EvoVLA-v2 assets
Model assets for EvoVLA-v2: self-supervised RL (SSRL) side path for post-training VLA policies on RLinf.
Contents
r3m_18/
Mirror of the facebookresearch/r3m resnet18 weights (model.pt, config.yaml), originally distributed via Google Drive links in the upstream repo.
Used by the SSRL contrastive branch (r_con) as the frozen visual backbone. The r3m pip package reads weights from ~/.r3m/r3m_18/; download with:
hf download zetliu2001/EvoVLA-v2 r3m_18/ --local-dir ~/.r3m/r3m_18
Upstream code and weights are MIT-licensed; see the upstream repository.
Related official assets (download separately)
| Asset | Source |
|---|---|
| Οβ.β CALVIN ABC-D SFT checkpoint | RLinf/RLinf-Pi05-CALVIN-ABC-D-SFT |
| OpenPI tokenizer | RLinf/openpi_tokenizer |
| CLIP text encoder | openai/clip-vit-base-patch32 |
Inference Providers NEW
This model isn't deployed by any Inference Provider. π Ask for provider support