{"full_name":"shibing624/MedicalGPT","owner":"shibing624","name":"MedicalGPT","shard":"1","url":"https://github.com/shibing624/MedicalGPT","description":"MedicalGPT: Training Your Own Medical GPT Model with ChatGPT Training Pipeline. 训练医疗大模型，实现了包括增量预训练(PT)、有监督微调(SFT)、RLHF、DPO、ORPO、GRPO。","homepage":"","topics":["llama","chatgpt","gpt","llm","medical","dpo","medicalgpt"],"tag_pages":["chatgpt","gpt","llama","llm"],"language":"Python","license":"Apache-2.0","avatar":"https://avatars.githubusercontent.com/u/10249622?u=39133b98c78fe7cbaf0458be30352f1c5b1197c4&v=4","created_at":"2023-06-02T06:07:11Z","pushed_at":"2026-09-15T13:46:54Z","metrics":{"stars":5896,"forks":803,"watchers":39,"issues":455,"prs":41,"releases":17,"issues_open":6,"issues_closed":449,"prs_open":0,"prs_closed":7,"prs_merged":34,"discussions":0},"growth":{"day":{"stars":1,"forks":0,"base_day":"2026-11-15"}},"history":{"dates":["2026-11-10","2026-11-11","2026-11-12","2026-11-13","2026-11-14","2026-11-15","2026-11-16"],"stars":[5890,5891,5892,5892,5893,5895,5896],"forks":[804,803,803,803,803,803,803],"watchers":[null,39,39,39,39,39,39],"issues_open":[null,6,6,6,6,6,6],"prs_open":[null,0,0,0,0,0,0]},"boards":{},"same_owner":[],"issues":[{"number":492,"title":"请问sft训练几个epoch比较好呢","comments":5,"created_at":"2026-05-07T08:05:56Z"},{"number":445,"title":"CPT阶段，采用36w条医疗数据集，训练一轮后loss从23.1-17.5左右正常吗","comments":5,"created_at":"2025-03-02T10:29:21Z"},{"number":439,"title":"reward model 训练数据量的问题","comments":2,"created_at":"2025-02-16T13:41:29Z"},{"number":494,"title":"DPO/OPD 训练入口可复现性修复：seed/revision 传递、参数去重、冻结验证","comments":1,"created_at":"2026-07-21T08:43:41Z"},{"number":462,"title":"老师您好，请问本项目是否支持多机多卡的分布式训练，是否有SFT的多机多卡Demo","comments":1,"created_at":"2025-05-13T08:57:19Z"},{"number":450,"title":"请问run_training_dpo_pipeline中的增量预训练阶段中的训练数据只能是txt格式吗","comments":1,"created_at":"2025-03-15T08:42:37Z"}],"discussions":[],"latest_release":{"name":"v2.7.0","tag":"2.7.0","published_at":"2026-04-20T04:11:20Z","body":"## What's New\n- 支持独立 OPD(On-Policy Distillation) 蒸馏训练\n- 新增 training/opd_training.py 训练入口，基于 TRL GKDTrainer 实现 student/teacher 蒸馏\n- 新增 scripts/run_opd.sh 启动脚本，可直接复用现有 ShareGPT/SFT 数据格式\n- 补充 README 与训练参数文档，说明 OPD 参数、部署方式与 teacher/student 使用建议\n- 依赖升级到 trl>=0.29.0 以支持 GKDTrainer\n\n## Notes\n- OPD v1 为独立训练链路，不改动现有 PPO/GRPO 主流程\n- 训练产物与 SFT/LoRA 一样可独立部署或合并后部署\n","is_prerelease":false},"readme":true,"readme_zh":false,"news":[],"stale":false,"snapshot_at":"2026-11-16T09:23:14Z","history_start":"2026-09-06"}