Instructions to use allenai/MolmoAct2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use allenai/MolmoAct2 with Transformers:
# pip install -U transformers accelerate # Load model directly from transformers import AutoModelForImageTextToText model = AutoModelForImageTextToText.from_pretrained("allenai/MolmoAct2", trust_remote_code=True, device_map="auto") - Notebooks
- Google Colab
- Kaggle
Commit History
-
Update MolmoAct2 action mode inference API
Verified signature46e066f
-
Remove unused action expert modules
Verified signature559eb2e -
Fix MolmoAct2 remote norm stats loading
Verified signatureAllow predict_action to load norm_stats.json from the Hub when a checkpoint is loaded by repo id instead of a local directory.1e8b98e -
Update MolmoAct2 config schema and bf16 inference fix
Verified signatureRemove obsolete action expert config fields, keep max action dimensions/horizons explicit, and make bf16/fp16 action postprocessing safe during inference.ba955ee
-
Enable CUDA graph AR decode for discrete actions
Verified signatureGeneralizes the existing single-token decode CUDA graph helper and uses it for no-depth discrete action autoregressive decoding when enable_cuda_graph=True. The original KV-cache path is preserved when enable_cuda_graph=False.c98c5ec
-
Add files using upload-large-folder tool
Verified signature75c5cf5