admin/lerobot
1
0
Fork 0
mirror of https://github.com/huggingface/lerobot.git synced 2026-06-18 16:57:12 +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
**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). |
| 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. |
| 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. |
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
@@ -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:
| Variant | LeRobot checkpoint |
|---|---|
| Variant | LeRobot checkpoint |
| ---------------------- | ---------------------------------- |
| LIBERO-Long post-train | `pepijn223/lingbot_va_libero_long` |
| RoboTwin post-train | `pepijn223/lingbot_va_robotwin` |
| Pretrained base | `pepijn223/lingbot_va_base` |
| RoboTwin post-train | `pepijn223/lingbot_va_robotwin` |
| Pretrained base | `pepijn223/lingbot_va_base` |
**Packaging:** only the trainable ~5B transformer is stored in the LeRobot
`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).
Requirements:
- 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).
- 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)
| Key | Value |
|---|---|
| height × width | 128 × 128 |
| cameras | `observation.images.image` (agentview), `observation.images.image2` (eye-in-hand) |
| action channels used | 0–6 (7-DoF arm + gripper) |
| action_per_frame / frame_chunk_size | 4 / 4 |
| attn_window | 30 |
| video / action denoising steps | 20 / 50 |
| guidance_scale / action_guidance_scale | 5 / 1 |
| snr_shift / action_snr_shift | 5.0 / 0.05 |
| Key | Value |
| -------------------------------------- | --------------------------------------------------------------------------------- |
| height × width | 128 × 128 |
| cameras | `observation.images.image` (agentview), `observation.images.image2` (eye-in-hand) |
| action channels used | 0–6 (7-DoF arm + gripper) |
| action_per_frame / frame_chunk_size | 4 / 4 |
| attn_window | 30 |
| video / action denoising steps | 20 / 50 |
| guidance_scale / action_guidance_scale | 5 / 1 |
| snr_shift / action_snr_shift | 5.0 / 0.05 |
These are the defaults of `LingBotVAConfig`; override any of them via `--policy.<name>=...`.