VR-X-SFT-RL

# VR-X: Visual Reasoning Benchmark for UniVR VR-X contains three independent data blocks: 1. SFT data organized by capability. 2. VR-X-RL data for visual-reasoning reinforcement learning. 3. VR-X-Eval held-out evaluation data. VR-X-RL and VR-X-Eval are independent from SFT and must be loaded separately. Public repository paths use anonymous source codes; no source-to-code mapping is published. ## Repository layout . ├── Robot Manipulation/ # SFT only │ └── RM-###/ │ └── train/ ├── VisualGuidance/ # SFT only │ └── VG-###/ │ └── train/ ├── General Reasoning/ # SFT only │ └── GR-###/ │ └── train/ ├── VR-X-RL/ # independent RL block │ └── RL-###/ │ └── part-xxxxx.parquet └── VR-X-Eval/ # independent evaluation block └── eval_combined.parquet The three capability directories contain all SFT data. Anonymous codes preserve source partitions without exposing source names. Parquet shard names are neutral. ## Parquet schema Core fields include question, problem_image_bytes, answer_image_bytes, problem_images, answer_images, height, width, and global_summary. Some anonymous codes contain additional metadata columns. ## Download hf download UniVRBD/UniVR_Data --repo-type dataset --local-dir datasets/UniVR_Data ## Citation @article{ren2026univr, title={UniVR: Thinking in Visual Space for Unified Visual Reasoning}, author={Zhongwei Ren and Yunchao Wei and Zhao Yao and Guixun Luo and Yao Zhao and Weibo Gong and Xiao Liu and Anran Wang and Xiangtai Li and Xiaojie Jin}, url={https://arxiv.org/abs/2607.12800}, year={2026}, } ## License This repository is released under the CC BY 4.0 License.