admin/lerobot
1
0
Fork 0
mirror of https://github.com/huggingface/lerobot.git synced 2026-05-17 17:50:09 +00:00
Code Issues Packages Projects Releases Wiki Activity

Convert tensors to scalars (#2903)

Browse Source 
Co-authored-by: Steven Palma <imstevenpmwork@ieee.org>
This commit is contained in:
Reece O'Mahoney
2026-02-09 13:46:12 +00:00
committed by GitHub
parent 97e7e0f9ed
commit e14bdf57d0
1 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/lerobot/policies/smolvla/modeling_smolvla.py
+3 -3
View File
@@ -378,16 +378,16 @@ class SmolVLAPolicy(PreTrainedPolicy):
actions_is_pad = batch.get("actions_id_pad") actions_is_pad = batch.get("actions_id_pad")
loss_dict = {} loss_dict = {}
losses = self.model.forward(images, img_masks, lang_tokens, lang_masks, state, actions, noise, time) losses = self.model.forward(images, img_masks, lang_tokens, lang_masks, state, actions, noise, time)
loss_dict["losses_after_forward"] = losses.clone() loss_dict["losses_after_forward"] = losses.clone().mean().item()
if actions_is_pad is not None: if actions_is_pad is not None:
in_episode_bound = ~actions_is_pad in_episode_bound = ~actions_is_pad
losses = losses * in_episode_bound.unsqueeze(-1) losses = losses * in_episode_bound.unsqueeze(-1)
loss_dict["losses_after_in_ep_bound"] = losses.clone() loss_dict["losses_after_in_ep_bound"] = losses.clone().mean().item()
# Remove padding # Remove padding
losses = losses[:, :, : self.config.max_action_dim] losses = losses[:, :, : self.config.max_action_dim]
loss_dict["losses_after_rm_padding"] = losses.clone() loss_dict["losses_after_rm_padding"] = losses.clone().mean().item()
if reduction == "none": if reduction == "none":
# Return per-sample losses (B,) by averaging over time and action dims # Return per-sample losses (B,) by averaging over time and action dims