Model: lerobot/act_aloha_sim_transfer_cube_human
Route: demonstration-policy (Action Chunking Transformer)
Hardware: Huawei Ascend 910B (Ascend910_9362) on npu:0
Revision: main (local dir /opt/atomgit/adapt-npu-agent/persist_models/act_aloha_sim_transfer_cube_human)
Weight: model.safetensors 198MB, config act chunk_size 100
AlohaTransferCube-v0 (gym-aloha)。observation.images.top : RGB [3,480,640] 单相机 top 视角observation.state : 14-dim 双臂 proprio (joint positions)action chunk [batch, chunk_size=100, action_dim=14] 连续关节位置序列,n_action_steps=100lerobot.policies.act.ACTPolicy.from_pretrained + torch_npu, patched to avoid online ResNet downloadtorch.npu.is_available()=True, device_count=2, device_name=Ascend910_9362npu:0, backbone npu:0, 输入 Tensor npu:0, 输出 action npu:0torch.npu.synchronize()[1,100,14] count 1400 finite true3.85e-04 mean_abs 1.02e-04 tolerance 1.46e-03 (atol 1e-4 rtol 1e-3) PASS7.28e-02 also PASSactions/s 7903 chunks/s 79.03scripts/benchmark_policy.py --latencies-ms /tmp/latencies_ms.json --items-per-call 100 --unit actionspip install -r requirements.txt
python inference.pyinference.py 默认从本地 MODEL_DIR 加载,打印 model/revision/route/backend/device/dtype、输入摘要、CPU 与 NPU action 摘要、同步耗时与 PASS/FAIL。无需网络下载权重。
inference.py
readme.md
requirements.txt
assets/agent_workflow.png
assets/npu_device_call.png
assets/model_result.pngnpu:0 验证,显存峰值约 1.2GB (权重 198MB + 模型 51M params)assets/agent_workflow.png 侦察、资源、加载、NPU 推理、一致性、性能与校验assets/npu_device_call.png npu-smi、NPU availability、设备名、模型/输入/输出 deviceassets/model_result.png 默认 python inference.py 的真实 ActionPlan 与同步耗时
