Implementation:Unslothai Unsloth Pyproject Configuration
| Knowledge Sources | |
|---|---|
| Domains | Build_System, Configuration |
| Last Updated | 2026-02-07 08:40 GMT |
Overview
Project configuration file defining build system, package metadata, dependencies, and optional GPU-specific installation targets for the Unsloth package.
Description
The pyproject.toml file is the central build and packaging configuration for Unsloth. It uses setuptools with dynamic versioning via setuptools-scm, specifies Python >=3.9 compatibility, and defines an extensive set of optional dependency groups covering different CUDA versions (cu118 through cu130), Intel GPU, AMD GPU, and convenience targets for Colab and Kaggle environments. It also configures ruff linter/formatter settings with Unsloth's custom kwarg spacing rules.
Usage
Reference this configuration when installing Unsloth with specific GPU backends (e.g., pip install unsloth[cu124]) or when understanding the project's dependency matrix and build requirements.
Code Reference
Source Location
- Repository: Unslothai_Unsloth
- File: pyproject.toml
- Lines: 1-1034
Signature
[build-system]
requires = ["setuptools>=80.9.0", "setuptools-scm>=9.2.0"]
build-backend = "setuptools.build_meta"
[project]
name = "unsloth"
dynamic = ["version"]
requires-python = ">=3.9,<3.14"
license = "Apache-2.0"
description = "2-5X faster training, reinforcement learning & finetuning."
Import
pip install unsloth
# Or with specific GPU backend:
pip install "unsloth[cu124]"
pip install "unsloth[colab]"
I/O Contract
Inputs
| Name | Type | Required | Description |
|---|---|---|---|
| optional dependency group | string | No | GPU backend selector (cu118, cu121, cu124, cu126, cu128, cu130, colab, kaggle, amd, intel-gpu-*) |
| Python version | string | Yes | Must be >=3.9, <3.14 |
Outputs
| Name | Type | Description |
|---|---|---|
| Installed package | Python package | Unsloth with selected dependencies installed |
| Version | string | Dynamically determined via setuptools-scm from git tags |
Usage Examples
Standard Installation
# Install with CUDA 12.4 support
pip install "unsloth[cu124]"
# Install for Google Colab
pip install "unsloth[colab]"
# Install base (HuggingFace dependencies only)
pip install "unsloth[base]"
Development Installation
# Clone and install in editable mode
git clone https://github.com/unslothai/unsloth.git
cd unsloth
pip install -e ".[cu124]"