Install Unsloth on Windows

To install Unsloth Desktop on Windows, follow the steps below.

1

Download

Download Unsloth for Windows

2

Install

1. Open the Unsloth installer (.exe file)

2. Follow the setup prompts

3. Launch the app and wait for installation to complete

3

Choose a model

Open Select model or Model hub, choose a model and quantization, then download it. Once it finishes, start chatting - no setup required.

4

Unsloth is now ready

To start chatting, type a message and press Enter.

• Connect tools: Claude Code, Codex, web search, MCP and more

• Train models: Fine-tune text, diffusion, embedding, and more

• Generate media: Create and train images, video, TTS locally

Install Unsloth manually

The sections below cover the web-based Unsloth Studio and the original code-based Unsloth Core.

For Unsloth Core on Windows, choose Conda, Docker, or WSL below. If PyTorch is already installed, pip install unsloth may be enough.

Conda TutorialDocker TutorialWSL Tutorial

Unsloth Studio

We have launched a new web UI called Unsloth Studio that works on Windows out of the box:

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

Use the same command to update.

Then to launch every time:

unsloth studio -H 0.0.0.0 -p 8888

For detailed Unsloth Studio install instructions and requirements, view our guide.

Below are installation instructions for the original Unsloth Core:

Method #1 - Windows via Conda:

1

Install Miniconda (or Anaconda)

Download Anaconda here. Our suggestion is to use Miniconda. To use it, first enter Powershell - search "Windows Powershell" in Start:

Then it'll open up Powershell:

Then copy paste the below: CTRL+C, and paste it in Powershell CTRL+V:

Copy

Invoke-WebRequest -Uri "https://repo.anaconda.com/miniconda/Miniconda3-latest-Windows-x86_64.exe" -OutFile ".\miniconda.exe"
Start-Process -FilePath ".\miniconda.exe" -ArgumentList "/S" -Wait
del .\miniconda.exe

Accept the warning and press "Paste anyway" and wait.

It's downloading the installer like below:

After installing, open up open Anaconda Powershell Prompt to use Miniconda via Start -> Search for it:

Then you'll see:

2

Make conda environment

Copy

conda create --name unsloth_env python==3.12 -y
conda activate unsloth_env

You will see:

3

Check nvidia-smi to confirm you have a GPU, and look for the CUDA version

After typing nvidia-smi in Powershell, you should see something like below. If you don't have nvidia-smi or the below fails to pop up, you need to reinstall NVIDIA drivers.

4

Install PyTorch

When running nvidia-smi you will see at the top right corner: "CUDA Version: 13.0". Install PyTorch in PowerShell via. Change 130 to your CUDA version - ensure the version exists and matches your CUDA driver version.

Copy

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu130

You will see:

Try running this in Python via python after PyTorch is installed:

Copy

import torch
print(torch.cuda.is_available())
A = torch.ones((10, 10), device = "cuda")
B = torch.ones((10, 10), device = "cuda")
A @ B

You should see a matrix of 10s. Also verify True for the first.

5

Install Unsloth (only if PyTorch works!)

Confirm PyTorch works fine and runs - if not PyTorch is broken and that means your Windows machine might need a re installation of CUDA drivers unfortunately.

In Powershell (after exiting Python via exit() , do and wait:

Copy

pip install unsloth

6

Verify Unsloth works

Now use any script in Unsloth Notebooks (save to .py file), or use the below basic script:

You should see:

Copy

🦥 Unsloth: Will patch your computer to enable 2x faster free finetuning.
🦥 Unsloth Zoo will now patch everything to make training faster!
==((====))==  Unsloth 2026.1.4: Fast Gemma3 patching. Transformers: 4.57.6.
   \\   /|    NVIDIA GeForce RTX 3060. Num GPUs = 1. Max memory: 12.0 GB. Platform: Windows.
O^O/ \_/ \    Torch: 2.10.0+cu130. CUDA: 8.6. CUDA Toolkit: 13.0. Triton: 3.6.0
\        /    Bfloat16 = TRUE. FA [Xformers = 0.0.34. FA2 = False]
 "-____-"     Free license: http://github.com/unslothai/unsloth
Unsloth: Fast downloading is enabled - ignore downloading bars which are red colored!
Unsloth: Gemma3 does not support SDPA - switching to fast eager.
Unsloth: Making `model.base_model.model.model` require gradients
Unsloth: Tokenizing ["text"] (num_proc=1):   0%|                 | 0/210289 [00:00<?, ? examples/s]�  Unsloth: Will patch your computer to enable 2x faster free finetuning.
🦥 Unsloth: Will patch your computer to enable 2x faster free finetuning.

And training:

Method #2 - Docker:

Docker might be the easiest way for Windows users to get started with Unsloth as there is no setup needed or dependency issues. unsloth/unsloth is Unsloth's only Docker image. For Blackwell and 50-series GPUs, use this same image - no separate image needed.

For installation instructions, please follow our Docker guide, otherwise here is a quickstart guide:

Method #3 - WSL:

Troubleshooting / Advanced

For advanced installation instructions or if you see weird errors during installations:

1. Install torch and triton. Go to https://pytorch.org to install it. For example pip install torch torchvision torchaudio triton

2. Confirm if CUDA is installated correctly. Try nvcc. If that fails, you need to install cudatoolkit or CUDA drivers.

3. If using an Intel GPU, you will need to follow our Intel Windows guide

4. Install xformers manually. You can try installing vllm and seeing if vllm succeeds. Check if xformers succeeded with python -m xformers.info Go to https://github.com/facebookresearch/xformers. Another option is to install flash-attn for Ampere GPUs.

5. Double check that your versions of Python, CUDA, CUDNN, torch, triton, and xformers are compatible with one another. The PyTorch Compatibility Matrix may be useful.

6. Finally, install bitsandbytes and check it with python -m bitsandbytes

7. If Unsloth is not detecting or using your GPU and if you are using our Docker container on Windows, your CUDA toolkit version nvcc --version should match the version of CUDA shown by nvidia-smi on the host GPU support for Docker containers on Windows is not automatic. You need to follow Docker's guide.

Uninstall

Unsloth Desktop

1. Close Unsloth if it is running.

2. Open Settings in Windows.

3. Go to Apps → Installed apps.

4. Find Unsloth, select the three-dot menu, then choose Uninstall.

5. Follow the prompts to remove the app.

Uninstalling Unsloth Desktop does not remove downloaded models or your files.

Unsloth Studio (manual installation)

The recommended way to fully remove Unsloth Studio is the uninstall script for your OS. It stops any running servers, removes the app, CLI command, launcher data, shortcuts, and platform-specific entries (macOS .app bundle + Launch Services; Windows Start Menu + registry + PATH):

Manual uninstall

If you prefer to remove only specific parts:

1. Remove app only (keeps history, chats, checkpoints, and exports intact):

• Remove-Item -Recurse -Force "$HOME\.unsloth\studio\unsloth_studio"

2. Remove Unsloth entirely (keeps other Unsloth tools intact):

• Remove-Item -Recurse -Force "$HOME\.unsloth\studio"

3. Remove everything Unsloth-related:

• Remove-Item -Recurse -Force "$HOME\.unsloth"

4. Remove shortcuts and symlinks:

5. Remove the CLI command:

• Windows (PowerShell): The installer added the venv's Scripts directory to your User PATH. To remove it, open Settings → System → About → Advanced system settings → Environment Variables, find Path under User variables, and remove the entry pointing to .unsloth\studio\...\Scripts.

Deleting cached HF model files

You can delete old model files either from the bin icon in model search or by removing the relevant cached model folder from the default Hugging Face cache directory. By default, Hugging Face uses ~/.cache/huggingface/hub/ on macOS/Linux/WSL and C:\Users\<username>\.cache\huggingface\hub\ on Windows.

• Windows: %USERPROFILE%\.cache\huggingface\hub\

If HF_HUB_CACHE or HF_HOME is set, use that location instead. On Linux and WSL, XDG_CACHE_HOME can also change the default cache root.