admin/lerobot
1
0
Fork 0
mirror of https://github.com/huggingface/lerobot.git synced 2026-06-18 00:37:10 +00:00
Code Issues Packages Projects Releases Wiki Activity

cleanup

Browse Source
This commit is contained in:
Maxime Ellerbach
2026-06-15 14:50:23 +00:00
parent 3ce50c3468
commit b42d124007
2 changed files with 3 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/source/multi_gpu_training.mdx
+3 -3
View File
@@ -138,10 +138,10 @@ num_machines: 1
num_processes: 4
fsdp_config:
fsdp_version: 1
fsdp_sharding_strategy: FULL_SHARD # params + grads + optimizer (ZeRO-3)
fsdp_sharding_strategy: FULL_SHARD # params + grads + optimizer (ZeRO-3)
fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
fsdp_transformer_layer_cls_to_wrap: <YourTransformerBlock> # repeated block class to shard
fsdp_use_orig_params: true # required: optimizer is built pre-prepare
fsdp_transformer_layer_cls_to_wrap: <YourTransformerBlock> # repeated block class to shard
fsdp_use_orig_params: true # required: optimizer is built pre-prepare
fsdp_state_dict_type: FULL_STATE_DICT
```
src/lerobot/scripts/lerobot_train.py
-1
View File
@@ -198,7 +198,6 @@ def train(cfg: TrainPipelineConfig, accelerator: "Accelerator | None" = None):
# We set step_scheduler_with_optimizer=False to prevent accelerate from adjusting the lr_scheduler steps based on the num_processes
# We set find_unused_parameters=True to handle models with conditional computation
if accelerator is None:
ddp_kwargs = DistributedDataParallelKwargs(find_unused_parameters=True)
# Accelerate auto-detects the device based on the available hardware and ignores the policy.device setting.
# Force the device to be CPU when the active config's device is set to CPU (works for both policy and reward model training).