modelscope/

ms-swift

GitHub Website

ms-swift is a Python framework for fine-tuning and deploying large models (text and multimodal) with various training paradigms including LoRA/Full-parameter, GRPO, DPO, embedding/reranker tasks, and support for Megatron parallelism and vLLM-based acceleration.

15kstars

1.6kforks

647issues

Apache-2.0license

2023since

Star historydaily snapshots by VibeCrowd

Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).

Alternatives & relatedmatched by topic overlap

hiyouga/

LlamaFactory

▲ 162/wk

Unified Efficient Fine-Tuning of 100+ LLMs & VLMs (ACL 2024)

74k9.0kPython

chatchat-space/

Langchain-Chatchat

▲ 24/wk

Langchain-Chatchat（原Langchain-ChatGLM）基于 Langchain 与 ChatGLM, Qwen 与 Llama 等语言模型的 RAG 与 Agent 应用 | Langchain-Chatchat (formerly langchain-ChatGLM), local knowledge based LLM (like ChatGLM, Qwen and Llama) RAG and Agent app with langchain

39k6.3kPythonquiet

vllm-project/

vllm

▲ 704/wk

A high-throughput and memory-efficient inference and serving engine for LLMs

88k20kPython

unslothai/

unsloth

▲ 524/wk

Unsloth is a local UI for training and running Kimi K3, Gemma 4, Qwen3.6, DeepSeek-V4, GLM and other models.

70k6.3kPython

AstrBotDevs/

AstrBot

▲ 325/wk

AI Agent Assistant & development framework that integrates lots of IM platforms, LLMs, plugins and AI feature, and can be your openclaw alternative. ✨

39k2.8kPython

ollama/

ollama

▲ 640/wk

Get up and running with Kimi-K2.6, GLM-5.2, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.

178k17kGo

Comparems-swift vs LlamaFactory ms-swift vs Langchain-Chatchat ms-swift vs vllm

Reviewgenerated from repository data · Aug 5, 2026

What it is

ms-swift is a large-model fine-tuning and deployment framework that supports training, inference, evaluation, quantization, and deployment for 600+ text-only large models and 400+ multimodal large models. It integrates Megatron parallelism methods and GRPO-related reinforcement learning algorithms, and enables training tasks such as embedding, reranker, and sequence classification, with a Web-UI for training/inference/evaluation/quantization.

How it works

The project provides tooling for training and deployment across full pipelines, including acceleration through vLLM, LMDeploy, and SGLang, as well as model quantization using GPTQ/AWQ/BNB/FP8. It supports distributed training (DDP, FSDP, DeepSpeed ZeRO, Megatron), memory optimization techniques (Liger-Kernel, Flash-Attention), and multimodal training via data packing and mixed modality support.

Getting started

Installation

pip install ms-swift -U

# Using uv
pip install uv
uv pip install ms-swift -U --torch-backend=auto

To install from source:

# pip install git+https://github.com/modelscope/ms-swift.git

git clone https://github.com/modelscope/ms-swift.git
cd ms-swift
# The main branch is for swift 4.x. To install swift 3.x, please run the following command:
# git checkout release/3.12
pip install -e .

# Using uv
uv pip install -e . --torch-backend=auto

Recent releases

v4.4.2 Patch release v4.4.2 (2026-07-21): Full Changelog: https://github.com/modelscope/ms-swift/compare/v4.4.1...v4.4.2
v4.4.1 Patch release v4.4.1 (2026-07-13): Full Changelog: https://github.com/modelscope/ms-swift/compare/v4.4.0...v4.4.1
v4.4.0 v4.4.0 (2026-07-06): 中文版，新增 RL、GRPO/多教师训练等特性。

Traction

Stars: 15045


## Behind the repo
N/A

## Caveats
License: Apache-2.0
Age: Created 2023-08-01
Last push: 2026-08-04

SharePost on X LinkedIn