admin/lerobot
1
0
Fork 0
mirror of https://github.com/huggingface/lerobot.git synced 2026-07-05 09:07:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

Update default targets

Browse Source 
Removed ACT since it doesn't make sense to fine-tune ACT without having it pretrained beforehand.
SmolVLA and Pi0/0.5 are much more senseful targets.
This commit is contained in:
nemo
2025-10-16 16:13:41 +02:00
parent 7f42338ae9
commit de34e876cc
1 changed files with 16 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/lerobot/scripts/lerobot_train.py
+16 -12
View File
@@ -138,22 +138,26 @@ def get_default_peft_configuration(policy_type):
"""
if policy_type == "smolvla":
return {
"target_modules": r"(model\.vlm_with_expert\.lm_expert\..*\.(q_proj|v_proj)|model\.action_.*|model\.state_proj.*)",
"target_modules": r"(model\.vlm_with_expert\.lm_expert\..*\.(q_proj|v_proj))",
"modules_to_save": [
# These are inf on load otherwise
"normalize_inputs",
"normalize_targets",
"unnormalize_outputs",
# these are initialized randomly and need full-finetuning
"state_proj",
"action_in_proj",
"action_out_proj",
"action_time_mlp_in",
"action_time_mlp_out",
],
}
elif policy_type == "act":
elif policy_type in ("pi0", "pi05"):
return {
"target_modules": r"(.*_proj|.*\.action_head)",
"target_modules": r".*\.gemma_expert\..*\.self_attn.(q_proj|v_proj)",
"modules_to_save": [
# These are inf on load otherwise
"normalize_inputs",
"normalize_targets",
"unnormalize_outputs",
# these are initialized randomly and need full-finetuning
"state_proj",
"action_in_proj",
"action_out_proj",
"action_time_mlp_in",
"action_time_mlp_out",
],
}
@@ -170,7 +174,7 @@ def wrap_policy_in_peft_model(cfg, policy):
peft_config_policy = get_default_peft_configuration(cfg.policy.type)
peft_config_cli = dataclasses.asdict(cfg.peft) if cfg.peft else {}
peft_config_cli['modules_to_save'] = peft_config_cli['full_training_modules'] # compatibility with PEFT
peft_config_cli["modules_to_save"] = peft_config_cli["full_training_modules"] # compatibility with PEFT
peft_method_type = PeftType[peft_config_cli["method_type"].upper()]
peft_config_cls = PEFT_TYPE_TO_CONFIG_MAPPING[peft_method_type]