admin/lerobot
1
0
Fork 0
mirror of https://github.com/huggingface/lerobot.git synced 2026-07-24 10:16:09 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix dinov3

Browse Source
This commit is contained in:
Pepijn
2025-08-31 19:44:27 +02:00
parent e1d433cbfc
commit 79c3466f0f
1 changed files with 1 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/lerobot/policies/rlearn/modeling_rlearn.py
-2
View File
@@ -377,8 +377,6 @@ class RLearNPolicy(PreTrainedPolicy):
inputs = {k: v.to(device) for k, v in inputs.items()} inputs = {k: v.to(device) for k, v in inputs.items()}
# Process in batch through DINOv3 model # Process in batch through DINOv3 model
# Use inference mode for stable, fast frozen encoder forward
with torch.inference_mode():
vision_outputs = self.vision_model(**inputs) vision_outputs = self.vision_model(**inputs)
# Prefer mean-pooled patch tokens over pooler/CLS to ensure input-dependent variation # Prefer mean-pooled patch tokens over pooler/CLS to ensure input-dependent variation