ms-swift (modelscope/ms-swift) is an open-source AI project on GitHub. Repository summary: 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). Its focus includes retrieval-augmented generation, image and vision workflows. It is suitable for extension, integration, and iterative delivery in real workflows.
License
Apache-2.0
Stars
14,347
Features
- Core capability: 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).
- Supports vector retrieval and retrieval-augmented reasoning
- Supports image generation, editing, or vision understanding
- Repository: modelscope/ms-swift
- Primary language: Python
- Open-source license: Apache-2.0
Use Cases
- Builds enterprise knowledge Q&A and document retrieval systems
- Used for visual content production and model experimentation
- Build internal AI workflow prototypes with ms-swift
- Validate ms-swift in production-like engineering scenarios
- Image generation and visual creation
- Model evaluation and comparison
FAQ
Teams should first define integration boundaries and call patterns, then map repository capabilities into concrete interfaces, parameters, and access rules. GitHub repository: https://github.com/modelscope/ms-swift. Community traction is around 14,345 stars. License: Apache-2.0.
It usually works as an execution component or capability layer, with common deployment fits such as: Builds enterprise knowledge Q&A and document retrieval systems, Used for visual content production and model experimentation, Build internal AI workflow prototypes with ms-swift.
