mirror of
https://github.com/huggingface/lerobot.git
synced 2026-07-16 06:21:48 +00:00
Pepijn
4dfa8cea65
Port the LingBot-VA policy (Wan2.2 dual-stream video+action world model) into LeRobot, following the EO-1 / VLA-JEPA conventions. Covers inference, checkpoint conversion, and predicted-video saving (training is deferred to a follow-up PR). - Vendored Wan transformer/attention/flex/VAE/scheduler modules (key names preserved for near-identity conversion); torch SDPA default, flashattn/flex lazy-guarded. - LingBotVAConfig (registered "lingbot_va") + processor with fixed-quantile action unnormalization; full dual-stream sampling loop with CFG, two flow-matching schedulers and KV cache, mapped onto select_action with observed-keyframe feedback. - convert_lingbot_va_checkpoints.py (libero/robotwin variants): bundles the ~5B transformer, lazy-pulls the frozen VAE+UMT5 from the source repo. - Predicted-video plumbing in lerobot_eval (predicted_frames_callback; opt-in via --policy.save_predicted_video) and ConstantWithWarmupSchedulerConfig. - pyproject: widen diffusers-dep to <0.37, add lingbot_va + imageio-dep extras, add lingbot_va and (missing) eo1 to `all`. - Factory + policies/__init__ wiring, docs page + toctree, and tests. Note: the LIBERO success-rate correctness gate must be validated on a CUDA GPU with the converted checkpoint. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
70 lines
2.5 KiB
Python
70 lines
2.5 KiB
Python
#!/usr/bin/env python
|
|
|
|
# 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.
|
|
|
|
"""Pure-torch unit tests for the vendored LingBot-VA helper modules (no diffusers needed)."""
|
|
|
|
from __future__ import annotations
|
|
|
|
import torch
|
|
|
|
from lerobot.policies.lingbot_va.schedulers import FlowMatchScheduler
|
|
from lerobot.policies.lingbot_va.wan_utils import data_seq_to_patch, get_mesh_id
|
|
|
|
|
|
def test_flow_match_scheduler_timesteps_monotone_decreasing() -> None:
|
|
sch = FlowMatchScheduler(shift=5.0, sigma_min=0.0, extra_one_step=True)
|
|
sch.set_timesteps(20)
|
|
assert sch.timesteps.shape == (20,)
|
|
diffs = sch.timesteps[1:] - sch.timesteps[:-1]
|
|
assert torch.all(diffs <= 0) # decreasing
|
|
|
|
|
|
def test_flow_match_scheduler_step_preserves_shape() -> None:
|
|
sch = FlowMatchScheduler(shift=5.0, sigma_min=0.0, extra_one_step=True)
|
|
sch.set_timesteps(20)
|
|
sample = torch.zeros(1, 48, 4, 8, 16)
|
|
out = sch.step(torch.ones_like(sample), sch.timesteps[0], sample)
|
|
assert out.shape == sample.shape
|
|
|
|
|
|
def test_flow_match_scheduler_add_noise() -> None:
|
|
sch = FlowMatchScheduler(shift=5.0, sigma_min=0.0, extra_one_step=True)
|
|
sch.set_timesteps(20)
|
|
sample = torch.randn(1, 48, 4, 8, 16)
|
|
noise = torch.randn_like(sample)
|
|
noisy = sch.add_noise(sample, noise, sch.timesteps[:4], t_dim=2)
|
|
assert noisy.shape == sample.shape
|
|
|
|
|
|
def test_get_mesh_id_latent_shape() -> None:
|
|
grid = get_mesh_id(4, 8, 16, 0, 1, 0)
|
|
assert grid.shape == (4, 4 * 8 * 16) # (f, h, w, stream) x tokens
|
|
|
|
|
|
def test_get_mesh_id_action_shape() -> None:
|
|
grid = get_mesh_id(4, 4, 1, 1, 1, 0, action=True)
|
|
assert grid.shape == (4, 4 * 4 * 1)
|
|
# Action rows for h/w are sentinel -1.
|
|
assert torch.all(grid[1] < 0)
|
|
assert torch.all(grid[2] < 0)
|
|
|
|
|
|
def test_data_seq_to_patch_roundtrip_shape() -> None:
|
|
b, f, h, w, c = 1, 4, 8, 16, 48
|
|
seq = torch.arange(b * f * h * w * c, dtype=torch.float32).reshape(b, f * h * w, c)
|
|
out = data_seq_to_patch((1, 2, 2), seq, f, h, w, batch_size=b)
|
|
assert out.shape == (b, c, f, h, w)
|