chore: Merge origin/main into streaming branch

README.md

@@ -83,7 +83,7 @@ episode_index=0
 print(f"{dataset[episode_index]['action'].shape=}\n")
 ```
 
-Learn more about it in the [LeRobotDataset Documentation](https://huggingface.co/docs/lerobot/lerobot-dataset-v3)
+Learn more about it in the [LeRobotDataset Documentation](https://huggingface.co/docs/lerobot/lerobot-dataset-v3).
 
 ## SoTA Models
 
@@ -109,7 +109,7 @@ lerobot-train \
 | **World Models**           | [VLA-JEPA](./docs/source/vla_jepa.mdx), [LingBot-VA](./docs/source/lingbot_va.mdx), [FastWAM](./docs/source/fastwam.mdx)                                                                                                                                                                                                                                                                   |
 | **Reward Models**          | [SARM](./docs/source/sarm.mdx), [TOPReward](./docs/source/topreward.mdx), [Robometer](./docs/source/robometer.mdx)                                                                                                                                                                                                                                                                         |
 
-Similarly to the hardware, you can easily implement your own policy & leverage LeRobot's data collection, training, and visualization tools, and share your model to the HF Hub
+Similarly to the hardware, you can easily implement your own policy & leverage LeRobot's data collection, training, and visualization tools, and share your model to the HF Hub.
 
 For detailed policy setup guides, see the [Policy Documentation](https://huggingface.co/docs/lerobot/bring_your_own_policies). For GPU/RAM requirements and expected training time per policy, see the [Compute Hardware Guide](https://huggingface.co/docs/lerobot/hardware_guide).
 
@@ -126,7 +126,7 @@ lerobot-eval \
   --eval.n_episodes=10
 ```
 
-Learn how to implement your own simulation environment or benchmark and distribute it from the HF Hub by following the [EnvHub Documentation](https://huggingface.co/docs/lerobot/envhub)
+Learn how to implement your own simulation environment or benchmark and distribute it from the HF Hub by following the [EnvHub Documentation](https://huggingface.co/docs/lerobot/envhub).
 
 ## Resources
 

docs/source/omx.mdx

@@ -1,3 +1,11 @@
+# OMX
+
+<img
+  src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/lerobot/omx_mainimage.png"
+  alt="OMX"
+  width=600
+/>
+
 ## Order and Assemble the parts
 
 First, assemble the OMX hardware following the official assembly guide.

src/lerobot/policies/pi05/modeling_pi05.py

@@ -524,8 +524,6 @@ class PI05Pytorch(nn.Module):  # see openpi `PI0Pytorch`
 
     def embed_suffix(self, noisy_actions, timestep):
         """Embed noisy_actions, timestep to prepare for Expert Gemma processing."""
-        embs = []
-        pad_masks = []
         att_masks = []
 
         # Embed timestep using sine-cosine positional encoding
@@ -551,23 +549,17 @@ class PI05Pytorch(nn.Module):  # see openpi `PI0Pytorch`
             return F.silu(x)
 
         time_emb = self._apply_checkpoint(time_mlp_func, time_emb)
-        action_time_emb = action_emb
         adarms_cond = time_emb
 
-        embs.append(action_time_emb)
+        bsize, action_time_dim = action_emb.shape[:2]
-        bsize, action_time_dim = action_time_emb.shape[:2]
+        pad_masks = torch.ones(bsize, action_time_dim, dtype=torch.bool, device=timestep.device)
-        action_time_mask = torch.ones(bsize, action_time_dim, dtype=torch.bool, device=timestep.device)
-        pad_masks.append(action_time_mask)
 
         # Set attention masks so that image, language and state inputs do not attend to action tokens
         att_masks += [1] + ([0] * (self.config.chunk_size - 1))
-
+        att_masks = torch.tensor(att_masks, dtype=action_emb.dtype, device=action_emb.device)
-        embs = torch.cat(embs, dim=1)
-        pad_masks = torch.cat(pad_masks, dim=1)
-        att_masks = torch.tensor(att_masks, dtype=embs.dtype, device=embs.device)
         att_masks = att_masks[None, :].expand(bsize, len(att_masks))
 
-        return embs, pad_masks, att_masks, adarms_cond
+        return action_emb, pad_masks, att_masks, adarms_cond
 
     def forward(self, images, img_masks, tokens, masks, actions, noise, time) -> Tensor:
         """Do a full training forward pass and compute the loss."""