admin/lerobot
1
0
Fork 0
mirror of https://github.com/huggingface/lerobot.git synced 2026-07-23 01:41:54 +00:00
Code Issues Packages Projects Releases Wiki Activity

huberman loss

Browse Source
This commit is contained in:
Pepijn
2025-09-01 11:59:20 +02:00
parent 4c4462edea
commit f84c20d403
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/lerobot/policies/rlearn/modeling_rlearn.py
+2 -2
View File
@@ -522,8 +522,8 @@ class RLearNPolicy(PreTrainedPolicy):
if mismatch_tensor.any(): if mismatch_tensor.any():
eps = self.config.logit_eps eps = self.config.logit_eps
zeros_target_logits = torch.logit(torch.full_like(mismatch_raw_logits, eps)) zeros_target_logits = torch.logit(torch.full_like(mismatch_raw_logits, eps))
mismatch_loss_per_sample = F.mse_loss( mismatch_loss_per_sample = F.smooth_l1_loss(
mismatch_raw_logits, zeros_target_logits, reduction='none' mismatch_raw_logits, zeros_target_logits, beta=0.5, reduction='none'
).mean(dim=1) ).mean(dim=1)
L_mismatch = mismatch_loss_per_sample[mismatch_tensor].mean() L_mismatch = mismatch_loss_per_sample[mismatch_tensor].mean()