# How to Fine-Tune LLMs on Windows with Unsloth (Step-by-Step Guide)

You can now fine-tune models directly on your local Windows device without WSL by using [Unsloth](https://github.com/unslothai/unsloth). For this guide, there are 3 main methods which you can use ([Conda](#method-1-windows-via-conda), [Docker](#method-2-docker) and [WSL](#method-3-wsl)).\
If you already have PyTorch installed on Windows, `pip install unsloth` should work. Otherwise, follow our guides below:

<a href="/pages/Sv0QKHkAGvwTK47OKX2A#method-1-windows-via-conda" class="button secondary">Conda Tutorial</a><a href="/pages/Sv0QKHkAGvwTK47OKX2A#method-2-docker" class="button secondary">Docker Tutorial</a><a href="/pages/Sv0QKHkAGvwTK47OKX2A#method-3-wsl" class="button secondary">WSL Tutorial</a>

### Unsloth Studio

We have launched a new web UI called [Unsloth Studio](/docs/new/studio/install.md) that works on Windows out of the box:

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

Use the same command to update or use `unsloth studio update`.

Then to launch every time:

```bash
unsloth studio -H 0.0.0.0 -p 8888
```

For detailed Unsloth Studio install instructions and requirements, [view our guide](/docs/new/studio/install.md).

Below are installation instructions for the original **Unsloth Core**:

### Method #1 - Windows via Conda:

{% stepper %}
{% step %}
**Install Miniconda (or Anaconda)**

Download Anaconda [here](https://www.anaconda.com/download). Our suggestion is to use [Miniconda](https://www.anaconda.com/docs/getting-started/miniconda/install#quickstart-install-instructions). To use it, first enter Powershell - search "Windows Powershell" in Start:

<figure><img src="/files/Xot07GgiCwWdtVKZZTAP" alt="" width="375"><figcaption></figcaption></figure>

Then it'll open up Powershell:

<figure><img src="/files/CxXUDcpKsOg2265REFbR" alt="" width="375"><figcaption></figcaption></figure>

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

{% code overflow="wrap" %}

```ps
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
```

{% endcode %}

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

<figure><img src="/files/z1KogGBNHT3zMCtIVHfq" alt="" width="375"><figcaption></figcaption></figure>

It's downloading the installer like below:

<figure><img src="/files/cZJT0CEY1U2PHmIwKrkL" alt="" width="375"><figcaption></figcaption></figure>

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

<figure><img src="/files/6foCIaehafkRkrwlCdwu" alt="" width="375"><figcaption></figcaption></figure>

Then you'll see:

<figure><img src="/files/9qGWVCobI7ivA3jnCl6A" alt="" width="375"><figcaption></figcaption></figure>
{% endstep %}

{% step %}
**Make conda environment**

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

**You will see:**

<figure><img src="/files/KUZgILrbLHG0Cy9meVNU" alt="" width="375"><figcaption></figcaption></figure>
{% endstep %}

{% step %}
**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](https://www.nvidia.com/en-us/drivers/).

<figure><img src="/files/C2UV1JPhXnbyZhxdT7M3" alt="" width="375"><figcaption></figcaption></figure>
{% endstep %}

{% step %}
**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](https://pytorch.org/) and matches your CUDA driver version.

{% code overflow="wrap" %}

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

{% endcode %}

You will see:

<figure><img src="/files/A4rR5H99Lvc7ShUN9ew2" alt="" width="563"><figcaption></figcaption></figure>

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

{% code overflow="wrap" %}

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

{% endcode %}

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

<figure><img src="/files/qfv5SU5Gv3x7YvPFk23X" alt="" width="375"><figcaption></figcaption></figure>
{% endstep %}

{% step %}
**Install Unsloth (only if PyTorch works!)**

{% hint style="danger" %}
**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.**
{% endhint %}

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

```bash
pip install unsloth
```

{% endstep %}

{% step %}
**Verify Unsloth works**

Now use any script in [Unsloth Notebooks](/docs/get-started/unsloth-notebooks.md) (save to .py file), or use the below basic script:

{% code expandable="true" %}

```python
from unsloth import FastLanguageModel, FastModel
import torch
from trl import SFTTrainer, SFTConfig
from datasets import load_dataset
max_seq_length = 512
url = "https://huggingface.co/datasets/laion/OIG/resolve/main/unified_chip2.jsonl"
dataset = load_dataset("json", data_files = {"train" : url}, split = "train")

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "unsloth/gemma-3-270m-it",
    max_seq_length = max_seq_length, # Choose any for long context!
    load_in_4bit = True,  # 4-bit quantization. False = 16-bit LoRA.
    load_in_8bit = False, # 8-bit quantization
    load_in_16bit = False, # 16-bit LoRA
    full_finetuning = False, # Use for full fine-tuning.
    trust_remote_code = False, # Enable to support new models
    # token = "hf_...", # use one if using gated models
)

# Do model patching and add fast LoRA weights
model = FastLanguageModel.get_peft_model(
    model,
    r = 16,
    target_modules = ["q_proj", "k_proj", "v_proj", "o_proj",
                      "gate_proj", "up_proj", "down_proj",],
    lora_alpha = 16,
    lora_dropout = 0, # Supports any, but = 0 is optimized
    bias = "none",    # Supports any, but = "none" is optimized
    # [NEW] "unsloth" uses 30% less VRAM, fits 2x larger batch sizes!
    use_gradient_checkpointing = "unsloth", # True or "unsloth" for very long context
    random_state = 3407,
    max_seq_length = max_seq_length,
    use_rslora = False,  # We support rank stabilized LoRA
    loftq_config = None, # And LoftQ
)

trainer = SFTTrainer(
    model = model,
    train_dataset = dataset,
    tokenizer = tokenizer,
    args = SFTConfig(
        max_seq_length = max_seq_length,
        per_device_train_batch_size = 2,
        gradient_accumulation_steps = 4,
        warmup_steps = 10,
        max_steps = 60,
        logging_steps = 1,
        output_dir = "outputs",
        optim = "adamw_8bit",
        seed = 3407,
        dataset_num_proc = 1,
    ),
)
trainer.train()
```

{% endcode %}

You should see:

```bash
🦥 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:

<figure><img src="/files/i6m1ueGdmV0fgRYBkaQZ" alt="" width="375"><figcaption></figcaption></figure>
{% endstep %}
{% endstepper %}

### 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`**](https://hub.docker.com/r/unsloth/unsloth) is Unsloth's only Docker image. For [Blackwell](/docs/blog/fine-tuning-llms-with-blackwell-rtx-50-series-and-unsloth.md) and 50-series GPUs, use this same image - no separate image needed.

For installation instructions, please follow our [Docker guide](/docs/blog/how-to-fine-tune-llms-with-unsloth-and-docker.md), otherwise here is a quickstart guide:

{% stepper %}
{% step %}
**Install Docker and NVIDIA Container Toolkit.**

Install Docker via [Linux](https://docs.docker.com/engine/install/) or [Desktop](https://docs.docker.com/desktop/) (other). Then install [NVIDIA Container Toolkit](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html#installation):

```bash
export NVIDIA_CONTAINER_TOOLKIT_VERSION=1.17.8-1
sudo apt-get update && sudo apt-get install -y \
  nvidia-container-toolkit=${NVIDIA_CONTAINER_TOOLKIT_VERSION} \
  nvidia-container-toolkit-base=${NVIDIA_CONTAINER_TOOLKIT_VERSION} \
  libnvidia-container-tools=${NVIDIA_CONTAINER_TOOLKIT_VERSION} \
  libnvidia-container1=${NVIDIA_CONTAINER_TOOLKIT_VERSION}
```

{% endstep %}

{% step %}
**Run the container.**

[**`unsloth/unsloth`**](https://hub.docker.com/r/unsloth/unsloth) is Unsloth's only Docker image.

```bash
docker run -d -e JUPYTER_PASSWORD="mypassword" \
  -p 8888:8888 -p 2222:22 \
  -v $(pwd)/work:/workspace/work \
  --gpus all \
  unsloth/unsloth
```

{% endstep %}

{% step %}
**Access Jupyter Lab**

Go to [http://localhost:8888](http://localhost:8888/) and open Unsloth. Access the `unsloth-notebooks` tabs to see Unsloth notebooks.
{% endstep %}

{% step %}
**Start training with Unsloth**

If you're new, follow our step-by-step [Fine-tuning Guide](/docs/get-started/fine-tuning-llms-guide.md), [RL Guide](/docs/get-started/reinforcement-learning-rl-guide.md) or just save/copy any of our premade [notebooks](/docs/get-started/unsloth-notebooks.md).
{% endstep %}

{% step %}
**Docker issues - GPU not discovered?**

Try doing WSL via [#method-2-wsl](#method-2-wsl "mention")
{% endstep %}
{% endstepper %}

### Method #3 - WSL:

{% stepper %}
{% step %}
**Install WSL**

Open up Command Prompt, the Terminal, and install Ubuntu. Set the password if asked.

```bash
wsl.exe --install Ubuntu-24.04
wsl.exe -d Ubuntu-24.04
```

{% endstep %}

{% step %} <mark style="color:$primary;background-color:orange;">**If you did NOT do (1), so you already installed WSL**</mark>**, enter WSL by typing `wsl` and ENTER in the command prompt**

```bash
wsl
```

{% endstep %}

{% step %}
**Install Python**

{% code overflow="wrap" %}

```bash
sudo apt update
sudo apt install python3 python3-full python3-pip python3-venv -y
```

{% endcode %}
{% endstep %}

{% step %}
**Install PyTorch**

{% code overflow="wrap" %}

```bash
pip install torch torchvision --force-reinstall --index-url https://download.pytorch.org/whl/cu130
```

{% endcode %}

If you encounter permission issues, use `–break-system-packages` so `pip install torch torchvision --force-reinstall --index-url https://download.pytorch.org/whl/cu130 –break-system-packages`
{% endstep %}

{% step %}
**Install Unsloth and Jupyter Notebook**

{% code overflow="wrap" %}

```bash
pip install unsloth jupyter
```

{% endcode %}

If you encounter permission issues, use `–-break-system-packages` so `pip install unsloth jupyter –-break-system-packages`
{% endstep %}

{% step %}
**Launch Unsloth via Jupyter Notebook**

{% code overflow="wrap" %}

```bash
jupyter notebook
```

{% endcode %}

Then open up our notebooks within [Unsloth Notebooks](/docs/get-started/unsloth-notebooks.md)and load them up! You can also go to Colab notebooks and download > download .ipynb and load them.

![](/files/mA0LUf7uMj3tLkxKzvw6)
{% endstep %}
{% endstepper %}

{% hint style="warning" %}
If you're using GRPO or plan to use vLLM, currently vLLM does not support Windows directly but only via WSL or Linux.
{% endhint %}

### **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](/docs/get-started/install/intel.md#windows-only-runtime-configurations)
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](https://github.com/pytorch/pytorch/blob/main/RELEASE.md#release-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](https://docs.docker.com/desktop/features/gpu/).

### Uninstall Unsloth Studio

To uninstall Unsloth Studio on Windows, follow these 4 steps:

#### **1. Remove the application**

* WSL: `rm -rf ~/.unsloth/studio/unsloth_studio`
* Windows (PowerShell): `Remove-Item -Recurse -Force "$HOME\.unsloth\studio\unsloth_studio"`&#x20;

This removes the application but keeps your model checkpoints, exports, history, cache, and chats intact.

#### **2. Remove shortcuts and symlinks**

**WSL / Windows (PowerShell):**

```bash
Remove-Item -Force "$HOME\Desktop\Unsloth Studio.lnk"
Remove-Item -Force "$env:APPDATA\Microsoft\Windows\Start Menu\Programs\Unsloth Studio.lnk"
```

#### **3. Remove the CLI command**

**WSL:**

```bash
rm -f ~/.local/bin/unsloth
```

**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`.

#### **4. Remove everything (optional)**

To also delete history, cache, chats, model checkpoints, and model exports, delete the entire Unsloth folder:

* WSL, Linux: `rm -rf ~/.unsloth`
* Windows (PowerShell): `Remove-Item -Recurse -Force "$HOME\.unsloth"`&#x20;

Note that downloaded HF model files are stored separately in the Hugging Face cache - none of the steps above will remove them. See **Deleting model files** below if you want to reclaim that disk space.


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://unsloth.ai/docs/get-started/install/windows-installation.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.