Cosmos LIBERO-40 reviewed503 joint policy
Community implementationCosmos-Data trained this Cosmos3-Nano video and action policy from scratch. Training combined original LIBERO-40 video–action pairs with 503 reviewed round-two out-of-distribution videos. The publisher reports a 5:1 sampling ratio, with video-only supervision on the additional clips. Step 3,479 is available as a native training checkpoint.
Resume joint video and action training on LIBERO-40.
Official resource (6)
Papers and technical reports
Paper / report unconfirmed
Official introductory materials are recorded; no version-specific paper or technical report has been verified.
Last checked: 2026-11-22
- Official source · huggingface.co ↗
Source publication date: 2026-11-22
How to access and use
- Public code resources
- Public access
- Model files
- Public access
- Training data
- Public access
Training code for this version is unconfirmed. Public code resources do not establish access to the exact training implementation.
Training / fine-tuning code
Model files
| File | Access | Weights license |
|---|---|---|
| Step 3479 · PyTorch DCP | Access | Weights license unknown |
SourceVerified on · 2026-11-22
Checks before deployment
The publisher’s pinned framework revision was unavailable during verification. A standalone inference export and the checkpoint’s weight license remain unverified. Available weights or code do not establish plug-and-play robot compatibility. Check the required configuration, data and interfaces in the source documentation.
Deployment and usage resources
- Runtime and frameworks
- PyTorch
Architecture and inputs / outputs
- Model type
- Action policy
- Architecture
- WAM
- Released
- November 22, 2026
- Main category · Architecture: Cosmos-Data
Scope and training
Record identity
Community implementation
Output
Actions / Video
Task scope
Multiple tasks
- Record identity: Cosmos-Data
- Record identity · Output · Task scope: Cosmos-Data
- Task scope: NVIDIA