█

AI/TLDR

Unsloth

Train and fine-tune LLMs faster with less GPU memory

github.com/unslothai/unsloth★ 66.9kunsloth.ai

Overview

Unsloth is a library for fine-tuning and running large language models with lower memory use and shorter training times. It focuses on LoRA, QLoRA, and other parameter-efficient methods, along with full fine-tuning, reinforcement learning, and 4-bit, 16-bit, and FP8 training. According to the project, it can train and run RL on 500+ models up to 2x faster while using up to 70% less VRAM, with no accuracy loss.

It comes in two forms. Unsloth Studio is a web UI that runs on Windows, Linux, WSL, and macOS, letting you search, download, run, and train text, vision, audio, and embedding models, and export them to formats like GGUF and 16-bit safetensors. Unsloth Core is the code-based version installed with pip or uv, aimed at people who want to script training and inference directly.

It fits the fine-tuning frameworks category by lowering the hardware bar for training. With its custom Triton kernels and memory savings, you can fine-tune many open models on a single consumer GPU instead of a multi-GPU cluster.

What it does

  • Trains and runs RL on 500+ models up to 2x faster with up to 70% less VRAM, per the project
  • Supports LoRA, QLoRA, full fine-tuning, RL, pretraining, and 4-bit, 16-bit, and FP8 training
  • Custom Triton and math kernels, including collaborations with PyTorch and Hugging Face
  • Unsloth Studio web UI to search, download, run, train, and export models on Windows, Linux, WSL, and macOS
  • Exports trained models to GGUF, 16-bit safetensors, and other formats
  • Reinforcement Learning support using up to 80% less VRAM for methods like GRPO

Getting started

You can use Unsloth either through the Studio web UI or as a code-based install. The commands below come straight from the README.

Install Unsloth Studio (macOS, Linux, WSL)

Run the install script. Use the same command later to update.

bashbash
curl -fsSL https://unsloth.ai/install.sh | sh

Install Unsloth Studio on Windows

Use the PowerShell install command instead.

bashbash
irm https://unsloth.ai/install.ps1 | iex

Launch the Studio

Start the web UI on a port. It is local-only by default; add -H 0.0.0.0 for cloud or global access.

bashbash
unsloth studio -p 8888

Or install Unsloth Core (code-based)

For the Python library, install with uv. On Windows, pip install unsloth also works if PyTorch is already installed.

bashbash
uv venv unsloth_env --python 3.13
source unsloth_env/bin/activate
uv pip install unsloth --torch-backend=auto

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

When to use it

  • Fine-tune an open LLM with LoRA or QLoRA on a single consumer GPU when VRAM is limited
  • Run reinforcement learning (for example GRPO) on a model with reduced memory use
  • Train and export a custom model to GGUF or safetensors for local inference
  • Search, download, and run text, vision, audio, or embedding models locally through the Studio web UI

How Unsloth compares

Unsloth alongside other open-source fine-tuning frameworks tools AI/TLDR tracks, ranked by GitHub stars.

ToolStarsWhat it does
LLaMA-Factory★ 72.3kAn end-to-end training suite with a web UI that covers pre-training, supervised fine-tuning, and RLHF for hundreds of LLMs and multimodal models.
Unsloth★ 66.9kTrain and fine-tune LLMs faster with less GPU memory
PEFT★ 21.3kHugging Face's library of parameter-efficient fine-tuning methods such as LoRA, DoRA, and prompt tuning that train small adapters instead of full models.
FinGPT★ 20.5kFinGPT is an open-source project of financial LLMs, fine-tuned with LoRA on news and tweet data for tasks like sentiment analysis, relation extraction, and stock-move forecasting.
ms-swift★ 14.6kModelScope's framework for fine-tuning and deploying 600+ LLMs and 300+ multimodal models, supporting PEFT and full-parameter SFT, DPO, and GRPO.
LitGPT★ 13.4kAn open-source toolkit from Lightning AI to pretrain, finetune, and serve 20+ large language models, each written from scratch for speed and full control.
Axolotl★ 12.1kA config-driven tool for fine-tuning and post-training open LLMs that supports SFT, LoRA/QLoRA, DPO, GRPO, and multi-GPU training across many model families.
Ludwig★ 11.7kLudwig is a low-code framework that lets you train, fine-tune, and deploy LLMs, multimodal, and tabular models using a YAML config instead of boilerplate Python.