epch:8.71 loss:0.086 grdn:0.232 lr:2.5e-06 updt_s:0 -.336 data_s:0.177 -INFO 2025-09-08 19:14:52 celerate.py:281 step:37K smpl:2M ep:15K epch:8.75 loss:0.086 grdn:0.235 lr:2.5e-06 updt_s:0 -.344 data_s:0.174 -INFO 2025-09-08 19:16:35 celerate.py:281 step:38K smpl:2M ep:15K epch:8.80 loss:0.086 grdn:0.231 lr:2.5e-06 updt_s:0 -.332 data_s:0.182 -INFO 2025-09-08 19:18:18 celerate.py:281 step:38K smpl:2M ep:15K epch:8.85 loss:0.086 grdn:0.234 lr:2.5e-06 updt_s:0 -.332 data_s:0.182 -INFO 2025-09-08 19:20:01 celerate.py:281 step:38K smpl:2M ep:15K epch:8.89 loss:0.087 grdn:0.233 lr:2.5e-06 updt_s:0 -.337 data_s:0.177 -INFO 2025-09-08 19:21:45 celerate.py:281 step:38K smpl:2M ep:15K epch:8.94 loss:0.086 grdn:0.234 lr:2.5e-06 updt_s:0 -.495 data_s:0.022 -INFO 2025-09-08 19:23:29 celerate.py:281 step:38K smpl:2M ep:15K epch:8.99 loss:0.086 grdn:0.235 lr:2.5e-06 updt_s:0 -.513 data_s:0.003 -INFO 2025-09-08 19:25:15 celerate.py:281 step:39K smpl:2M ep:15K epch:9.03 loss:0.086 grdn:0.234 lr:2.5e-06 updt_s:0 -.436 data_s:0.097 -INFO 2025-09-08 19:26:59 celerate.py:281 step:39K smpl:2M ep:15K epch:9.08 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s:0 -.378 data_s:0.138 -INFO 2025-09-08 19:28:43 celerate.py:281 step:39K smpl:2M ep:15K epch:9.13 loss:0.087 grdn:0.241 lr:2.5e-06 updt_s:0 -.497 data_s:0.023 -INFO 2025-09-08 19:30:27 celerate.py:281 step:39K smpl:3M ep:16K epch:9.17 loss:0.086 grdn:0.235 lr:2.5e-06 updt_s:0 -.515 data_s:0.003 -INFO 2025-09-08 19:32:14 celerate.py:281 step:39K smpl:3M ep:16K epch:9.22 loss:0.086 grdn:0.233 lr:2.5e-06 updt_s:0 -.526 data_s:0.003 -INFO 2025-09-08 19:33:57 celerate.py:281 step:40K smpl:3M ep:16K epch:9.27 loss:0.087 grdn:0.234 lr:2.5e-06 updt_s:0 -.506 data_s:0.011 -INFO 2025-09-08 19:35:42 celerate.py:281 step:40K smpl:3M ep:16K epch:9.31 loss:0.087 grdn:0.229 lr:2.5e-06 updt_s:0 -.455 data_s:0.066 -INFO 2025-09-08 19:37:26 celerate.py:281 step:40K smpl:3M ep:16K epch:9.36 loss:0.086 grdn:0.238 lr:2.5e-06 updt_s:0 -.493 data_s:0.026 -INFO 2025-09-08 19:37:26 celerate.py:295 Checkpoint policy after step 40000 -INFO 2025-09-08 19:39:10 celerate.py:281 step:40K smpl:3M ep:16K epch:9.41 loss:0.086 grdn:0.236 lr:2.5e-06 updt_s:0 -.397 data_s:0.114 -INFO 2025-09-08 19:40:53 celerate.py:281 step:40K smpl:3M ep:16K epch:9.45 loss:0.086 grdn:0.239 lr:2.5e-06 updt_s:0 -.344 data_s:0.168 -INFO 2025-09-08 19:42:37 celerate.py:281 step:41K smpl:3M ep:16K epch:9.50 loss:0.086 grdn:0.234 lr:2.5e-06 updt_s:0 -.480 data_s:0.036 -INFO 2025-09-08 19:44:21 celerate.py:281 step:41K smpl:3M ep:16K epch:9.55 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s:0 -.517 data_s:0.003 -INFO 2025-09-08 19:46:05 celerate.py:281 step:41K smpl:3M ep:16K epch:9.60 loss:0.086 grdn:0.233 lr:2.5e-06 updt_s:0 -.517 data_s:0.003 -INFO 2025-09-08 19:47:49 celerate.py:281 step:41K smpl:3M ep:16K epch:9.64 loss:0.086 grdn:0.234 lr:2.5e-06 updt_s:0 -.513 data_s:0.003 -INFO 2025-09-08 19:49:33 celerate.py:281 step:41K smpl:3M ep:16K epch:9.69 loss:0.087 grdn:0.236 lr:2.5e-06 updt_s:0 -.515 data_s:0.003 -INFO 2025-09-08 19:51:17 celerate.py:281 step:42K smpl:3M ep:16K epch:9.74 loss:0.086 grdn:0.228 lr:2.5e-06 updt_s:0 -.515 data_s:0.003 -INFO 2025-09-08 19:53:00 celerate.py:281 step:42K smpl:3M ep:17K epch:9.78 loss:0.086 grdn:0.232 lr:2.5e-06 updt_s:0 -.513 data_s:0.003 -INFO 2025-09-08 19:54:44 celerate.py:281 step:42K smpl:3M ep:17K epch:9.83 loss:0.086 grdn:0.233 lr:2.5e-06 updt_s:0 -.516 data_s:0.003 -INFO 2025-09-08 19:56:28 celerate.py:281 step:42K smpl:3M ep:17K epch:9.88 loss:0.087 grdn:0.237 lr:2.5e-06 updt_s:0 -.512 data_s:0.003 -INFO 2025-09-08 19:58:11 celerate.py:281 step:42K smpl:3M ep:17K epch:9.92 loss:0.086 grdn:0.233 lr:2.5e-06 updt_s:0 -.514 data_s:0.003 -INFO 2025-09-08 19:59:55 celerate.py:281 step:43K smpl:3M ep:17K epch:9.97 loss:0.087 grdn:0.238 lr:2.5e-06 updt_s:0 -.514 data_s:0.003 -INFO 2025-09-08 20:01:42 celerate.py:281 step:43K smpl:3M ep:17K epch:10.02 loss:0.087 grdn:0.234 lr:2.5e-06 updt_s: -0.476 data_s:0.057 -INFO 2025-09-08 20:03:25 celerate.py:281 step:43K smpl:3M ep:17K epch:10.06 loss:0.087 grdn:0.239 lr:2.5e-06 updt_s: -0.471 data_s:0.043 -INFO 2025-09-08 20:05:09 celerate.py:281 step:43K smpl:3M ep:17K epch:10.11 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.515 data_s:0.004 -INFO 2025-09-08 20:06:53 celerate.py:281 step:43K smpl:3M ep:17K epch:10.16 loss:0.087 grdn:0.237 lr:2.5e-06 updt_s: -0.505 data_s:0.013 -INFO 2025-09-08 20:08:36 celerate.py:281 step:44K smpl:3M ep:17K epch:10.20 loss:0.087 grdn:0.237 lr:2.5e-06 updt_s: -0.511 data_s:0.003 -INFO 2025-09-08 20:10:20 celerate.py:281 step:44K smpl:3M ep:17K epch:10.25 loss:0.086 grdn:0.236 lr:2.5e-06 updt_s: -0.516 data_s:0.003 -INFO 2025-09-08 20:12:04 celerate.py:281 step:44K smpl:3M ep:17K epch:10.30 loss:0.086 grdn:0.232 lr:2.5e-06 updt_s: -0.511 data_s:0.003 -INFO 2025-09-08 20:13:47 celerate.py:281 step:44K smpl:3M ep:18K epch:10.34 loss:0.087 grdn:0.235 lr:2.5e-06 updt_s: -0.503 data_s:0.011 -INFO 2025-09-08 20:15:31 celerate.py:281 step:44K smpl:3M ep:18K epch:10.39 loss:0.086 grdn:0.233 lr:2.5e-06 updt_s: -0.416 data_s:0.102 -INFO 2025-09-08 20:17:15 celerate.py:281 step:45K smpl:3M ep:18K epch:10.44 loss:0.087 grdn:0.235 lr:2.5e-06 updt_s: -0.502 data_s:0.017 -INFO 2025-09-08 20:18:58 celerate.py:281 step:45K smpl:3M ep:18K epch:10.48 loss:0.086 grdn:0.234 lr:2.5e-06 updt_s: -0.512 data_s:0.003 -INFO 2025-09-08 20:20:41 celerate.py:281 step:45K smpl:3M ep:18K epch:10.53 loss:0.086 grdn:0.238 lr:2.5e-06 updt_s: -0.496 data_s:0.017 -INFO 2025-09-08 20:22:24 celerate.py:281 step:45K smpl:3M ep:18K epch:10.58 loss:0.086 grdn:0.233 lr:2.5e-06 updt_s: -0.493 data_s:0.022 -INFO 2025-09-08 20:24:08 celerate.py:281 step:45K smpl:3M ep:18K epch:10.63 loss:0.086 grdn:0.236 lr:2.5e-06 updt_s: -0.485 data_s:0.031 -INFO 2025-09-08 20:25:52 celerate.py:281 step:46K smpl:3M ep:18K epch:10.67 loss:0.086 grdn:0.231 lr:2.5e-06 updt_s: -0.518 data_s:0.003 -INFO 2025-09-08 20:27:36 celerate.py:281 step:46K smpl:3M ep:18K epch:10.72 loss:0.085 grdn:0.228 lr:2.5e-06 updt_s: -0.513 data_s:0.003 -INFO 2025-09-08 20:29:19 celerate.py:281 step:46K smpl:3M ep:18K epch:10.77 loss:0.087 grdn:0.235 lr:2.5e-06 updt_s: -0.514 data_s:0.003 -INFO 2025-09-08 20:31:04 celerate.py:281 step:46K smpl:3M ep:18K epch:10.81 loss:0.086 grdn:0.242 lr:2.5e-06 updt_s: -0.516 data_s:0.003 -INFO 2025-09-08 20:32:47 celerate.py:281 step:46K smpl:3M ep:18K epch:10.86 loss:0.086 grdn:0.235 lr:2.5e-06 updt_s: -0.513 data_s:0.003 -INFO 2025-09-08 20:34:32 celerate.py:281 step:47K smpl:3M ep:18K epch:10.91 loss:0.086 grdn:0.234 lr:2.5e-06 updt_s: -0.518 data_s:0.003 -INFO 2025-09-08 20:36:15 celerate.py:281 step:47K smpl:3M ep:19K epch:10.95 loss:0.087 grdn:0.238 lr:2.5e-06 updt_s: -0.514 data_s:0.003 -INFO 2025-09-08 20:38:01 celerate.py:281 step:47K smpl:3M ep:19K epch:11.00 loss:0.086 grdn:0.239 lr:2.5e-06 updt_s: -0.416 data_s:0.114 -INFO 2025-09-08 20:39:45 celerate.py:281 step:47K smpl:3M ep:19K epch:11.05 loss:0.086 grdn:0.233 lr:2.5e-06 updt_s: -0.429 data_s:0.086 -INFO 2025-09-08 20:41:28 celerate.py:281 step:47K smpl:3M ep:19K epch:11.09 loss:0.086 grdn:0.229 lr:2.5e-06 updt_s: -0.409 data_s:0.106 -INFO 2025-09-08 20:43:12 celerate.py:281 step:48K smpl:3M ep:19K epch:11.14 loss:0.086 grdn:0.238 lr:2.5e-06 updt_s: -0.470 data_s:0.050 -INFO 2025-09-08 20:44:56 celerate.py:281 step:48K smpl:3M ep:19K epch:11.19 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.516 data_s:0.003 -INFO 2025-09-08 20:46:41 celerate.py:281 step:48K smpl:3M ep:19K epch:11.23 loss:0.087 grdn:0.243 lr:2.5e-06 updt_s: -0.517 data_s:0.003 -INFO 2025-09-08 20:48:25 celerate.py:281 step:48K smpl:3M ep:19K epch:11.28 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.518 data_s:0.003 -INFO 2025-09-08 20:50:08 celerate.py:281 step:48K smpl:3M ep:19K epch:11.33 loss:0.086 grdn:0.238 lr:2.5e-06 updt_s: -0.505 data_s:0.009 -INFO 2025-09-08 20:51:52 celerate.py:281 step:49K smpl:3M ep:19K epch:11.37 loss:0.087 grdn:0.245 lr:2.5e-06 updt_s: -0.449 data_s:0.067 -INFO 2025-09-08 20:53:35 celerate.py:281 step:49K smpl:3M ep:19K epch:11.42 loss:0.086 grdn:0.238 lr:2.5e-06 updt_s: -0.420 data_s:0.094 -INFO 2025-09-08 20:55:19 celerate.py:281 step:49K smpl:3M ep:19K epch:11.47 loss:0.086 grdn:0.232 lr:2.5e-06 updt_s: -0.515 data_s:0.003 -INFO 2025-09-08 20:57:01 celerate.py:281 step:49K smpl:3M ep:19K epch:11.51 loss:0.087 grdn:0.236 lr:2.5e-06 updt_s: -0.505 data_s:0.003 -INFO 2025-09-08 20:58:44 celerate.py:281 step:49K smpl:3M ep:20K epch:11.56 loss:0.086 grdn:0.234 lr:2.5e-06 updt_s: -0.511 data_s:0.003 -INFO 2025-09-08 21:00:28 celerate.py:281 step:50K smpl:3M ep:20K epch:11.61 loss:0.087 grdn:0.239 lr:2.5e-06 updt_s: -0.516 data_s:0.003 -INFO 2025-09-08 21:02:11 celerate.py:281 step:50K smpl:3M ep:20K epch:11.65 loss:0.086 grdn:0.233 lr:2.5e-06 updt_s: -0.402 data_s:0.110 -INFO 2025-09-08 21:03:54 celerate.py:281 step:50K smpl:3M ep:20K epch:11.70 loss:0.086 grdn:0.238 lr:2.5e-06 updt_s: -0.332 data_s:0.184 -INFO 2025-09-08 21:05:37 celerate.py:281 step:50K smpl:3M ep:20K epch:11.75 loss:0.086 grdn:0.229 lr:2.5e-06 updt_s: -0.332 data_s:0.182 -INFO 2025-09-08 21:07:21 celerate.py:281 step:50K smpl:3M ep:20K epch:11.80 loss:0.087 grdn:0.237 lr:2.5e-06 updt_s: -0.466 data_s:0.049 -INFO 2025-09-08 21:09:05 celerate.py:281 step:51K smpl:3M ep:20K epch:11.84 loss:0.086 grdn:0.238 lr:2.5e-06 updt_s: -0.517 data_s:0.003 -INFO 2025-09-08 21:10:49 celerate.py:281 step:51K smpl:3M ep:20K epch:11.89 loss:0.087 grdn:0.240 lr:2.5e-06 updt_s: -0.512 data_s:0.004 -INFO 2025-09-08 21:12:32 celerate.py:281 step:51K smpl:3M ep:20K epch:11.94 loss:0.085 grdn:0.234 lr:2.5e-06 updt_s: -0.484 data_s:0.032 -INFO 2025-09-08 21:14:17 celerate.py:281 step:51K smpl:3M ep:20K epch:11.98 loss:0.087 grdn:0.236 lr:2.5e-06 updt_s: -0.517 data_s:0.004 -INFO 2025-09-08 21:16:03 celerate.py:281 step:51K smpl:3M ep:20K epch:12.03 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.424 data_s:0.105 -INFO 2025-09-08 21:17:46 celerate.py:281 step:52K smpl:3M ep:20K epch:12.08 loss:0.086 grdn:0.236 lr:2.5e-06 updt_s: -0.442 data_s:0.073 -INFO 2025-09-08 21:19:30 celerate.py:281 step:52K smpl:3M ep:21K epch:12.12 loss:0.086 grdn:0.231 lr:2.5e-06 updt_s: -0.511 data_s:0.007 -INFO 2025-09-08 21:21:15 celerate.py:281 step:52K smpl:3M ep:21K epch:12.17 loss:0.086 grdn:0.241 lr:2.5e-06 updt_s: -0.520 data_s:0.003 -INFO 2025-09-08 21:22:59 celerate.py:281 step:52K smpl:3M ep:21K epch:12.22 loss:0.086 grdn:0.242 lr:2.5e-06 updt_s: -0.515 data_s:0.003 -INFO 2025-09-08 21:24:43 celerate.py:281 step:52K smpl:3M ep:21K epch:12.26 loss:0.086 grdn:0.241 lr:2.5e-06 updt_s: -0.518 data_s:0.003 -INFO 2025-09-08 21:26:27 celerate.py:281 step:53K smpl:3M ep:21K epch:12.31 loss:0.086 grdn:0.241 lr:2.5e-06 updt_s: -0.514 data_s:0.003 -INFO 2025-09-08 21:28:11 celerate.py:281 step:53K smpl:3M ep:21K epch:12.36 loss:0.087 grdn:0.241 lr:2.5e-06 updt_s: -0.517 data_s:0.003 -INFO 2025-09-08 21:29:55 celerate.py:281 step:53K smpl:3M ep:21K epch:12.40 loss:0.087 grdn:0.236 lr:2.5e-06 updt_s: -0.514 data_s:0.003 -INFO 2025-09-08 21:31:39 celerate.py:281 step:53K smpl:3M ep:21K epch:12.45 loss:0.086 grdn:0.240 lr:2.5e-06 updt_s: -0.517 data_s:0.003 -INFO 2025-09-08 21:33:23 celerate.py:281 step:53K smpl:3M ep:21K epch:12.50 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.515 data_s:0.003 -INFO 2025-09-08 21:35:08 celerate.py:281 step:54K smpl:3M ep:21K epch:12.54 loss:0.086 grdn:0.236 lr:2.5e-06 updt_s: -0.519 data_s:0.003 -INFO 2025-09-08 21:36:51 celerate.py:281 step:54K smpl:3M ep:21K epch:12.59 loss:0.086 grdn:0.232 lr:2.5e-06 updt_s: -0.511 data_s:0.003 -INFO 2025-09-08 21:38:36 celerate.py:281 step:54K smpl:3M ep:21K epch:12.64 loss:0.087 grdn:0.236 lr:2.5e-06 updt_s: -0.517 data_s:0.003 -INFO 2025-09-08 21:40:18 celerate.py:281 step:54K smpl:3M ep:21K epch:12.68 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.473 data_s:0.038 -INFO 2025-09-08 21:42:02 celerate.py:281 step:54K smpl:3M ep:22K epch:12.73 loss:0.085 grdn:0.232 lr:2.5e-06 updt_s: -0.408 data_s:0.109 -INFO 2025-09-08 21:43:45 celerate.py:281 step:55K smpl:3M ep:22K epch:12.78 loss:0.086 grdn:0.244 lr:2.5e-06 updt_s: -0.377 data_s:0.136 -INFO 2025-09-08 21:45:29 celerate.py:281 step:55K smpl:4M ep:22K epch:12.83 loss:0.087 grdn:0.233 lr:2.5e-06 updt_s: -0.497 data_s:0.022 -INFO 2025-09-08 21:47:12 celerate.py:281 step:55K smpl:4M ep:22K epch:12.87 loss:0.086 grdn:0.235 lr:2.5e-06 updt_s: -0.512 data_s:0.003 -INFO 2025-09-08 21:48:56 celerate.py:281 step:55K smpl:4M ep:22K epch:12.92 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.429 data_s:0.086 -INFO 2025-09-08 21:50:39 celerate.py:281 step:55K smpl:4M ep:22K epch:12.97 loss:0.087 grdn:0.241 lr:2.5e-06 updt_s: -0.454 data_s:0.059 -INFO 2025-09-08 21:52:25 celerate.py:281 step:56K smpl:4M ep:22K epch:13.01 loss:0.086 grdn:0.235 lr:2.5e-06 updt_s: -0.459 data_s:0.072 -INFO 2025-09-08 21:54:08 celerate.py:281 step:56K smpl:4M ep:22K epch:13.06 loss:0.086 grdn:0.232 lr:2.5e-06 updt_s: -0.382 data_s:0.132 -INFO 2025-09-08 21:55:51 celerate.py:281 step:56K smpl:4M ep:22K epch:13.11 loss:0.086 grdn:0.236 lr:2.5e-06 updt_s: -0.500 data_s:0.016 -INFO 2025-09-08 21:57:36 celerate.py:281 step:56K smpl:4M ep:22K epch:13.15 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.517 data_s:0.003 -INFO 2025-09-08 21:59:20 celerate.py:281 step:56K smpl:4M ep:22K epch:13.20 loss:0.085 grdn:0.235 lr:2.5e-06 updt_s: -0.518 data_s:0.003 -INFO 2025-09-08 22:01:03 celerate.py:281 step:57K smpl:4M ep:22K epch:13.25 loss:0.086 grdn:0.244 lr:2.5e-06 updt_s: -0.510 data_s:0.003 -INFO 2025-09-08 22:02:48 celerate.py:281 step:57K smpl:4M ep:23K epch:13.29 loss:0.086 grdn:0.242 lr:2.5e-06 updt_s: -0.517 data_s:0.003 -INFO 2025-09-08 22:04:32 celerate.py:281 step:57K smpl:4M ep:23K epch:13.34 loss:0.086 grdn:0.236 lr:2.5e-06 updt_s: -0.515 data_s:0.003 -INFO 2025-09-08 22:06:14 celerate.py:281 step:57K smpl:4M ep:23K epch:13.39 loss:0.087 grdn:0.244 lr:2.5e-06 updt_s: -0.505 data_s:0.005 -bINFO 2025-09-08 22:07:59 celerate.py:281 step:57K smpl:4M ep:23K epch:13.43 loss:0.086 grdn:0.239 lr:2.5e-06 updt_s -:0.496 data_s:0.026 -INFO 2025-09-08 22:09:43 celerate.py:281 step:58K smpl:4M ep:23K epch:13.48 loss:0.087 grdn:0.239 lr:2.5e-06 updt_s: -0.438 data_s:0.080 -INFO 2025-09-08 22:11:27 celerate.py:281 step:58K smpl:4M ep:23K epch:13.53 loss:0.087 grdn:0.240 lr:2.5e-06 updt_s: -0.444 data_s:0.073 -INFO 2025-09-08 22:13:11 celerate.py:281 step:58K smpl:4M ep:23K epch:13.57 loss:0.086 grdn:0.240 lr:2.5e-06 updt_s: -0.515 data_s:0.003 -INFO 2025-09-08 22:14:55 celerate.py:281 step:58K smpl:4M ep:23K epch:13.62 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.518 data_s:0.003 -INFO 2025-09-08 22:16:39 celerate.py:281 step:58K smpl:4M ep:23K epch:13.67 loss:0.086 grdn:0.233 lr:2.5e-06 updt_s: -0.513 data_s:0.003 -INFO 2025-09-08 22:18:22 celerate.py:281 step:59K smpl:4M ep:23K epch:13.71 loss:0.087 grdn:0.240 lr:2.5e-06 updt_s: -0.513 data_s:0.003 -INFO 2025-09-08 22:20:05 celerate.py:281 step:59K smpl:4M ep:23K epch:13.76 loss:0.087 grdn:0.239 lr:2.5e-06 updt_s: -0.508 data_s:0.003 -INFO 2025-09-08 22:21:48 celerate.py:281 step:59K smpl:4M ep:23K epch:13.81 loss:0.086 grdn:0.228 lr:2.5e-06 updt_s: -0.505 data_s:0.008 -INFO 2025-09-08 22:23:31 celerate.py:281 step:59K smpl:4M ep:23K epch:13.85 loss:0.086 grdn:0.238 lr:2.5e-06 updt_s: -0.505 data_s:0.008 -INFO 2025-09-08 22:25:15 celerate.py:281 step:59K smpl:4M ep:24K epch:13.90 loss:0.086 grdn:0.240 lr:2.5e-06 updt_s: -0.515 data_s:0.003 -INFO 2025-09-08 22:26:58 celerate.py:281 step:60K smpl:4M ep:24K epch:13.95 loss:0.087 grdn:0.235 lr:2.5e-06 updt_s: -0.493 data_s:0.022 -INFO 2025-09-08 22:28:42 celerate.py:281 step:60K smpl:4M ep:24K epch:14.00 loss:0.086 grdn:0.231 lr:2.5e-06 updt_s: -0.515 data_s:0.003 -INFO 2025-09-08 22:30:29 celerate.py:281 step:60K smpl:4M ep:24K epch:14.04 loss:0.086 grdn:0.235 lr:2.5e-06 updt_s: -0.435 data_s:0.101 -INFO 2025-09-08 22:30:29 celerate.py:295 Checkpoint policy after step 60000 -INFO 2025-09-08 22:32:14 celerate.py:281 step:60K smpl:4M ep:24K epch:14.09 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.508 data_s:0.003 -INFO 2025-09-08 22:33:58 celerate.py:281 step:60K smpl:4M ep:24K epch:14.14 loss:0.086 grdn:0.238 lr:2.5e-06 updt_s: -0.516 data_s:0.003 -INFO 2025-09-08 22:35:42 celerate.py:281 step:61K smpl:4M ep:24K epch:14.18 loss:0.086 grdn:0.241 lr:2.5e-06 updt_s: -0.515 data_s:0.003 -INFO 2025-09-08 22:37:25 celerate.py:281 step:61K smpl:4M ep:24K epch:14.23 loss:0.086 grdn:0.240 lr:2.5e-06 updt_s: -0.513 data_s:0.003 -INFO 2025-09-08 22:39:09 celerate.py:281 step:61K smpl:4M ep:24K epch:14.28 loss:0.087 grdn:0.235 lr:2.5e-06 updt_s: -0.514 data_s:0.003 -INFO 2025-09-08 22:40:52 celerate.py:281 step:61K smpl:4M ep:24K epch:14.32 loss:0.086 grdn:0.242 lr:2.5e-06 updt_s: -0.509 data_s:0.003 -INFO 2025-09-08 22:42:35 celerate.py:281 step:61K smpl:4M ep:24K epch:14.37 loss:0.086 grdn:0.235 lr:2.5e-06 updt_s: -0.513 data_s:0.003 -INFO 2025-09-08 22:44:18 celerate.py:281 step:62K smpl:4M ep:24K epch:14.42 loss:0.086 grdn:0.233 lr:2.5e-06 updt_s: -0.511 data_s:0.003 -INFO 2025-09-08 22:46:02 celerate.py:281 step:62K smpl:4M ep:24K epch:14.46 loss:0.086 grdn:0.235 lr:2.5e-06 updt_s: -0.515 data_s:0.003 -INFO 2025-09-08 22:47:47 celerate.py:281 step:62K smpl:4M ep:25K epch:14.51 loss:0.086 grdn:0.242 lr:2.5e-06 updt_s: -0.517 data_s:0.003 -INFO 2025-09-08 22:49:30 celerate.py:281 step:62K smpl:4M ep:25K epch:14.56 loss:0.086 grdn:0.234 lr:2.5e-06 updt_s: -0.512 data_s:0.003 -INFO 2025-09-08 22:51:14 celerate.py:281 step:62K smpl:4M ep:25K epch:14.60 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.514 data_s:0.003 -INFO 2025-09-08 22:52:58 celerate.py:281 step:63K smpl:4M ep:25K epch:14.65 loss:0.087 grdn:0.245 lr:2.5e-06 updt_s: -0.484 data_s:0.033 -INFO 2025-09-08 22:54:41 celerate.py:281 step:63K smpl:4M ep:25K epch:14.70 loss:0.086 grdn:0.239 lr:2.5e-06 updt_s: -0.501 data_s:0.016 -INFO 2025-09-08 22:56:25 celerate.py:281 step:63K smpl:4M ep:25K epch:14.74 loss:0.086 grdn:0.233 lr:2.5e-06 updt_s: -0.436 data_s:0.081 -INFO 2025-09-08 22:58:08 celerate.py:281 step:63K smpl:4M ep:25K epch:14.79 loss:0.087 grdn:0.235 lr:2.5e-06 updt_s: -0.436 data_s:0.080 -INFO 2025-09-08 22:59:51 celerate.py:281 step:63K smpl:4M ep:25K epch:14.84 loss:0.086 grdn:0.234 lr:2.5e-06 updt_s: -0.344 data_s:0.168 -INFO 2025-09-08 23:01:34 celerate.py:281 step:64K smpl:4M ep:25K epch:14.88 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.479 data_s:0.035 -INFO 2025-09-08 23:03:18 celerate.py:281 step:64K smpl:4M ep:25K epch:14.93 loss:0.086 grdn:0.241 lr:2.5e-06 updt_s: -0.514 data_s:0.003 -INFO 2025-09-08 23:05:02 celerate.py:281 step:64K smpl:4M ep:25K epch:14.98 loss:0.086 grdn:0.230 lr:2.5e-06 updt_s: -0.513 data_s:0.003 -INFO 2025-09-08 23:06:47 celerate.py:281 step:64K smpl:4M ep:25K epch:15.02 loss:0.086 grdn:0.234 lr:2.5e-06 updt_s: -0.405 data_s:0.119 -INFO 2025-09-08 23:08:30 celerate.py:281 step:64K smpl:4M ep:26K epch:15.07 loss:0.087 grdn:0.248 lr:2.5e-06 updt_s: -0.393 data_s:0.121 -INFO 2025-09-08 23:10:13 celerate.py:281 step:65K smpl:4M ep:26K epch:15.12 loss:0.087 grdn:0.242 lr:2.5e-06 updt_s: -0.369 data_s:0.142 -INFO 2025-09-08 23:11:56 celerate.py:281 step:65K smpl:4M ep:26K epch:15.17 loss:0.085 grdn:0.230 lr:2.5e-06 updt_s: -0.360 data_s:0.156 -INFO 2025-09-08 23:13:40 celerate.py:281 step:65K smpl:4M ep:26K epch:15.21 loss:0.086 grdn:0.241 lr:2.5e-06 updt_s: -0.333 data_s:0.182 -INFO 2025-09-08 23:15:23 celerate.py:281 step:65K smpl:4M ep:26K epch:15.26 loss:0.087 grdn:0.241 lr:2.5e-06 updt_s: -0.376 data_s:0.136 -INFO 2025-09-08 23:17:05 celerate.py:281 step:65K smpl:4M ep:26K epch:15.31 loss:0.087 grdn:0.239 lr:2.5e-06 updt_s: -0.439 data_s:0.069 -INFO 2025-09-08 23:18:49 celerate.py:281 step:66K smpl:4M ep:26K epch:15.35 loss:0.086 grdn:0.242 lr:2.5e-06 updt_s: -0.512 data_s:0.006 -INFO 2025-09-08 23:20:32 celerate.py:281 step:66K smpl:4M ep:26K epch:15.40 loss:0.087 grdn:0.242 lr:2.5e-06 updt_s: -0.511 data_s:0.003 -INFO 2025-09-08 23:22:15 celerate.py:281 step:66K smpl:4M ep:26K epch:15.45 loss:0.086 grdn:0.236 lr:2.5e-06 updt_s: -0.508 data_s:0.004 -INFO 2025-09-08 23:23:59 celerate.py:281 step:66K smpl:4M ep:26K epch:15.49 loss:0.086 grdn:0.235 lr:2.5e-06 updt_s: -0.474 data_s:0.046 -INFO 2025-09-08 23:25:42 celerate.py:281 step:66K smpl:4M ep:26K epch:15.54 loss:0.087 grdn:0.238 lr:2.5e-06 updt_s: -0.510 data_s:0.003 -INFO 2025-09-08 23:27:25 celerate.py:281 step:67K smpl:4M ep:26K epch:15.59 loss:0.086 grdn:0.240 lr:2.5e-06 updt_s: -0.487 data_s:0.025 -INFO 2025-09-08 23:29:08 celerate.py:281 step:67K smpl:4M ep:26K epch:15.63 loss:0.086 grdn:0.241 lr:2.5e-06 updt_s: -0.510 data_s:0.005 -INFO 2025-09-08 23:30:50 celerate.py:281 step:67K smpl:4M ep:27K epch:15.68 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.506 data_s:0.003 -INFO 2025-09-08 23:32:33 celerate.py:281 step:67K smpl:4M ep:27K epch:15.73 loss:0.086 grdn:0.245 lr:2.5e-06 updt_s: -0.509 data_s:0.003 -INFO 2025-09-08 23:34:16 celerate.py:281 step:67K smpl:4M ep:27K epch:15.77 loss:0.087 grdn:0.244 lr:2.5e-06 updt_s: -0.503 data_s:0.014 -INFO 2025-09-08 23:35:59 celerate.py:281 step:68K smpl:4M ep:27K epch:15.82 loss:0.086 grdn:0.236 lr:2.5e-06 updt_s: -0.512 data_s:0.003 -INFO 2025-09-08 23:37:42 celerate.py:281 step:68K smpl:4M ep:27K epch:15.87 loss:0.086 grdn:0.233 lr:2.5e-06 updt_s: -0.509 data_s:0.003 -INFO 2025-09-08 23:39:26 celerate.py:281 step:68K smpl:4M ep:27K epch:15.91 loss:0.085 grdn:0.235 lr:2.5e-06 updt_s: -0.512 data_s:0.003 -INFO 2025-09-08 23:41:10 celerate.py:281 step:68K smpl:4M ep:27K epch:15.96 loss:0.086 grdn:0.241 lr:2.5e-06 updt_s: -0.517 data_s:0.003 -INFO 2025-09-08 23:42:56 celerate.py:281 step:68K smpl:4M ep:27K epch:16.01 loss:0.086 grdn:0.242 lr:2.5e-06 updt_s: -0.469 data_s:0.061 -INFO 2025-09-08 23:44:40 celerate.py:281 step:69K smpl:4M ep:27K epch:16.05 loss:0.086 grdn:0.235 lr:2.5e-06 updt_s: -0.339 data_s:0.179 -INFO 2025-09-08 23:46:22 celerate.py:281 step:69K smpl:4M ep:27K epch:16.10 loss:0.086 grdn:0.242 lr:2.5e-06 updt_s: -0.386 data_s:0.124 -INFO 2025-09-08 23:48:06 celerate.py:281 step:69K smpl:4M ep:27K epch:16.15 loss:0.085 grdn:0.239 lr:2.5e-06 updt_s: -0.515 data_s:0.003 -INFO 2025-09-08 23:49:50 celerate.py:281 step:69K smpl:4M ep:27K epch:16.20 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.516 data_s:0.003 -INFO 2025-09-08 23:51:33 celerate.py:281 step:69K smpl:4M ep:27K epch:16.24 loss:0.087 grdn:0.246 lr:2.5e-06 updt_s: -0.501 data_s:0.014 -INFO 2025-09-08 23:53:17 celerate.py:281 step:70K smpl:4M ep:28K epch:16.29 loss:0.086 grdn:0.236 lr:2.5e-06 updt_s: -0.513 data_s:0.003 -INFO 2025-09-08 23:54:59 celerate.py:281 step:70K smpl:4M ep:28K epch:16.34 loss:0.087 grdn:0.245 lr:2.5e-06 updt_s: -0.507 data_s:0.005 -INFO 2025-09-08 23:56:43 celerate.py:281 step:70K smpl:4M ep:28K epch:16.38 loss:0.086 grdn:0.238 lr:2.5e-06 updt_s: -0.516 data_s:0.003 -INFO 2025-09-08 23:58:27 celerate.py:281 step:70K smpl:4M ep:28K epch:16.43 loss:0.086 grdn:0.235 lr:2.5e-06 updt_s: -0.514 data_s:0.003 -INFO 2025-09-09 00:00:12 celerate.py:281 step:70K smpl:5M ep:28K epch:16.48 loss:0.086 grdn:0.240 lr:2.5e-06 updt_s: -0.518 data_s:0.003 -INFO 2025-09-09 00:01:55 celerate.py:281 step:71K smpl:5M ep:28K epch:16.52 loss:0.086 grdn:0.241 lr:2.5e-06 updt_s: -0.512 data_s:0.003 -INFO 2025-09-09 00:03:37 celerate.py:281 step:71K smpl:5M ep:28K epch:16.57 loss:0.087 grdn:0.241 lr:2.5e-06 updt_s: -0.508 data_s:0.003 -INFO 2025-09-09 00:05:20 celerate.py:281 step:71K smpl:5M ep:28K epch:16.62 loss:0.086 grdn:0.238 lr:2.5e-06 updt_s: -0.509 data_s:0.003 -INFO 2025-09-09 00:07:04 celerate.py:281 step:71K smpl:5M ep:28K epch:16.66 loss:0.086 grdn:0.235 lr:2.5e-06 updt_s: -0.519 data_s:0.003 -INFO 2025-09-09 00:08:47 celerate.py:281 step:71K smpl:5M ep:28K epch:16.71 loss:0.087 grdn:0.240 lr:2.5e-06 updt_s: -0.509 data_s:0.003 -INFO 2025-09-09 00:10:30 celerate.py:281 step:72K smpl:5M ep:28K epch:16.76 loss:0.087 grdn:0.245 lr:2.5e-06 updt_s: -0.511 data_s:0.003 -INFO 2025-09-09 00:12:13 celerate.py:281 step:72K smpl:5M ep:28K epch:16.80 loss:0.085 grdn:0.230 lr:2.5e-06 updt_s: -0.510 data_s:0.003 -INFO 2025-09-09 00:13:58 celerate.py:281 step:72K smpl:5M ep:29K epch:16.85 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.522 data_s:0.003 -INFO 2025-09-09 00:15:40 celerate.py:281 step:72K smpl:5M ep:29K epch:16.90 loss:0.086 grdn:0.239 lr:2.5e-06 updt_s: -0.506 data_s:0.003 -INFO 2025-09-09 00:17:23 celerate.py:281 step:72K smpl:5M ep:29K epch:16.94 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.510 data_s:0.004 -INFO 2025-09-09 00:19:05 celerate.py:281 step:73K smpl:5M ep:29K epch:16.99 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.505 data_s:0.003 -INFO 2025-09-09 00:20:51 celerate.py:281 step:73K smpl:5M ep:29K epch:17.04 loss:0.086 grdn:0.239 lr:2.5e-06 updt_s: -0.437 data_s:0.092 -INFO 2025-09-09 00:22:34 celerate.py:281 step:73K smpl:5M ep:29K epch:17.08 loss:0.087 grdn:0.241 lr:2.5e-06 updt_s: -0.489 data_s:0.025 -INFO 2025-09-09 00:24:17 celerate.py:281 step:73K smpl:5M ep:29K epch:17.13 loss:0.086 grdn:0.243 lr:2.5e-06 updt_s: -0.511 data_s:0.003 -INFO 2025-09-09 00:26:01 celerate.py:281 step:73K smpl:5M ep:29K epch:17.18 loss:0.087 grdn:0.245 lr:2.5e-06 updt_s: -0.513 data_s:0.003 -INFO 2025-09-09 00:27:45 celerate.py:281 step:74K smpl:5M ep:29K epch:17.22 loss:0.087 grdn:0.246 lr:2.5e-06 updt_s: -0.515 data_s:0.003 -INFO 2025-09-09 00:29:28 celerate.py:281 step:74K smpl:5M ep:29K epch:17.27 loss:0.086 grdn:0.239 lr:2.5e-06 updt_s: -0.509 data_s:0.003 -INFO 2025-09-09 00:31:11 celerate.py:281 step:74K smpl:5M ep:29K epch:17.32 loss:0.086 grdn:0.238 lr:2.5e-06 updt_s: -0.512 data_s:0.003 -INFO 2025-09-09 00:32:54 celerate.py:281 step:74K smpl:5M ep:29K epch:17.37 loss:0.087 grdn:0.237 lr:2.5e-06 updt_s: -0.510 data_s:0.003 -INFO 2025-09-09 00:34:38 celerate.py:281 step:74K smpl:5M ep:29K epch:17.41 loss:0.086 grdn:0.240 lr:2.5e-06 updt_s: -0.515 data_s:0.003 -INFO 2025-09-09 00:36:21 celerate.py:281 step:75K smpl:5M ep:30K epch:17.46 loss:0.086 grdn:0.241 lr:2.5e-06 updt_s: -0.513 data_s:0.003 -INFO 2025-09-09 00:38:05 celerate.py:281 step:75K smpl:5M ep:30K epch:17.51 loss:0.086 grdn:0.244 lr:2.5e-06 updt_s: -0.512 data_s:0.003 -INFO 2025-09-09 00:39:49 celerate.py:281 step:75K smpl:5M ep:30K epch:17.55 loss:0.086 grdn:0.251 lr:2.5e-06 updt_s: -0.517 data_s:0.003 -INFO 2025-09-09 00:41:32 celerate.py:281 step:75K smpl:5M ep:30K epch:17.60 loss:0.086 grdn:0.243 lr:2.5e-06 updt_s: -0.506 data_s:0.005 -INFO 2025-09-09 00:43:14 celerate.py:281 step:75K smpl:5M ep:30K epch:17.65 loss:0.086 grdn:0.240 lr:2.5e-06 updt_s: -0.390 data_s:0.122 -INFO 2025-09-09 00:44:58 celerate.py:281 step:76K smpl:5M ep:30K epch:17.69 loss:0.086 grdn:0.232 lr:2.5e-06 updt_s: -0.410 data_s:0.107 -INFO 2025-09-09 00:46:41 celerate.py:281 step:76K smpl:5M ep:30K epch:17.74 loss:0.087 grdn:0.247 lr:2.5e-06 updt_s: -0.427 data_s:0.085 -INFO 2025-09-09 00:48:24 celerate.py:281 step:76K smpl:5M ep:30K epch:17.79 loss:0.085 grdn:0.241 lr:2.5e-06 updt_s: -0.492 data_s:0.025 -INFO 2025-09-09 00:50:08 celerate.py:281 step:76K smpl:5M ep:30K epch:17.83 loss:0.086 grdn:0.238 lr:2.5e-06 updt_s: -0.514 data_s:0.003 -INFO 2025-09-09 00:51:51 celerate.py:281 step:76K smpl:5M ep:30K epch:17.88 loss:0.087 grdn:0.241 lr:2.5e-06 updt_s: -0.510 data_s:0.003 -INFO 2025-09-09 00:53:33 celerate.py:281 step:77K smpl:5M ep:30K epch:17.93 loss:0.086 grdn:0.234 lr:2.5e-06 updt_s: -0.510 data_s:0.003 -INFO 2025-09-09 00:55:18 celerate.py:281 step:77K smpl:5M ep:30K epch:17.97 loss:0.086 grdn:0.240 lr:2.5e-06 updt_s: -0.519 data_s:0.003 -INFO 2025-09-09 00:57:05 celerate.py:281 step:77K smpl:5M ep:31K epch:18.02 loss:0.086 grdn:0.236 lr:2.5e-06 updt_s: -0.486 data_s:0.046 -INFO 2025-09-09 00:58:47 celerate.py:281 step:77K smpl:5M ep:31K epch:18.07 loss:0.086 grdn:0.239 lr:2.5e-06 updt_s: -0.509 data_s:0.003 -INFO 2025-09-09 01:00:31 celerate.py:281 step:77K smpl:5M ep:31K epch:18.11 loss:0.086 grdn:0.244 lr:2.5e-06 updt_s: -0.514 data_s:0.003 -INFO 2025-09-09 01:02:15 celerate.py:281 step:78K smpl:5M ep:31K epch:18.16 loss:0.086 grdn:0.241 lr:2.5e-06 updt_s: -0.515 data_s:0.003 -INFO 2025-09-09 01:03:56 celerate.py:281 step:78K smpl:5M ep:31K epch:18.21 loss:0.086 grdn:0.233 lr:2.5e-06 updt_s: -0.498 data_s:0.008 -INFO 2025-09-09 01:05:40 celerate.py:281 step:78K smpl:5M ep:31K epch:18.25 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.508 data_s:0.010 -INFO 2025-09-09 01:07:24 celerate.py:281 step:78K smpl:5M ep:31K epch:18.30 loss:0.085 grdn:0.240 lr:2.5e-06 updt_s: -0.515 data_s:0.003 -INFO 2025-09-09 01:09:07 celerate.py:281 step:78K smpl:5M ep:31K epch:18.35 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.513 data_s:0.003 -INFO 2025-09-09 01:10:51 celerate.py:281 step:79K smpl:5M ep:31K epch:18.40 loss:0.086 grdn:0.236 lr:2.5e-06 updt_s: -0.515 data_s:0.003 -INFO 2025-09-09 01:12:35 celerate.py:281 step:79K smpl:5M ep:31K epch:18.44 loss:0.086 grdn:0.239 lr:2.5e-06 updt_s: -0.514 data_s:0.003 -INFO 2025-09-09 01:14:17 celerate.py:281 step:79K smpl:5M ep:31K epch:18.49 loss:0.086 grdn:0.247 lr:2.5e-06 updt_s: -0.508 data_s:0.003 -INFO 2025-09-09 01:16:01 celerate.py:281 step:79K smpl:5M ep:31K epch:18.54 loss:0.086 grdn:0.244 lr:2.5e-06 updt_s: -0.513 data_s:0.003 -INFO 2025-09-09 01:17:43 celerate.py:281 step:79K smpl:5M ep:31K epch:18.58 loss:0.086 grdn:0.233 lr:2.5e-06 updt_s: -0.507 data_s:0.003 -INFO 2025-09-09 01:19:26 celerate.py:281 step:80K smpl:5M ep:32K epch:18.63 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.511 data_s:0.003 -INFO 2025-09-09 01:21:09 celerate.py:281 step:80K smpl:5M ep:32K epch:18.68 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.513 data_s:0.003 -INFO 2025-09-09 01:22:52 celerate.py:281 step:80K smpl:5M ep:32K epch:18.72 loss:0.086 grdn:0.240 lr:2.5e-06 updt_s: -0.501 data_s:0.011 -INFO 2025-09-09 01:22:52 celerate.py:295 Checkpoint policy after step 80000 -INFO 2025-09-09 01:24:37 celerate.py:281 step:80K smpl:5M ep:32K epch:18.77 loss:0.086 grdn:0.241 lr:2.5e-06 updt_s: -0.510 data_s:0.003 -INFO 2025-09-09 01:26:20 celerate.py:281 step:80K smpl:5M ep:32K epch:18.82 loss:0.087 grdn:0.242 lr:2.5e-06 updt_s: -0.514 data_s:0.003 -INFO 2025-09-09 01:28:05 celerate.py:281 step:81K smpl:5M ep:32K epch:18.86 loss:0.086 grdn:0.241 lr:2.5e-06 updt_s: -0.517 data_s:0.003 -INFO 2025-09-09 01:29:48 celerate.py:281 step:81K smpl:5M ep:32K epch:18.91 loss:0.086 grdn:0.243 lr:2.5e-06 updt_s: -0.511 data_s:0.003 -INFO 2025-09-09 01:31:32 celerate.py:281 step:81K smpl:5M ep:32K epch:18.96 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.517 data_s:0.003 -INFO 2025-09-09 01:33:18 celerate.py:281 step:81K smpl:5M ep:32K epch:19.00 loss:0.087 grdn:0.244 lr:2.5e-06 updt_s: -0.487 data_s:0.042 -INFO 2025-09-09 01:35:01 celerate.py:281 step:81K smpl:5M ep:32K epch:19.05 loss:0.086 grdn:0.236 lr:2.5e-06 updt_s: -0.461 data_s:0.054 -INFO 2025-09-09 01:36:46 celerate.py:281 step:82K smpl:5M ep:32K epch:19.10 loss:0.086 grdn:0.239 lr:2.5e-06 updt_s: -0.450 data_s:0.071 -INFO 2025-09-09 01:38:29 celerate.py:281 step:82K smpl:5M ep:32K epch:19.14 loss:0.086 grdn:0.241 lr:2.5e-06 updt_s: -0.451 data_s:0.064 -INFO 2025-09-09 01:40:12 celerate.py:281 step:82K smpl:5M ep:32K epch:19.19 loss:0.086 grdn:0.234 lr:2.5e-06 updt_s: -0.512 data_s:0.003 -INFO 2025-09-09 01:41:54 celerate.py:281 step:82K smpl:5M ep:33K epch:19.24 loss:0.085 grdn:0.237 lr:2.5e-06 updt_s: -0.480 data_s:0.030 -INFO 2025-09-09 01:43:37 celerate.py:281 step:82K smpl:5M ep:33K epch:19.28 loss:0.086 grdn:0.242 lr:2.5e-06 updt_s: -0.434 data_s:0.080 -INFO 2025-09-09 01:45:20 celerate.py:281 step:83K smpl:5M ep:33K epch:19.33 loss:0.086 grdn:0.243 lr:2.5e-06 updt_s: -0.511 data_s:0.003 -INFO 2025-09-09 01:47:04 celerate.py:281 step:83K smpl:5M ep:33K epch:19.38 loss:0.086 grdn:0.243 lr:2.5e-06 updt_s: -0.511 data_s:0.003 -INFO 2025-09-09 01:48:49 celerate.py:281 step:83K smpl:5M ep:33K epch:19.42 loss:0.086 grdn:0.243 lr:2.5e-06 updt_s: -0.520 data_s:0.003 -INFO 2025-09-09 01:50:32 celerate.py:281 step:83K smpl:5M ep:33K epch:19.47 loss:0.086 grdn:0.238 lr:2.5e-06 updt_s: -0.511 data_s:0.003 -INFO 2025-09-09 01:52:15 celerate.py:281 step:83K smpl:5M ep:33K epch:19.52 loss:0.086 grdn:0.243 lr:2.5e-06 updt_s: -0.513 data_s:0.003 -INFO 2025-09-09 01:53:59 celerate.py:281 step:84K smpl:5M ep:33K epch:19.57 loss:0.085 grdn:0.236 lr:2.5e-06 updt_s: -0.513 data_s:0.003 -INFO 2025-09-09 01:55:41 celerate.py:281 step:84K smpl:5M ep:33K epch:19.61 loss:0.086 grdn:0.238 lr:2.5e-06 updt_s: -0.508 data_s:0.003 -INFO 2025-09-09 01:57:24 celerate.py:281 step:84K smpl:5M ep:33K epch:19.66 loss:0.087 grdn:0.237 lr:2.5e-06 updt_s: -0.500 data_s:0.013 -INFO 2025-09-09 01:59:07 celerate.py:281 step:84K smpl:5M ep:33K epch:19.71 loss:0.087 grdn:0.245 lr:2.5e-06 updt_s: -0.510 data_s:0.003 -INFO 2025-09-09 02:00:50 celerate.py:281 step:84K smpl:5M ep:33K epch:19.75 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.512 data_s:0.003 -INFO 2025-09-09 02:02:34 celerate.py:281 step:85K smpl:5M ep:34K epch:19.80 loss:0.086 grdn:0.240 lr:2.5e-06 updt_s: -0.516 data_s:0.003 -INFO 2025-09-09 02:04:17 celerate.py:281 step:85K smpl:5M ep:34K epch:19.85 loss:0.086 grdn:0.244 lr:2.5e-06 updt_s: -0.511 data_s:0.003 -INFO 2025-09-09 02:05:59 celerate.py:281 step:85K smpl:5M ep:34K epch:19.89 loss:0.086 grdn:0.243 lr:2.5e-06 updt_s: -0.507 data_s:0.004 -INFO 2025-09-09 02:07:43 celerate.py:281 step:85K smpl:5M ep:34K epch:19.94 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.505 data_s:0.011 -INFO 2025-09-09 02:09:26 celerate.py:281 step:85K smpl:5M ep:34K epch:19.99 loss:0.086 grdn:0.242 lr:2.5e-06 updt_s: -0.392 data_s:0.124 -INFO 2025-09-09 02:11:14 celerate.py:281 step:86K smpl:5M ep:34K epch:20.03 loss:0.086 grdn:0.240 lr:2.5e-06 updt_s: -0.435 data_s:0.100 -INFO 2025-09-09 02:12:57 celerate.py:281 step:86K smpl:5M ep:34K epch:20.08 loss:0.086 grdn:0.242 lr:2.5e-06 updt_s: -0.514 data_s:0.003 -INFO 2025-09-09 02:14:41 celerate.py:281 step:86K smpl:6M ep:34K epch:20.13 loss:0.085 grdn:0.234 lr:2.5e-06 updt_s: -0.516 data_s:0.003 -INFO 2025-09-09 02:16:25 celerate.py:281 step:86K smpl:6M ep:34K epch:20.17 loss:0.085 grdn:0.234 lr:2.5e-06 updt_s: -0.514 data_s:0.003 -INFO 2025-09-09 02:18:09 celerate.py:281 step:86K smpl:6M ep:34K epch:20.22 loss:0.085 grdn:0.238 lr:2.5e-06 updt_s: -0.506 data_s:0.010 -INFO 2025-09-09 02:19:53 celerate.py:281 step:87K smpl:6M ep:34K epch:20.27 loss:0.086 grdn:0.238 lr:2.5e-06 updt_s: -0.515 data_s:0.003 -INFO 2025-09-09 02:21:35 celerate.py:281 step:87K smpl:6M ep:34K epch:20.31 loss:0.086 grdn:0.243 lr:2.5e-06 updt_s: -0.501 data_s:0.007 -INFO 2025-09-09 02:23:17 celerate.py:281 step:87K smpl:6M ep:34K epch:20.36 loss:0.085 grdn:0.242 lr:2.5e-06 updt_s: -0.432 data_s:0.080 -INFO 2025-09-09 02:25:01 celerate.py:281 step:87K smpl:6M ep:35K epch:20.41 loss:0.086 grdn:0.235 lr:2.5e-06 updt_s: -0.445 data_s:0.073 -INFO 2025-09-09 02:26:43 celerate.py:281 step:87K smpl:6M ep:35K epch:20.45 loss:0.085 grdn:0.239 lr:2.5e-06 updt_s: -0.401 data_s:0.107 -INFO 2025-09-09 02:28:26 celerate.py:281 step:88K smpl:6M ep:35K epch:20.50 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.502 data_s:0.009 -INFO 2025-09-09 02:30:09 celerate.py:281 step:88K smpl:6M ep:35K epch:20.55 loss:0.087 grdn:0.248 lr:2.5e-06 updt_s: -0.491 data_s:0.021 -INFO 2025-09-09 02:31:52 celerate.py:281 step:88K smpl:6M ep:35K epch:20.59 loss:0.086 grdn:0.243 lr:2.5e-06 updt_s: -0.391 data_s:0.124 -INFO 2025-09-09 02:33:35 celerate.py:281 step:88K smpl:6M ep:35K epch:20.64 loss:0.086 grdn:0.238 lr:2.5e-06 updt_s: -0.332 data_s:0.180 -INFO 2025-09-09 02:35:18 celerate.py:281 step:88K smpl:6M ep:35K epch:20.69 loss:0.087 grdn:0.243 lr:2.5e-06 updt_s: -0.333 data_s:0.181 -INFO 2025-09-09 02:37:02 celerate.py:281 step:89K smpl:6M ep:35K epch:20.74 loss:0.086 grdn:0.245 lr:2.5e-06 updt_s: -0.332 data_s:0.185 -INFO 2025-09-09 02:38:47 celerate.py:281 step:89K smpl:6M ep:35K epch:20.78 loss:0.086 grdn:0.241 lr:2.5e-06 updt_s: -0.332 data_s:0.190 -INFO 2025-09-09 02:40:30 celerate.py:281 step:89K smpl:6M ep:35K epch:20.83 loss:0.086 grdn:0.240 lr:2.5e-06 updt_s: -0.330 data_s:0.187 -INFO 2025-09-09 02:42:12 celerate.py:281 step:89K smpl:6M ep:35K epch:20.88 loss:0.085 grdn:0.243 lr:2.5e-06 updt_s: -0.331 data_s:0.177 -INFO 2025-09-09 02:43:56 celerate.py:281 step:89K smpl:6M ep:35K epch:20.92 loss:0.086 grdn:0.244 lr:2.5e-06 updt_s: -0.330 data_s:0.185 -INFO 2025-09-09 02:45:39 celerate.py:281 step:90K smpl:6M ep:36K epch:20.97 loss:0.087 grdn:0.252 lr:2.5e-06 updt_s: -0.335 data_s:0.182 -INFO 2025-09-09 02:47:25 celerate.py:281 step:90K smpl:6M ep:36K epch:21.02 loss:0.087 grdn:0.247 lr:2.5e-06 updt_s: -0.330 data_s:0.197 -INFO 2025-09-09 02:49:08 celerate.py:281 step:90K smpl:6M ep:36K epch:21.06 loss:0.086 grdn:0.240 lr:2.5e-06 updt_s: -0.330 data_s:0.182 -INFO 2025-09-09 02:50:51 celerate.py:281 step:90K smpl:6M ep:36K epch:21.11 loss:0.086 grdn:0.243 lr:2.5e-06 updt_s: -0.334 data_s:0.181 -INFO 2025-09-09 02:52:34 celerate.py:281 step:90K smpl:6M ep:36K epch:21.16 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.330 data_s:0.183 -INFO 2025-09-09 02:54:17 celerate.py:281 step:91K smpl:6M ep:36K epch:21.20 loss:0.085 grdn:0.235 lr:2.5e-06 updt_s: -0.342 data_s:0.173 -INFO 2025-09-09 02:56:01 celerate.py:281 step:91K smpl:6M ep:36K epch:21.25 loss:0.086 grdn:0.246 lr:2.5e-06 updt_s: -0.331 data_s:0.189 -INFO 2025-09-09 02:57:44 celerate.py:281 step:91K smpl:6M ep:36K epch:21.30 loss:0.086 grdn:0.236 lr:2.5e-06 updt_s: -0.329 data_s:0.182 -INFO 2025-09-09 02:59:27 celerate.py:281 step:91K smpl:6M ep:36K epch:21.34 loss:0.087 grdn:0.246 lr:2.5e-06 updt_s: -0.341 data_s:0.175 -INFO 2025-09-09 03:01:11 celerate.py:281 step:91K smpl:6M ep:36K epch:21.39 loss:0.086 grdn:0.242 lr:2.5e-06 updt_s: -0.330 data_s:0.188 -INFO 2025-09-09 03:02:54 celerate.py:281 step:92K smpl:6M ep:36K epch:21.44 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.328 data_s:0.186 -INFO 2025-09-09 03:04:39 celerate.py:281 step:92K smpl:6M ep:36K epch:21.48 loss:0.086 grdn:0.244 lr:2.5e-06 updt_s: -0.329 data_s:0.195 -INFO 2025-09-09 03:06:22 celerate.py:281 step:92K smpl:6M ep:36K epch:21.53 loss:0.086 grdn:0.244 lr:2.5e-06 updt_s: -0.330 data_s:0.183 -INFO 2025-09-09 03:08:04 celerate.py:281 step:92K smpl:6M ep:37K epch:21.58 loss:0.086 grdn:0.241 lr:2.5e-06 updt_s: -0.462 data_s:0.051 -INFO 2025-09-09 03:09:48 celerate.py:281 step:92K smpl:6M ep:37K epch:21.62 loss:0.086 grdn:0.248 lr:2.5e-06 updt_s: -0.407 data_s:0.108 -INFO 2025-09-09 03:11:32 celerate.py:281 step:93K smpl:6M ep:37K epch:21.67 loss:0.086 grdn:0.232 lr:2.5e-06 updt_s: -0.333 data_s:0.185 -INFO 2025-09-09 03:13:15 celerate.py:281 step:93K smpl:6M ep:37K epch:21.72 loss:0.085 grdn:0.242 lr:2.5e-06 updt_s: -0.329 data_s:0.187 -INFO 2025-09-09 03:14:58 celerate.py:281 step:93K smpl:6M ep:37K epch:21.77 loss:0.086 grdn:0.240 lr:2.5e-06 updt_s: -0.357 data_s:0.156 -INFO 2025-09-09 03:16:41 celerate.py:281 step:93K smpl:6M ep:37K epch:21.81 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.487 data_s:0.027 -INFO 2025-09-09 03:18:25 celerate.py:281 step:93K smpl:6M ep:37K epch:21.86 loss:0.086 grdn:0.240 lr:2.5e-06 updt_s: -0.512 data_s:0.003 -INFO 2025-09-09 03:20:08 celerate.py:281 step:94K smpl:6M ep:37K epch:21.91 loss:0.087 grdn:0.247 lr:2.5e-06 updt_s: -0.512 data_s:0.003 -INFO 2025-09-09 03:21:51 celerate.py:281 step:94K smpl:6M ep:37K epch:21.95 loss:0.086 grdn:0.238 lr:2.5e-06 updt_s: -0.508 data_s:0.004 -INFO 2025-09-09 03:23:38 celerate.py:281 step:94K smpl:6M ep:37K epch:22.00 loss:0.085 grdn:0.239 lr:2.5e-06 updt_s: -0.429 data_s:0.104 -INFO 2025-09-09 03:25:20 celerate.py:281 step:94K smpl:6M ep:37K epch:22.05 loss:0.086 grdn:0.243 lr:2.5e-06 updt_s: -0.328 data_s:0.183 -INFO 2025-09-09 03:27:04 celerate.py:281 step:94K smpl:6M ep:37K epch:22.09 loss:0.086 grdn:0.246 lr:2.5e-06 updt_s: -0.329 data_s:0.191 -INFO 2025-09-09 03:28:47 celerate.py:281 step:95K smpl:6M ep:37K epch:22.14 loss:0.086 grdn:0.246 lr:2.5e-06 updt_s: -0.329 data_s:0.186 -INFO 2025-09-09 03:30:30 celerate.py:281 step:95K smpl:6M ep:38K epch:22.19 loss:0.086 grdn:0.241 lr:2.5e-06 updt_s: -0.346 data_s:0.166 -INFO 2025-09-09 03:32:13 celerate.py:281 step:95K smpl:6M ep:38K epch:22.23 loss:0.086 grdn:0.247 lr:2.5e-06 updt_s: -0.334 data_s:0.181 -INFO 2025-09-09 03:33:56 celerate.py:281 step:95K smpl:6M ep:38K epch:22.28 loss:0.086 grdn:0.244 lr:2.5e-06 updt_s: -0.402 data_s:0.110 -INFO 2025-09-09 03:35:39 celerate.py:281 step:95K smpl:6M ep:38K epch:22.33 loss:0.085 grdn:0.237 lr:2.5e-06 updt_s: -0.353 data_s:0.161 -INFO 2025-09-09 03:37:22 celerate.py:281 step:96K smpl:6M ep:38K epch:22.37 loss:0.086 grdn:0.235 lr:2.5e-06 updt_s: -0.356 data_s:0.158 -INFO 2025-09-09 03:39:04 celerate.py:281 step:96K smpl:6M ep:38K epch:22.42 loss:0.086 grdn:0.244 lr:2.5e-06 updt_s: -0.379 data_s:0.131 -INFO 2025-09-09 03:40:49 celerate.py:281 step:96K smpl:6M ep:38K epch:22.47 loss:0.085 grdn:0.239 lr:2.5e-06 updt_s: -0.344 data_s:0.175 -INFO 2025-09-09 03:42:32 celerate.py:281 step:96K smpl:6M ep:38K epch:22.51 loss:0.086 grdn:0.236 lr:2.5e-06 updt_s: -0.331 data_s:0.185 -INFO 2025-09-09 03:44:15 celerate.py:281 step:96K smpl:6M ep:38K epch:22.56 loss:0.086 grdn:0.244 lr:2.5e-06 updt_s: -0.331 data_s:0.183 -INFO 2025-09-09 03:45:58 celerate.py:281 step:97K smpl:6M ep:38K epch:22.61 loss:0.086 grdn:0.238 lr:2.5e-06 updt_s: -0.330 data_s:0.184 -INFO 2025-09-09 03:47:43 celerate.py:281 step:97K smpl:6M ep:38K epch:22.65 loss:0.086 grdn:0.248 lr:2.5e-06 updt_s: -0.331 data_s:0.188 -INFO 2025-09-09 03:49:27 celerate.py:281 step:97K smpl:6M ep:38K epch:22.70 loss:0.086 grdn:0.240 lr:2.5e-06 updt_s: -0.333 data_s:0.185 -INFO 2025-09-09 03:51:10 celerate.py:281 step:97K smpl:6M ep:39K epch:22.75 loss:0.085 grdn:0.241 lr:2.5e-06 updt_s: -0.330 data_s:0.185 -INFO 2025-09-09 03:52:54 celerate.py:281 step:97K smpl:6M ep:39K epch:22.79 loss:0.086 grdn:0.247 lr:2.5e-06 updt_s: -0.330 data_s:0.192 -INFO 2025-09-09 03:54:37 celerate.py:281 step:98K smpl:6M ep:39K epch:22.84 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.329 data_s:0.185 -INFO 2025-09-09 03:56:21 celerate.py:281 step:98K smpl:6M ep:39K epch:22.89 loss:0.086 grdn:0.237 lr:2.5e-06 updt_s: -0.329 data_s:0.187 -INFO 2025-09-09 03:58:04 celerate.py:281 step:98K smpl:6M ep:39K epch:22.94 loss:0.086 grdn:0.242 lr:2.5e-06 updt_s: -0.329 data_s:0.185 -INFO 2025-09-09 03:59:46 celerate.py:281 step:98K smpl:6M ep:39K epch:22.98 loss:0.086 grdn:0.239 lr:2.5e-06 updt_s: -0.329 data_s:0.183 -INFO 2025-09-09 04:01:32 celerate.py:281 step:98K smpl:6M ep:39K epch:23.03 loss:0.087 grdn:0.250 lr:2.5e-06 updt_s: -0.376 data_s:0.151 -INFO 2025-09-09 04:03:16 celerate.py:281 step:99K smpl:6M ep:39K epch:23.08 loss:0.086 grdn:0.241 lr:2.5e-06 updt_s: -0.329 data_s:0.187 -INFO 2025-09-09 04:04:59 celerate.py:281 step:99K smpl:6M ep:39K epch:23.12 loss:0.086 grdn:0.243 lr:2.5e-06 updt_s: -0.379 data_s:0.136 -INFO 2025-09-09 04:06:42 celerate.py:281 step:99K smpl:6M ep:39K epch:23.17 loss:0.086 grdn:0.240 lr:2.5e-06 updt_s: -0.513 data_s:0.003 -INFO 2025-09-09 04:08:25 celerate.py:281 step:99K smpl:6M ep:39K epch:23.22 loss:0.086 grdn:0.242 lr:2.5e-06 updt_s: -0.510 data_s:0.003 -INFO 2025-09-09 04:10:07 celerate.py:281 step:99K smpl:6M ep:39K epch:23.26 loss:0.087 grdn:0.242 lr:2.5e-06 updt_s: -0.470 data_s:0.039 -INFO 2025-09-09 04:11:50 celerate.py:281 step:100K smpl:6M ep:39K epch:23.31 loss:0.086 grdn:0.247 lr:2.5e-06 updt_s -:0.347 data_s:0.169 -INFO 2025-09-09 04:13:34 celerate.py:281 step:100K smpl:6M ep:40K epch:23.36 loss:0.085 grdn:0.237 lr:2.5e-06 updt_s -:0.330 data_s:0.185 -INFO 2025-09-09 04:15:17 celerate.py:281 step:100K smpl:6M ep:40K epch:23.40 loss:0.086 grdn:0.239 lr:2.5e-06 updt_s -:0.356 data_s:0.156 -INFO 2025-09-09 04:15:17 celerate.py:295 Checkpoint policy after step 100000 -INFO 2025-09-09 04:15:18 celerate.py:359 End of training -(lerobot) jade_choghari@hf-dgx-01:~/lerobot$ clear -(lerobot) jade_choghari@hf-dgx-01:~/lerobot$ tmux capture-pane -pS - > tmux_log.txt - - - - - - - - - -