Instructions to use adipotnis/Cosmos3-Edge-Charging with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- LeRobot
How to use adipotnis/Cosmos3-Edge-Charging with LeRobot:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Notebooks
- Google Colab
- Kaggle
Cosmos3 Edge Charging
Final 10,000-step checkpoint, completed successfully on November 11, 2026. Fine-tuned on 51 charging demonstrations (21,895 frames) with low and wrist cameras, matching the MolmoAct2-Charging dataset selection. No robot evaluation result is claimed.
The repository root contains the inference model, tokenizer, processors, normalization
statistics and training configuration. training_state/ preserves optimizer, scheduler,
RNG and step state. To recreate the training checkpoint layout for resuming, place root
model files under 010000/pretrained_model/ and state files under 010000/training_state/.
Recipe: batch 16, action horizon 16, learning rate 5e-5, 500 warmup steps, linear decay through 10,000 steps, bf16, gradient checkpointing, state input, quantile state/action normalization, seed 42. Each camera is resized to 192x256, low left and wrist right.
- Downloads last month
- 7