YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
LiLa-WAM all-store memory — Press Button smoke-debug
This public bundle is an infrastructure-only smoke checkpoint for developing the RMBench inference loop. It is not a trained policy and must not be used for model-quality or benchmark claims.
- Source job:
pt-j4po9mia(sgres, 1x A100-80GB) - Optimizer steps: 2
- Training world size: 1
- Effective global batch: 16
- LiLa-WAM source revision:
7549e94a3068c50d02859f3c767027a714dfe9fd - Checkpoint SHA-256:
a74a4e2335eedf03ed9fe9525839fd26c2619e6ae22228f872a6838f5e0ffdd6
The checkpoint contains the trainable 212M-parameter action model only. The
frozen DINOv3-L backbone is intentionally not redistributed here; use the
official facebook/dinov3-vitl16-pretrain-lvd1689m repository.
Contract
- RMBench
press_button - RGB mosaic: 320x384 (W,H), left/right wrist above head camera
- State/action: 14-D model order
[L arm6, R arm6, L grip, R grip] - Action chunk: 32, execute horizon: 16
- Memory: one shared-adapter 64-token page every prediction; retain all pages
- First request of every episode must explicitly reset memory
manifest.json records the exact hashes and debug-only boundary.
- Downloads last month
- 6
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support