torchtune vs Qwen
LLM training workflows compared
torchtune and Qwen are Python projects for working with large language models, including fine-tuning. torchtune centers on editable PyTorch post-training recipes across supported model families, while Qwen focuses on running, adapting, and deploying Qwen checkpoints.
torchtune: Fine-Tune and Post-Train Large Language Models
torchtune is a PyTorch library for configuring and running LLM post-training workflows, from supervised fine-tuning to preference optimization. It is aimed at developers who want editable recipes and model-specific configs, but is no longer actively maintained.
Qwen: Run and Fine-Tune Pretrained Language Models
Qwen is Alibaba Cloud’s Python repository for running and adapting pretrained and chat language models, with an emphasis on Chinese and English. It includes inference, quantization, fine-tuning, and deployment paths, but is no longer actively maintained; the project points users to Qwen2.
| torchtune | Qwen | |
|---|---|---|
| Language | Python | Python |
| License | BSD-3-Clause | Apache-2.0 |
| Stars | 5.8k | 21.9k |
| Forks | 756 | 2k |
| Last analyzed | Oct 3, 2026 | Oct 3, 2026 |
Key differences
- torchtune provides recipes for methods including supervised fine-tuning, distillation, DPO, PPO, GRPO, and quantization-aware training; Qwen provides tooling for its base and chat checkpoints, including inference and fine-tuning.
- torchtune emphasizes editable YAML configurations and PyTorch implementations; Qwen documents inference and deployment paths using tools such as Transformers, ModelScope, vLLM, and FastChat.
- Qwen includes web, CLI, and OpenAI-style API deployment examples, plus CPU and multi-GPU inference routes; torchtune focuses on training workflows across single-device, multi-device, and some multi-node setups.
- torchtune is licensed under BSD-3-Clause, while Qwen is licensed under Apache-2.0.
- Both projects are no longer actively maintained; torchtune’s development wound down in 2025, while Qwen directs users to Qwen2 as its successor project.
Choose torchtune if you…
- need editable PyTorch recipes for post-training methods across supported model families.
- want to inspect or adapt YAML training configurations, including for LoRA, QLoRA, or preference optimization.
- are maintaining an existing torchtune workflow and can plan around limited upstream support.
Choose Qwen if you…
- want to run or adapt Qwen base and chat checkpoints, with inference examples for conversational and batch use.
- need documented options for quantization, fine-tuning, or deployment through web, CLI, or API examples.
- are evaluating local inference paths such as CPU, multiple GPUs, or vLLM and can account for the repository’s maintenance status.
This comparison is generated with AI from the OSRepos analyses of both projects. Always check each project's repository and documentation before choosing.