MedicalGPT

MedicalGPT is an open-source training pipeline designed for creating and fine-tuning domain-specific Large Language Models through SFT, DPO, RLHF, and ORPO stages.

MedicalGPT provides a comprehensive pipeline for training and fine-tuning Large Language Models, with specialized support for the medical domain. The toolkit implements multi-stage training processes including incremental pretraining, supervised fine-tuning, and alignment methods like RLHF, DPO, ORPO, and GRPO. Users can train models for specific tasks such as agent tool calling and retrieval-augmented generation. It supports a wide range of architectures including Llama, Qwen, Baichuan, and ChatGLM. The framework processes data in JSONL format and allows for LoRA or full-parameter tuning to optimize model performance for domain-specific applications.

MedicalGPT at a glance

Best for: Researcher, Software engineer
Runs on: open source, CLI, self-hosted
Pricing: Free
Released: June 2023
Category: AI coding tool
Website: github.com/shibing624/MedicalGPT

Key features

Incremental pretraining
Supervised fine-tuning
RLHF and reward modeling
Direct Preference Optimization (DPO)
ORPO and GRPO alignment methods
Agent tool calling fine-tuning
On-Policy Distillation (OPD)
Retrieval-augmented generation capabilities

Use it for

Training medical-specific large language models
Fine-tuning models for agent function calling
Implementing file-based Q&A via RAG
Aligning models with human preference data

Alternatives to MedicalGPT

LitGPT — Readable single-file LLM code with tested finetuning recipes
GLiNER — Enables zero-shot entity extraction without labeled training data on consumer CPUs.
ART — Enhances LLM agent performance and reliability using reinforcement learning techniques.
XTuner — High-efficiency training for 200B+ scale MoE models without expert parallelism.
DataFlow — Automates and scales data preparation through reusable, operator-based pipelines.
xTuring — Provides complete control over customizing open-source LLMs.
PEFT — Reduces computational requirements and storage needs for large model training.
OpenCode — Works with 75+ model providers, local models, and existing subscriptions
Soup — Fine-tune and align models like Llama-3.1-8B on a 4 GB laptop GPU.
Genkit — Provides unified APIs and local developer tools for AI application workflows.
Ludwig — Train and fine-tune complex AI models using a single YAML configuration file.
jevgrep — Reduces coding agent costs by providing targeted, relevant repository context.

All MedicalGPT alternatives