Curious-VLA-dev

This repository hosts the development release assets for Curious-VLA, including:

• SFT training data
• SFT-stage model weights

For full documentation, environment setup, evaluation, SFT training, and GRPO training, please refer to the main project:

• GitHub: github.com/Mashiroln/Curious-VLA

Recommended Usage

• baseline data + Qwen2.5-VL
• golden data + Qwen3-VL

Notes

• The SFT-stage model can be used as the starting point for GRPO training in the main Curious-VLA project.
• For RL training data, released token filters, and end-to-end scripts, see the main repository.