act-so101-multi-task-gguf

ACT as vla.simd GGUF for CPU inference. Each file is self-contained.

File	Converted from
act-so101-multi-task.gguf	khanhnd61/act-matched_so101-multi-task-clean
act-so101-multi-task-tape-into-box.gguf	khanhnd61/act-matched_so101-multi-task-clean_tape-into-box
act-so101-multi-task-cup-into-box.gguf	khanhnd61/act-matched_so101-multi-task-clean_cup-into-box
act-so101-multi-task-tape-into-cup.gguf	khanhnd61/act-matched_so101-multi-task-clean_tape-into-cup

Serve

Install vla.simd with the serve extra, then:

hf download khanhnd61/act-so101-multi-task-gguf --local-dir act-so101-multi-task-gguf
OMP_NUM_THREADS=$CORES vla-simd-serve --model act \
    --model-dir act-so101-multi-task-gguf/act-so101-multi-task.gguf --port 8080

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'

On the robot, with the arm and both cameras attached:

lerobot-vla-simd --server_address=127.0.0.1:8080 --policy_type=act \
    --robot.type=so101_follower --robot.port=/dev/ttyACM0 --robot.id=my_arm \
    --robot.cameras="{ front: {type: opencv, index_or_path: 0, width: 640, height: 480, fps: 30}, wrist: {type: opencv, index_or_path: 2, width: 640, height: 480, fps: 30} }" \
    --actions_per_chunk=50