Chinese contentHuggingFace 开源 SmolDataEnvs:5000+ 可验证数据科学 RL 训练任务
HuggingFace 开源了 5000 个数据科学 RL 任务,源自真实 Kaggle 笔记本,评分不用 LLM 判卷,还配了 SFT 轨迹和 Colab 脚本,想自己练小模型可以直接上手。
HuggingFace 开源了 5000 个数据科学 RL 任务,源自真实 Kaggle 笔记本,评分不用 LLM 判卷,还配了 SFT 轨迹和 Colab 脚本,想自己练小模型可以直接上手。
Verified RL environments for code and data science give small-model trainers a ready-made hill-climbing curriculum instead of building reward harnesses from scratch. Practitioners fine-tuning compact models can take the environments, evals and training pipeline and adapt them directly. The full open-source release also makes it a useful reference point against closed RL training suites.