Torchtune
fine-tuning5,768BSD-3-Clause

Torchtune

PyTorch-native library for easily fine-tuning LLMs with memory-efficient techniques and recipe-based configs.

Updated Jun 7, 2026
Website GitHub Docs
Platforms
linux, macos
Pricing
free-open-source
Status
active
License
BSD-3-Clause

What it does

Core capabilities at a glance

  • Native PyTorch implementations with no extra frameworks
  • LoRA, QLoRA, and full fine-tuning support
  • Memory-efficient with FSDP and activation offloading
  • Recipe-based configurable training
  • Supports Llama, Mistral, Gemma, Qwen, and more
  • Eval harness and checkpoint conversion utilities

Deep dive

The full breakdown - performance, comparisons, and setup

Torchtune

Torchtune is PyTorch's official LLM fine-tuning library. It provides clean, readable PyTorch implementations of training recipes without the abstraction layers found in other frameworks.

What it is

Torchtune is developed by the PyTorch team at Meta. It focuses on providing reference-quality implementations of fine-tuning techniques like LoRA, QLoRA, and full fine-tuning, optimized for memory efficiency.

Get started

pip install torchtune
tune download meta-llama/Meta-Llama-3-8B-Instruct
tune run lora_finetune_single_device --config llama3_8B_lora_single_device

When to use something else

Frequently asked

Quick answers to common questions

What is Torchtune?

Torchtune is a fine-tuning tool for local AI workloads. PyTorch-native library for easily fine-tuning LLMs with memory-efficient techniques and recipe-based configs.

Is Torchtune free and open source?

Yes, Torchtune has 5,768 GitHub stars and is licensed under BSD-3-Clause. You can self-host it for free on linux, macos.

What platforms does Torchtune support?

Torchtune runs on linux, macos.

What hardware do I need for Torchtune?

The hardware requirements depend on which models you run. Check our hardware directory for compatible GPUs and systems. Torchtune has 5,768 GitHub stars and an active community.

Does Torchtune support GPU acceleration?

Torchtune supports GPU acceleration via CUDA, Metal, or Vulkan depending on your platform. For the best performance, pair it with an NVIDIA RTX 4090 or 5090.

What are the best alternatives to Torchtune?

Popular alternatives include other fine-tuning tools in our directory. Browse our full collection at /tool for comparisons, community reviews, and benchmark data to find the right fit for your workflow.

How much does Torchtune cost?

Torchtune is free-open-source. It is completely free and open source to self-host.

Pairs well with

Complementary tools, models, and hardware

Comments coming soon

Configure NEXT_PUBLIC_GISCUS_REPO_ID and NEXT_PUBLIC_GISCUS_CATEGORY_ID at giscus.app to enable.