GH Repository · InternLM
xtuner
A Next-Generation Training Engine Built for Ultra-Large MoE Models
- stars
- 5,195
- 30-day movement
- +10/day
- Related entries
- 69
- Connections
- 2
dockerreinforcement-learningagentgpt-ossinternvlkimi-k2multimodaldeepseek-v3qwen3-moepythonllmintern-s1Pythonqwen3-vlagent-app
xtuner is a training engine from the InternLM team aimed at ultra-large Mixture-of-Experts models. Based on its topics, it covers LLM and multimodal training with reinforcement-learning support, referencing models such as DeepSeek-V3, Qwen3-MoE, Kimi-K2, InternVL, and GPT-OSS.
Reach for it when you need a training engine designed around very large MoE architectures rather than dense models.
Use it to
- Fine-tune ultra-large MoE language models
- Train multimodal models like InternVL and Qwen3-VL
- Run reinforcement-learning training workflows
- Work with MoE models such as DeepSeek-V3 and Kimi-K2
For ML engineers training large MoE and multimodal models
- Role
- agent-app
- Language
- Python
- Licence
- Apache-2.0
- Forks
- 448
- Open issues
- 242
- Last push
- 2026-09-15
- Latest release
- v0.1.0 · 2023-08-30
- Skills shipped
- 9
topicstraining-enginemoellmmultimodalreinforcement-learningfine-tuning