admin/lerobot
1
0
Fork 0
mirror of https://github.com/huggingface/lerobot.git synced 2026-07-24 02:06:15 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix(lingbot_va): CI quality gate + fast-test collection

Browse Source 
- Add tests/policies/lingbot_va/__init__.py so the test files don't clash by basename
  with tests/policies/vla_jepa/* under pytest's default import mode (fast-test collection error).
- Fix vendored typos flagged by the typos hook (pach_scale->patch_scale, total_tolen->
  total_token_len, stablized->stabilized) and a mypy union-attr in RoboTwinEnv._read_eef_pose.
- Apply Prettier formatting to docs/source/lingbot_va.mdx.

Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
pepijn223
2026-06-06 15:46:37 +02:00
parent 71aacda05e
commit f9d12db9cf
4 changed files with 48 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/source/lingbot_va.mdx
+19 -18
View File
@@ -14,11 +14,11 @@ LingBot-VA is a **dual-stream "mixture-of-transformers"**: a video/latent stream
text conditioning. Actions are produced by the dedicated `action_proj_out` head — they are text conditioning. Actions are produced by the dedicated `action_proj_out` head — they are
**not** decoded from predicted pixels, though video and action are co-trained. **not** decoded from predicted pixels, though video and action are co-trained.
| Component | Class | Role | | Component | Class | Role |
|---|---|---| | ------------------------ | ----------------------- | -------------------------------------------------------------------------------------- |
| DiT backbone (trainable) | `WanTransformer3DModel` | ~5B-param dual-stream transformer (the only weights stored in the LeRobot checkpoint). | | DiT backbone (trainable) | `WanTransformer3DModel` | ~5B-param dual-stream transformer (the only weights stored in the LeRobot checkpoint). |
| VAE (frozen) | `AutoencoderKLWan` | Wan2.2 VAE, `z_dim=48`. Lazy-pulled from the source repo. | | VAE (frozen) | `AutoencoderKLWan` | Wan2.2 VAE, `z_dim=48`. Lazy-pulled from the source repo. |
| Text encoder (frozen) | `UMT5EncoderModel` | UMT5-XXL, `d_model=4096`. Lazy-pulled from the source repo. | | Text encoder (frozen) | `UMT5EncoderModel` | UMT5-XXL, `d_model=4096`. Lazy-pulled from the source repo. |
At inference the policy runs an autoregressive loop per chunk: it denoises the video-latent At inference the policy runs an autoregressive loop per chunk: it denoises the video-latent
stream (CFG, ~20 steps) and the action stream (~50 steps) with two independent stream (CFG, ~20 steps) and the action stream (~50 steps) with two independent
@@ -50,11 +50,11 @@ pip install -e ".[lingbot_va,libero]"
The released upstream checkpoints have been converted to LeRobot format and pushed to the Hub: The released upstream checkpoints have been converted to LeRobot format and pushed to the Hub:
| Variant | LeRobot checkpoint | | Variant | LeRobot checkpoint |
|---|---| | ---------------------- | ---------------------------------- |
| LIBERO-Long post-train | `pepijn223/lingbot_va_libero_long` | | LIBERO-Long post-train | `pepijn223/lingbot_va_libero_long` |
| RoboTwin post-train | `pepijn223/lingbot_va_robotwin` | | RoboTwin post-train | `pepijn223/lingbot_va_robotwin` |
| Pretrained base | `pepijn223/lingbot_va_base` | | Pretrained base | `pepijn223/lingbot_va_base` |
**Packaging:** only the trainable ~5B transformer is stored in the LeRobot **Packaging:** only the trainable ~5B transformer is stored in the LeRobot
`model.safetensors`. The frozen VAE + UMT5 + tokenizer (~20 GB) are **lazily pulled** from `model.safetensors`. The frozen VAE + UMT5 + tokenizer (~20 GB) are **lazily pulled** from
@@ -112,6 +112,7 @@ transformer's block-causal training pass and returns `(loss, metrics)`. Optimize
with a linear-warmup-then-constant schedule (matching upstream). with a linear-warmup-then-constant schedule (matching upstream).
Requirements: Requirements:
- The block-causal masks use PyTorch **flex-attention**, so build the policy with - The block-causal masks use PyTorch **flex-attention**, so build the policy with
`--policy.attn_mode=flex` for training (the default `torch` SDPA is inference-only). `--policy.attn_mode=flex` for training (the default `torch` SDPA is inference-only).
- The full 5B DiT does not fit a single 24–32 GB GPU under AdamW; fine-tune with **LoRA** - The full 5B DiT does not fit a single 24–32 GB GPU under AdamW; fine-tune with **LoRA**
@@ -131,16 +132,16 @@ The dataset must provide camera clips (a temporal window per camera, VAE-encoded
## Inference Hyperparameters (LIBERO) ## Inference Hyperparameters (LIBERO)
| Key | Value | | Key | Value |
|---|---| | -------------------------------------- | --------------------------------------------------------------------------------- |
| height × width | 128 × 128 | | height × width | 128 × 128 |
| cameras | `observation.images.image` (agentview), `observation.images.image2` (eye-in-hand) | | cameras | `observation.images.image` (agentview), `observation.images.image2` (eye-in-hand) |
| action channels used | 0–6 (7-DoF arm + gripper) | | action channels used | 0–6 (7-DoF arm + gripper) |
| action_per_frame / frame_chunk_size | 4 / 4 | | action_per_frame / frame_chunk_size | 4 / 4 |
| attn_window | 30 | | attn_window | 30 |
| video / action denoising steps | 20 / 50 | | video / action denoising steps | 20 / 50 |
| guidance_scale / action_guidance_scale | 5 / 1 | | guidance_scale / action_guidance_scale | 5 / 1 |
| snr_shift / action_snr_shift | 5.0 / 0.05 | | snr_shift / action_snr_shift | 5.0 / 0.05 |
These are the defaults of `LingBotVAConfig`; override any of them via `--policy.<name>=...`. These are the defaults of `LingBotVAConfig`; override any of them via `--policy.<name>=...`.
src/lerobot/envs/robotwin.py
+1
View File
@@ -359,6 +359,7 @@ class RoboTwinEnv(gym.Env):
def _read_eef_pose(self) -> np.ndarray: def _read_eef_pose(self) -> np.ndarray:
"""Read the current 16-d dual-arm eef pose [left(xyz+quat)+grip, right(xyz+quat)+grip].""" """Read the current 16-d dual-arm eef pose [left(xyz+quat)+grip, right(xyz+quat)+grip]."""
assert self._env is not None, "_read_eef_pose called before _ensure_env()"
ep = self._env.get_obs()["endpose"] ep = self._env.get_obs()["endpose"]
pose = ( pose = (
list(ep["left_endpose"]) list(ep["left_endpose"])
src/lerobot/policies/lingbot_va/modeling_lingbot_va.py
+15 -15
View File
@@ -185,12 +185,12 @@ class FlowMatchScheduler:
prev_sample = sample + model_output * (sigma_ - sigma) prev_sample = sample + model_output * (sigma_ - sigma)
return prev_sample return prev_sample
def return_to_timestep(self, timestep, sample, sample_stablized): def return_to_timestep(self, timestep, sample, sample_stabilized):
if isinstance(timestep, torch.Tensor): if isinstance(timestep, torch.Tensor):
timestep = timestep.cpu() timestep = timestep.cpu()
timestep_id = torch.argmin((self.timesteps - timestep).abs()) timestep_id = torch.argmin((self.timesteps - timestep).abs())
sigma = self.sigmas[timestep_id] sigma = self.sigmas[timestep_id]
model_output = (sample - sample_stablized) / sigma model_output = (sample - sample_stabilized) / sigma
return model_output return model_output
def add_noise(self, original_samples, noise, timestep, t_dim=2): def add_noise(self, original_samples, noise, timestep, t_dim=2):
@@ -518,15 +518,15 @@ class WanAttention(nn.Module):
return return
self.attn_caches[cache_name] = None self.attn_caches[cache_name] = None
def init_kv_cache(self, cache_name, total_tolen, num_head, head_dim, device, dtype, batch_size): def init_kv_cache(self, cache_name, total_token_len, num_head, head_dim, device, dtype, batch_size):
if self.attn_caches is None: if self.attn_caches is None:
return return
self.attn_caches[cache_name] = { self.attn_caches[cache_name] = {
"k": torch.empty([batch_size, total_tolen, num_head, head_dim], device=device, dtype=dtype), "k": torch.empty([batch_size, total_token_len, num_head, head_dim], device=device, dtype=dtype),
"v": torch.empty([batch_size, total_tolen, num_head, head_dim], device=device, dtype=dtype), "v": torch.empty([batch_size, total_token_len, num_head, head_dim], device=device, dtype=dtype),
"id": torch.full((total_tolen,), -1, device=device), "id": torch.full((total_token_len,), -1, device=device),
"mask": torch.zeros((total_tolen,), dtype=torch.bool, device=device), "mask": torch.zeros((total_token_len,), dtype=torch.bool, device=device),
"is_pred": torch.zeros((total_tolen,), dtype=torch.bool, device=device), "is_pred": torch.zeros((total_token_len,), dtype=torch.bool, device=device),
} }
def allocate_slots(self, cache_name, key_size): def allocate_slots(self, cache_name, key_size):
@@ -831,13 +831,13 @@ class WanTransformer3DModel(ModelMixin, ConfigMixin):
dtype, dtype,
batch_size, batch_size,
): ):
total_tolen = (attn_window // 2) * latent_token_per_chunk + ( total_token_len = (attn_window // 2) * latent_token_per_chunk + (
attn_window // 2 attn_window // 2
) * action_token_per_chunk ) * action_token_per_chunk
for block in self.blocks: for block in self.blocks:
block.attn1.init_kv_cache( block.attn1.init_kv_cache(
cache_name, cache_name,
total_tolen, total_token_len,
self.num_attention_heads, self.num_attention_heads,
self.attention_head_dim, self.attention_head_dim,
device, device,
@@ -866,9 +866,9 @@ class WanTransformer3DModel(ModelMixin, ConfigMixin):
return hidden_states return hidden_states
def _time_embed(self, timesteps, H, W, dtype, action_mode=False): def _time_embed(self, timesteps, H, W, dtype, action_mode=False):
pach_scale_h, pach_scale_w = (1, 1) if action_mode else (self.patch_size[1], self.patch_size[2]) patch_scale_h, patch_scale_w = (1, 1) if action_mode else (self.patch_size[1], self.patch_size[2])
latent_time_steps = torch.repeat_interleave( latent_time_steps = torch.repeat_interleave(
timesteps, (H // pach_scale_h) * (W // pach_scale_w), dim=1 timesteps, (H // patch_scale_h) * (W // patch_scale_w), dim=1
) )
current_condition_embedder = ( current_condition_embedder = (
self.condition_embedder_action if action_mode else self.condition_embedder self.condition_embedder_action if action_mode else self.condition_embedder
@@ -1012,12 +1012,12 @@ class WanTransformer3DModel(ModelMixin, ConfigMixin):
latent_grid_id = input_dict["grid_id"] latent_grid_id = input_dict["grid_id"]
rotary_emb = self.rope(latent_grid_id)[:, :, None] # 1 L 1 C rotary_emb = self.rope(latent_grid_id)[:, :, None] # 1 L 1 C
pach_scale_h, pach_scale_w = (1, 1) if action_mode else (self.patch_size[1], self.patch_size[2]) patch_scale_h, patch_scale_w = (1, 1) if action_mode else (self.patch_size[1], self.patch_size[2])
latent_time_steps = torch.repeat_interleave( latent_time_steps = torch.repeat_interleave(
input_dict["timesteps"], input_dict["timesteps"],
(input_dict["noisy_latents"].shape[-2] // pach_scale_h) (input_dict["noisy_latents"].shape[-2] // patch_scale_h)
* (input_dict["noisy_latents"].shape[-1] // pach_scale_w), * (input_dict["noisy_latents"].shape[-1] // patch_scale_w),
dim=1, dim=1,
) # L ) # L
current_condition_embedder = ( current_condition_embedder = (
tests/policies/lingbot_va/__init__.py
+13
View File
@@ -0,0 +1,13 @@
# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.