Fine-tune LLMs twice as fast with 70% less VRAM, so even a single GPU can train.
TRL
基于 Transformers 的强化学习后训练库（SFT、DPO、GRPO）。
- Category
- 🏋️ Training, Fine-Tuning & MLOps
- Repository
- huggingface/trl
- Stars
- ★ 19,533
- Forks
- 3,070
- Language
- Python
- License
- Apache-2.0
- Last updated
- 2026-10-16
- Visits
- 0
RLHF
Similar projects
Unified, efficient fine-tuning of 100+ LLMs with a no-code web UI, the most popular fine-tuning tool in China.
A unified AI compute engine for distributed training, tuning, inference and serving.
机器学习与 GenAI 生命周期管理平台，含追踪与评测。
🧠 Train a 64M-parameter LLM from scratch in just 2h!
微软的深度学习优化库，让分布式训练与推理简单高效。