| license: other | |
| license_name: nvidia-open-model-license | |
| license_link: https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/ | |
| base_model: nvidia/Cosmos-Predict2.5-2B | |
| tags: | |
| - world-model | |
| - robotics | |
| - video-prediction | |
| - action-conditioned | |
| # Robot World Model (Cosmos-Predict2.5-2B fine-tune) | |
| Built on NVIDIA Cosmos. EMA weights of a fully fine-tuned | |
| Cosmos-Predict2.5-2B action-conditioned world model, post-trained on the RT-1 | |
| robot manipulation dataset by an AIBuildAI agent, autonomously, in a single | |
| 29-hour run on 8x A100. | |
| Official IRASim short-trajectory benchmark (RT-1 test, 4,799 clips): | |
| **PSNR 25.56, SSIM 0.845** (highest SSIM reported on the benchmark). | |
| Usage, training code, and the official evaluation harness: | |
| https://github.com/aibuildai/AI-Build-AI/tree/main/tasks/robot-world-model-post-training | |
| These weights are a derivative model of Cosmos-Predict2.5-2B and are governed | |
| by the NVIDIA Open Model License (see NOTICE). | |