M
ms-swift
modelscope/ms-swift
Use PEFT or Full-parameter to CPT/SFT/DPO/GRPO 600+ LLMs (Qwen3.6, DeepSeek-V4, GLM-5.1, InternLM3, Llama4, ...) and 300+ MLLMs (Qwen3-VL, Qwen3-Omni, InternVL3.5, Ovis2.5, GLM4.5v, Gemma4, Llava, Phi4, ...) (AAAI 2025).
★15.1kstars
Python
Apache-2.0
Updated: Today
📋 Project at a Glance
Tap to expand
What's this?Data & Infrastructure project leveraging deepseek-r1 and embedding, built with Python, open-source
Who made it?Maintained by modelscope team, 15.1K⭐ on GitHub, #122 out of 3133 in Data & Infrastructure
Why does it exist?In the Data & Infrastructure space, deepseek-r1 workflows faced efficiency bottlenecks. ms-swift was built by modelscope to address these embedding challenges.
What can it do?Key use cases: grpo, internvl, liger
How to install with AI?Use an AI coding assistant (Claude Code, Cursor, Copilot) to automatically set up pip dependencies and virtual env. Follow the README — the AI handles the rest.
🔗 github.com/modelscope/ms-swift | 官网 https://swift.readthedocs.io/zh-cn/latest/
🔗 github.com/modelscope/ms-swift | 官网 https://swift.readthedocs.io/zh-cn/latest/
Topics
deepseek-r1embeddinggrpointernvlligerllamallama4llmloramegatronmoemultimodalopen-r1peftqwen3qwen3-6qwen3-omniqwen3-vlrerankersft