MolmoAct2 BimanualYAM dynamic engines for Jetson AGX Thor

This is the dynamic-prompt TensorRT build of allenai/MolmoAct2-BimanualYAM for vla-edge. Use it when an instruction may exceed the fixed 704-token bracket of the faster champion build. It accepts instructions up to the engine profile's 1024-token bound.

Download and verify

This repository is about 11.2 GB.

hf download agents2agents/MolmoAct2-BimanualYAM-Dynamic-Jetson-Thor \
  --local-dir vla-edge-yam-dynamic
cd vla-edge-yam-dynamic
python -c "from vla_edge.backends.tensorrt import artifacts; \
artifacts.check_compatible('.'); artifacts.verify_checksums('.'); \
print('bundle verified')"

Run

vla-edge-serve --embodiment bimanual-yam --backend tensorrt \
  --engine-dir /path/to/vla-edge-yam-dynamic/yam

The server reads the prompt policy from yam/serving.json. No manual padding flag is needed.

Hardware and contents

These plans require an NVIDIA Jetson AGX Thor Developer Kit with JetPack R39 rev 2.1 and TensorRT 10.16.2.10.

yam/                dynamic BimanualYAM TensorRT engines
host/yam/           processor, normalization, embeddings, and flow weights
MANIFEST.json       compatibility requirements and checksums

Serving this bundle is local and does not download the upstream checkpoint.

License

Apache-2.0. See LICENSE and NOTICE.

The plans embed weights from allenai/MolmoAct2-BimanualYAM, released by the Allen Institute for AI under Apache-2.0. The conversion changes execution, not the checkpoint parameters.