admin/lerobot
1
0
Fork 0
mirror of https://github.com/huggingface/lerobot.git synced 2026-05-26 05:59:52 +00:00
Code Issues Packages Projects Releases Wiki Activity

Store policy config alongside PEFT checkpoint

Browse Source 
Before this change the PEFT-wrapped policy did not save the policy's config
alongside the adapter config / weights which prevented us from changing the
policy config. Now the policy config is saved both in full training and PEFT
training.

This change makes loading the PEFT policy adapter much easier as well.
This commit is contained in:
nemo
2025-06-22 19:46:10 +02:00
parent 7fd8b4c773
commit dc67b2ff3f
3 changed files with 27 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files
lerobot/common/utils/train_utils.py
+4
View File
@@ -99,6 +99,10 @@ def save_checkpoint(
pretrained_dir = checkpoint_dir / PRETRAINED_MODEL_DIR
policy.save_pretrained(pretrained_dir)
cfg.save_pretrained(pretrained_dir)
if cfg.use_peft:
# When using PEFT, policy.save_pretrained will only write the adapter weights + config, not the
# policy config which we need for loading the model. In this case we'll write it ourselves.
policy.config.save_pretrained(pretrained_dir)
save_training_state(checkpoint_dir, step, optimizer, scheduler)