Instructions to use GmanGmanGman/M2-VLA with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use GmanGmanGman/M2-VLA with Transformers:
# Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("GmanGmanGman/M2-VLA", dtype="auto") - Notebooks
- Google Colab
- Kaggle
M2-VLA Checkpoints
This repository hosts the public M2-VLA checkpoints for the LIBERO benchmark suites.
Checkpoints
| Folder | Suite |
|---|---|
M2VLA-spatial |
LIBERO Spatial |
M2VLA-object |
LIBERO Object |
M2VLA-goal |
LIBERO Goal |
M2VLA-long |
LIBERO Long |
Each checkpoint folder contains the model weights, tokenizer/processor files, dataset statistics, action head, and proprioception projector required by the inference code.
Notes
These checkpoints are released for research use with the M2-VLA project. Please refer to the project repository and paper for environment setup, evaluation scripts, and citation information.