impact-libero-gguf
IMPACT as vla.simd GGUF for CPU inference. Each file is self-contained.
| File | Converted from |
|---|---|
impact-libero-spatial.gguf | khanhnd61/impact_libero_spatial |
impact-libero-object.gguf | khanhnd61/impact_libero_object |
impact-libero-goal.gguf | khanhnd61/impact_libero_goal |
impact-libero-10.gguf | khanhnd61/impact_libero_10 |
Serve
Install vla.simd with the serve extra, then:
hf download khanhnd61/impact-libero-gguf --local-dir impact-libero-gguf
OMP_NUM_THREADS=$CORES vla-simd-serve --model impact \
--model-dir impact-libero-gguf/impact-libero-spatial.gguf --port 8080 \
--task "<instruction>"
Any file in the table works as --model-dir.
Rollout
The robot side is lerobot's async client, lerobot-vla-simd:
pip install 'lerobot[async,feetech] @ git+https://github.com/khanhnd61-vr/lerobot@4b33b84296c0880ebce778d69f16a38d33825575'
The client has no simulator mode. Replaying LIBERO frames checks that the server is up and returns sane actions next to the recorded ones; it is open-loop, not a success-rate evaluation.
lerobot-vla-simd --server_address=127.0.0.1:8080 --policy_type=impact \
--replay.repo_id=HuggingFaceVLA/libero --replay.episode=0