Install Pruna

Prerequisites

pruna is officially supported on Linux. While it can be used on macOS and Windows, compatibility depends on the specific algorithms you plan to use, as some may have platform-specific dependencies. If you don’t have access to a Linux environment, you can still easily experiment with pruna using cloud platforms like Google Colab.

Installing Pruna

If all prerequisites are met, you can simply install pruna with the following commands:

We support pip installation in both CPU and GPU environments. When using pip, we recommend using uv instead of the native pip command to avoid the performance penalty of rebuilding package environments. When you have uv installed, you can install pruna with the following command:

pip install pruna==0.3.4

pip install pruna==0.3.4 --extra-index-url https://download.pytorch.org/whl/cpu

Conda is the recommended way to create and manage your Python environments.

If you don’t already have it, you can download and install Miniconda from here. When you have Conda installed, you can create a new environment as follows:

conda create -n pruna python=3.11
conda activate pruna

If you are on a CUDA-enabled machine, you will need to make sure that the CUDA runtime toolkit is installed.

You can check that nvcc --version provides a version that matches your nvidia-smi version. If nvcc --version fails or there is a version mismatch, you should install the correct CUDA runtime toolkit as follows:

conda install nvidia/label/cuda-12.4.0::cuda

Additionally, there are a few algorithms that require additional dependencies.

Extra Name	Description
`stable-fast`	StableFast compiler for diffusion models
`llamacpp`	Llama.cpp conversion and inference
`gptq`	GPTQ quantization support
`awq`	AWQ quantization support
`ctranslate`	CTranslate2 backend for speech models
`upscale`	Real-ESRGAN upscaling support
`kvpress`	KV cache compression for LLMs

Tip

Install only the extras your algorithms need. For example:

pip install "pruna[stable-fast]"

You can find out if your desired algorithm requires additional dependencies by checking the Algorithms Overview page.