huggingface/hugging-face-model-trainer logo

huggingface/hugging-face-model-trainer

Free

Train models with TRL: SFT, DPO, GRPO

FreeFree tier
Type
Open Source

About huggingface/hugging-face-model-trainer

The Hugging Face Model Trainer is an agent skill designed for training Hugging Face models using the Transformer Reinforcement Learning (TRL) library. It supports three major training methods: Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and Generative Reinforcement Policy Optimization (GRPO). This tool enables users to fine-tune language models with reinforcement learning techniques, making it suitable for customizing models for specific tasks or improving alignment with human preferences. It is part of the awesome-agent-skills collection and is open source.

Key Features

Supports Supervised Fine-Tuning (SFT)
Supports Direct Preference Optimization (DPO)
Supports Generative Reinforcement Policy Optimization (GRPO)
Built on the TRL library for reinforcement learning with transformers
Open-source agent skill for integration into workflows

Pros & Cons

Pros
  • Supports multiple state-of-the-art training methods in one tool
  • Open source and free to use
  • Integrates with the Hugging Face ecosystem

Best For

Fine-tuning language models for specific tasks using supervised learningAligning model outputs with human preferences via DPOOptimizing model policies through generative reinforcement learning (GRPO)