[Bug] unsloth studio setup failed

[huainian6911]

I failed to install Unsloth Studio, the installation log is as follows.

GPU: RTX5060Ti

Mon Jul  6 15:54:50 2026
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 595.97                 Driver Version: 595.97         CUDA Version: 13.2     |
+-----------------------------------------+------------------------+----------------------+
| GPU  Name                  Driver-Model | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA GeForce RTX 5060 Ti   WDDM  |   00000000:01:00.0  On |                  N/A |
|  0%   40C    P3            N/A  /  N/A  |    1384MiB /  16311MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+

CUDA:

Copyright (c) 2005-2026 NVIDIA Corporation
Built on Mon_Mar__2_21:54:11_Pacific_Standard_Time_2026
Cuda compilation tools, release 13.2, V13.2.51
Build cuda_13.2.r13.2/compiler.37434383_0

Windows11 25H2

PS C:\Users\53106> irm https://unsloth.ai/install.ps1 | iex

  🦥 Unsloth Studio Installer (Windows)
  ────────────────────────────────────────────────────

  winget         available
  python         Python 3.13 already installed
  venv           creating Python 3.13 virtual environment
                 C:\Users\53106\.unsloth\studio\unsloth_studio
  gpu            NVIDIA GPU detected
                 installing PyTorch (https://download.pytorch.org/whl/cu130)...
                 installing unsloth (this may take a few minutes)...
                 PyTorch flavor mismatch (installed cpu, need cu130) -- reinstalling correct build...

  [WARN] PyTorch is CPU-only but a cu130 GPU build was expected for this machine.
  [WARN] Training and GPU inference will run on CPU until this is fixed.
  [WARN] Re-run this installer, or reinstall the GPU build manually for your GPU.
  setup          running unsloth studio setup...

  🦥 Unsloth Studio Setup
  ────────────────────────────────────────────────────
  gpu            NVIDIA GPU detected
  long paths     enabled
  git            git version 2.54.0.windows.1
  vcredist       present
  cmake          cmake version 4.3.3
  vs             Visual Studio 17 2022 (filesystem (BuildTools)) (only used if a source build is needed)
                 cl.exe: C:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\VC\Tools\MSVC\14.44.
35207\bin\Hostx64\x64\cl.exe
                 Node v24.11.0 and npm 11.6.2 already meet requirements (system).
                 Python 3.13.14

  system         prerequisites ready

  frontend       up to date
  node           v24.11.0 | npm 11.6.2 (system)
                 installing OXC validator runtime...
  oxc runtime    installed

                 setting up Python environment...
                 Python found: C:\Users\53106\.unsloth\studio\unsloth_studio\Scripts\python.exe
                 Stale venv detected (torch cpu != required cu130).
   [ERROR] The existing Studio environment needs repair.
           Re-run install.ps1 so it can replace the environment safely with rollback.
[ERROR] unsloth studio setup failed (exit code 1)

[Etherll]

Thanks for the full log - this looks like a bug. The machine is correctly selecting cu130, but Torch is ending up CPU-only and then setup rejects the venv; could you rerun with $env:UNSLOTH_VERBOSE='1'; irm https://unsloth.ai/install.ps1 | iex and paste the PyTorch reinstall part of the log? That should show why the CUDA Torch install is not sticking.

[huainian6911]

@Etherll

PS C:\Users\53106> $env:UNSLOTH_VERBOSE='1'; irm https://unsloth.ai/install.ps1 | iex

  🦥 Unsloth Studio Installer (Windows)
  ────────────────────────────────────────────────────

  winget         available
  python         Python 3.13 already installed
  venv           creating Python 3.13 virtual environment
                 C:\Users\53106\.unsloth\studio\unsloth_studio
uv : Using CPython 3.13.14 interpreter at: AppData\Local\Programs\Python\Python313\python.exe
所在位置 行:1494 字符: 45
+ ... nstallCommand { uv venv $VenvDir --python "$($DetectedPython.Path)" }
+                     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
    + CategoryInfo          : NotSpecified: (Using CPython 3...n313\python.exe:String) [], RemoteException
    + FullyQualifiedErrorId : NativeCommandError

Creating virtual environment at: .unsloth\studio\unsloth_studio
Activate with: .unsloth\studio\unsloth_studio\Scripts\activate
  gpu            NVIDIA GPU detected
                 installing PyTorch (https://download.pytorch.org/whl/cu130)...
uv : Using Python 3.13.14 environment at: .unsloth\studio\unsloth_studio
所在位置 行:2226 字符: 87
+ ...  PyTorch" { uv pip install --python $VenvPython "torch>=2.4,<2.11.0"  ...
+                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
    + CategoryInfo          : NotSpecified: (Using Python 3....\unsloth_studio:String) [], RemoteException
    + FullyQualifiedErrorId : NativeCommandError

Resolved 14 packages in 608ms
Prepared 1 package in 32ms
Installed 14 packages in 5.57s
 + filelock==3.29.7
 + fsspec==2026.6.0
 + jinja2==3.1.6
 + markupsafe==3.0.3
 + mpmath==1.3.0
 + networkx==3.6.1
 + numpy==2.5.1
 + pillow==12.3.0
 + setuptools==83.0.0
 + sympy==1.14.0
 + torch==2.10.0
 + torchaudio==2.11.0
 + torchvision==0.25.0
 + typing-extensions==4.16.0
                 installing unsloth (this may take a few minutes)...
uv : Using Python 3.13.14 environment at: .unsloth\studio\unsloth_studio
所在位置 行:2252 字符: 86
+ ...  unsloth" { uv pip install --python $VenvPython --upgrade-package uns ...
+                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
    + CategoryInfo          : NotSpecified: (Using Python 3....\unsloth_studio:String) [], RemoteException
    + FullyQualifiedErrorId : NativeCommandError

Resolved 84 packages in 1.10s
Downloading unsloth (71.3MiB)
Downloading unsloth-zoo (1.5MiB)
 Downloaded unsloth-zoo
 Downloaded unsloth
Prepared 5 packages in 2.69s
Uninstalled 1 package in 3ms
Installed 72 packages in 2.49s
 + accelerate==1.14.0
 + aiohappyeyeballs==2.7.1
 + aiohttp==3.14.1
 + aiosignal==1.4.0
 + annotated-doc==0.0.4
 + annotated-types==0.7.0
 + anyio==4.14.1
 + attrs==26.1.0
 + bitsandbytes==0.49.2
 + certifi==2026.6.17
 + charset-normalizer==3.4.9
 + click==8.4.2
 + colorama==0.4.6
 + cut-cross-entropy==25.1.1
 + datasets==4.3.0
 + diffusers==0.39.0
 + dill==0.4.0
 + docstring-parser==0.18.0
 + frozenlist==1.8.0
 - fsspec==2026.6.0
 + fsspec==2025.9.0
 + h11==0.16.0
 + hf-transfer==0.1.9
 + hf-xet==1.5.1
 + httpcore==1.0.9
 + httpx==0.28.1
 + huggingface-hub==1.22.0
 + idna==3.18
 + importlib-metadata==9.0.0
 + markdown-it-py==4.2.0
 + mdurl==0.1.2
 + msgspec==0.21.1
 + multidict==6.7.1
 + multiprocess==0.70.16
 + nest-asyncio==1.6.0
 + packaging==26.2
 + pandas==3.0.3
 + peft==0.19.1
 + propcache==0.5.2
 + protobuf==7.35.1
 + psutil==7.2.2
 + pyarrow==24.0.0
 + pydantic==2.13.4
 + pydantic-core==2.46.4
 + pygments==2.20.0
 + python-dateutil==2.9.0.post0
 + pyyaml==6.0.3
 + regex==2026.6.28
 + requests==2.34.2
 + rich==15.0.0
 + safetensors==0.8.0
 + sentencepiece==0.2.1
 + shellingham==1.5.4
 + six==1.17.0
 + tokenizers==0.22.2
 + torchao==0.17.0
 + tqdm==4.68.4
 + transformers==5.5.0
 + triton-windows==3.7.1.post27
 + trl==0.24.0
 + typeguard==4.5.2
 + typer==0.26.8
 + typing-inspection==0.4.2
 + tyro==1.0.15
 + tzdata==2026.2
 + unsloth==2026.7.1
 + unsloth-zoo==2026.7.1
 + urllib3==2.7.0
 + wheel==0.47.0
 + xformers==0.0.35
 + xxhash==3.8.1
 + yarl==1.24.2
 + zipp==4.1.0
                 PyTorch flavor mismatch (installed cpu, need cu130) -- reinstalling correct build...
uv : Using Python 3.13.14 environment at: .unsloth\studio\unsloth_studio
所在位置 行:2334 字符: 61
+ ... llCommand { uv pip install --python $VenvPython "torch>=2.4,<2.11.0"  ...
+                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
    + CategoryInfo          : NotSpecified: (Using Python 3....\unsloth_studio:String) [], RemoteException
    + FullyQualifiedErrorId : NativeCommandError

Resolved 14 packages in 94ms
Prepared 3 packages in 12ms
Uninstalled 3 packages in 1.26s
Installed 3 packages in 1.61s
 ~ torch==2.10.0
 ~ torchaudio==2.11.0
 ~ torchvision==0.25.0

  [WARN] PyTorch is CPU-only but a cu130 GPU build was expected for this machine.
  [WARN] Training and GPU inference will run on CPU until this is fixed.
  [WARN] Re-run this installer, or reinstall the GPU build manually for your GPU.
  setup          running unsloth studio setup...

  🦥 Unsloth Studio Setup
  ────────────────────────────────────────────────────
  gpu            NVIDIA GPU detected
  long paths     enabled
  git            git version 2.54.0.windows.1
  vcredist       present
  cmake          cmake version 4.3.3
  vs             Visual Studio 17 2022 (filesystem (BuildTools)) (only used if a source build is needed)
                 cl.exe: C:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\VC\Tools\MSVC\14.44.
35207\bin\Hostx64\x64\cl.exe
                 Node v24.11.0 and npm 11.6.2 already meet requirements (system).
                 Python 3.13.14

  system         prerequisites ready

  frontend       up to date
  node           v24.11.0 | npm 11.6.2 (system)
                 installing OXC validator runtime...

added 7 packages in 1s

3 packages are looking for funding
  run `npm fund` for details
node.exe : npm notice
所在位置 行:1 字符: 1
+ & "E:\Program Files\nodejs/node.exe" "C:\Users\53106\AppData\Roaming\ ...
+ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
    + CategoryInfo          : NotSpecified: (npm notice:String) [], RemoteException
    + FullyQualifiedErrorId : NativeCommandError

npm notice New minor version of npm available! 11.6.2 -> 11.18.0
npm notice Changelog: https://github.com/npm/cli/releases/tag/v11.18.0
npm notice To update run: npm install -g npm@11.18.0
npm notice
  oxc runtime    installed

                 setting up Python environment...
                 Python found: C:\Users\53106\.unsloth\studio\unsloth_studio\Scripts\python.exe
                 Stale venv detected (torch cpu != required cu130).
   [ERROR] The existing Studio environment needs repair.
           Re-run install.ps1 so it can replace the environment safely with rollback.
[ERROR] unsloth studio setup failed (exit code 1)

[alkinun]

heres my diagnosis.
installer detects cuda 13.2 correctly and expects cu130 but the later uv pip install unsloth step appears to resolve/keep default PyPI CPU torch (torch==2.10.0) instead of torch==2.10.0+cu130. setup.ps1 then correctly aborts because it sees torch cpu != required cu130.
PyTorch does publish Windows Python 3.13 cu130 wheels, so Python 3.13 itself is likely not the blocker.

Possible workaround you can try:

$env:UV_TORCH_BACKEND="cu130"
irm https://unsloth.ai/install.ps1 | iex

Likely fix is for the installer to pass the resolved torch backend into the later Unsloth install step, or fail immediately if the CUDA repair still leaves CPU torch installed.

[huainian6911]

I suddenly remembered that I had set UV_DEFAULT_INDEX in the system environment variables. After I deleted this environment variable, Unsloth Studio installed successfully.

Can the installation script be modified so that it can still install correctly when UV_DEFAULT_INDEX is set?

[alkinun]

Thanks, that explains it. This is still something the installer should handle.

I reproduced the uv behavior: if UV_DEFAULT_INDEX is set, the current installer command using --index-url https://download.pytorch.org/whl/cu130 can still resolve CPU torch==2.10.0 instead of torch==2.10.0+cu130.

Using uv's newer --default-index https://download.pytorch.org/whl/cu130 does resolve the correct CUDA wheel even when UV_DEFAULT_INDEX is set.

So the installer should likely either:

• use --default-index instead of deprecated --index-url for uv PyTorch installs, or
• temporarily clear inherited UV_DEFAULT_INDEX / UV_INDEX_URL around installer-managed uv calls.

That would make Studio install correctly even when users have a global uv mirror/default index configured.