MedicalGPT

Why we included this project

MedicalGPT is a full training pipeline for teams that want to build a medical language model from their own data rather than just call an API. It covers the standard fine-tuning stages, from continued pretraining on domain documents through supervised instruction tuning to preference alignment via RLHF, DPO, ORPO, GRPO, and on-policy distillation. The repo ships ready-to-run shell scripts for each stage, plus LoRA and full-parameter training support for models like Qwen, Llama, Baichuan, and ChatGLM, and it handles chat templates and data conversion for you. That makes it a practical fit for researchers and engineers who have medical corpora and GPU time and want reproducible, scripted experiments instead of hand-rolling training loops. Just don't expect a finished chatbot: this is a toolkit you run to produce one, so it suits people comfortable with configs and data curation.