X-Coder-RL-40k

# X-Coder-RL-40k X-Coder-RL-40k is a fully synthetic reinforcement learning dataset for competitive programming, containing 40k high-quality tasks with verified test cases. ## Dataset Description - **Tasks**: Synthesized by **o3-mini** - **Test Cases**: Synthesized by **Gemini-2.5-Pro** - **Total Samples**: ~40k - **Purpose**: RLVR training for code generation models ## Dataset Structure The dataset is organized by difficulty level: | File | Difficulty | |------|------------| | `part_0000.parquet` | Easiest | | `part_0001.parquet` | Easy | | `part_0002.parquet` | Medium | | `part_0003.parquet` | Hard | | `part_0004.parquet` | Hardest | ## Task Difficulty Distribution **Table: Distribution of Proprietary LLMs' First-Try Pass Rates on Test Cases** | Pass Rate Range | Qwen3-Max | Gemini-2.5-Pro | GPT5-High | |-----------------|-----------|----------------|-----------| | (0–20) | 11.06% | 9.57% | 3.07% | | [20–40) | 16.44% | 14.38% | 4.83% | | [40–60) | 18.59% | 17.17% | 6.49% | | [60–80) | 16.36% | 15.80% | 7.80% | | [80–100) | 14.39% | 14.90% | 10.82% | | 100 | 23.16% | 28.18% | 66.98% | ## Usage ```python from datasets import load_dataset # Load all data dataset = load_dataset("IIGroup/X-Coder-RL-40k") # Load specific difficulty level easy_tasks = load_dataset("IIGroup/X-Coder-RL-40k", data_files="part_0000.parquet") hard_tasks = load_dataset("IIGroup/X-Coder-RL-40k", data_files="part_0004.parquet") ``` ## Related Resources - **GitHub**: [X-Coder](https://github.com/JieWu02/X-Coder) - **SFT Data**: [IIGroup/X-Coder-SFT-376k](https://huggingface.co/datasets/IIGroup/X-Coder-SFT-376k) - **Models**: - [IIGroup/X-Coder-RL-Qwen3-8B](https://huggingface.co/IIGroup/X-Coder-RL-Qwen3-8B) - [IIGroup/X-Coder-RL-Qwen2.5-7B](https://huggingface.co/IIGroup/X-Coder-RL-Qwen2.5-7B) ## Citation ``` @article{wu2026x, title={X-Coder: Advancing Competitive Programming with Fully Synthetic Tasks, Solutions, and Tests}, author={Wu, Jie and Li, Haoling and Zhang, Xin and Guo, Jiani and Luo, Jane and Liu, Steven and Huang, Yangyu and Chu, Ruihang and Li, Scarlett and Yang, Yujiu}, journal={arXiv preprint arXiv:2601.06953}, year={2026} } ``` ## License This dataset is licensed under the Apache License 2.0.