QwenLM

Qwen: Run Alibaba Cloud's Qwen Language Models

Qwen is Alibaba Cloud’s repository for its Qwen pretrained and chat language models, with code for inference, fine-tuning, quantization, and deployment. It supports Transformers and ModelScope workflows, but the repository is no longer actively maintained.

View on GitHub Compare
Stars
21.9k
Forks
2k
Stars, last 30 days
+129
Last commit
8 months ago
Contributors
31

Our take

Quietno commits in the last 90 days and no release in a year

Health score

42/100

How it is scored
Activity
19
Community
69
Issues
59
Pull requests
20

+129 stars in the last 30 days

Qwen provides broad practical tooling around Alibaba Cloud’s Qwen models, but its quiet maintenance status makes the newer Qwen2 repository the safer starting point for new work.

Good fit if

  • You need the older Qwen model checkpoints or want to reproduce workflows documented for this model generation.
  • Your team can work with Python, PyTorch, Transformers, and GPU-oriented inference or training tools.
  • You value a permissively licensed codebase and can maintain or adapt it yourself; the repository is under Apache-2.0.

Look elsewhere if

  • You are starting a project that needs actively maintained code; the README says this repository is no longer actively maintained, with no recent commits or release in the measured period.
  • You need recent upstream changes or merged pull requests, since none were recorded in the last 90 days.
  • You want a ready-to-use CPU deployment from this repository; it warns that direct CPU inference can be very inefficient and recommends qwen.cpp instead.
All health signals
Last commit2026-03-05 (8 months ago)
Commits, last 90 days0
Releases, last 12 months0
Contributors31 (top contributor: 35% of commits)
Issues closed, last 90 days49 (typically closed in 38 days)
Pull requests merged, last 90 days0
Project age3 years

Checked on 2026-10-24 with the GitHub API.

Project history

Daily GitHub snapshots since Oct 7, 2026. History covers stars and forks; health signals are not tracked historically.

Selected snapshot

Stars

22,057

Forks

2,008

Change since Nov 17, 2026: +5 stars·+1 forks

Overview

Qwen is Alibaba Cloud’s Python repository for running and adapting the Qwen family of pretrained and chat language models. It addresses the practical work around using those models: loading checkpoints, serving chat, reducing inference memory with quantization, and fine-tuning for downstream tasks.

The repository describes Qwen 1.8B, 7B, 14B, and 72B models, with examples using Hugging Face Transformers and ModelScope. It also documents deployment paths including vLLM, FastChat, web and CLI demos, and an OpenAI-style API. The README says this repository is no longer actively maintained and directs users to Qwen2 for the newer codebase.

Key Features

  • Examples for inference with base and chat model checkpoints through Transformers or ModelScope.
  • Int4 and Int8 GPTQ model support, plus optional quantization of the attention KV cache.
  • Fine-tuning scripts for full-parameter training, LoRA, and Q-LoRA, with DeepSpeed and FSDP support described.
  • Deployment guidance for vLLM and FastChat, along with web, CLI, and OpenAI-style API demos.
  • Batch inference examples and optional FlashAttention support for GPU inference.
  • CPU usage instructions and a recommendation to use the separate qwen.cpp implementation for CPU deployment.
  • Links to model checkpoints on Hugging Face and ModelScope, and to the project’s technical report.

Use Cases

  • Researchers exploring Chinese-English language modeling can compare the listed Qwen model sizes and use the repository’s inference examples.
  • Application developers can adapt Qwen-Chat for conversation, summarization, translation, coding, or other text tasks.
  • Teams with limited GPU memory can evaluate the documented Int4/Int8 models or KV-cache quantization.
  • Model practitioners can use the provided scripts as a starting point for LoRA, Q-LoRA, or full-parameter fine-tuning.
  • Operators seeking a self-hosted chat endpoint can follow the vLLM, FastChat, or local API deployment instructions.

What you need

Detected in the repository

  • Python (from requirements.txt)
  • Automated checks on GitHub Actions

License in plain words

Apache-2.0permissive

  • Commercial use: yes
  • Modify and redistribute: yes
  • You must keep: the license, the NOTICE file and a note of your changes
  • Share your changes: no
  • Includes an explicit patent grant from the contributors:

A summary, not legal advice: the LICENSE file is what applies.

Getting Started

Install the Python dependencies:

pip install -r requirements.txt

The README’s quickstart shows how to load a Qwen checkpoint with Transformers. See the repository README for model selection, setup details, and deployment instructions.

Considerations

The measured project status is Quiet: there have been no commits in the last 90 days and no release in over a year. The repository has contributions from multiple people, and contribution activity is not concentrated in one person, but the measured period also shows no merged pull requests. Issues have continued to close, with a measured median close time of 38 days; this does not replace the missing recent code and release activity.

The code requires Python, PyTorch, and Transformers. GPU use is recommended for practical inference, and the README describes CUDA and optional FlashAttention setup; CPU execution is possible but warned to be extremely slow. The repository reports CI but no tests in the supplied project facts. Check the relevant model card and current library compatibility before adopting the older examples, which rely on trust_remote_code=True in the documented Transformers workflow.

Found this useful?

Share it with someone who would like Qwen.

Comparisons