Wall-OSS-1.0-lite — RoboDojo base
The pretrained model our RoboDojo training starts from.
It is a preview of Wall-OSS-1.0-lite, published ahead of that release and not identical to it: the two differ mainly in the number of training iterations.
Files
| File | What it is |
|---|---|
model.safetensors |
the weights, bf16 |
config.json |
the model configuration |
normalizer_action.pth, normalizer_propri.pth |
the action and state statistics |
preprocessor_config.json |
the image processor settings |
The tokenizer and the chat template are Qwen/Qwen3.5-4B's.
- Downloads last month
- -