[Bug] Windows 'https://unsloth.ai/install.ps1 '

[runxc1]

Note: Please do not remove the questions. Answer beside them.

1. Did you update? nope this was a clean install on Windows 11, 5080 GPU that failed. Seems it had a lot of install prompts as well, not a smooth experience at all. Below it the logs

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

             installing Python 3.13...

Found Python 3.13 [Python.Python.3.13] Version 3.13.13
This application is licensed to you by its owner.
Microsoft is not responsible for, nor does it grant any licenses to, third-party packages.
Downloading https://www.python.org/ftp/python/3.13.13/python-3.13.13-amd64.exe
██████████████████████████████ 27.7 MB / 27.7 MB
Successfully verified installer hash
Starting package install...
Successfully installed
venv creating Python 3.13 virtual environment
C:\Users\whodis.unsloth\studio\unsloth_studio
gpu NVIDIA GPU detected
installing PyTorch (https://download.pytorch.org/whl/cu130)...
installing unsloth (this may take a few minutes)...
setup running unsloth studio setup...

🦥 Unsloth Studio Setup
────────────────────────────────────────────────────
gpu NVIDIA GPU detected
Windows Long Paths not enabled (required for Triton compilation and deep dependency paths).
Requesting admin access to fix...
long paths enabled (via UAC)
git git version 2.50.1.windows.1
CMake not found -- installing via winget...
cmake installed
vs Visual Studio 17 2022 (vswhere)
driver supports up to CUDA 13.0
GPU Compute Capability = 12.0 (sm_120)
CUDA toolkit (nvcc) not found -- installing via winget...
Installing CUDA Toolkit 13.0 via winget...
CUDA Toolkit 13.0 installed (nvcc: C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.0\bin\nvcc.exe)
Persisted CUDA_PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.0 to user environment
Set CUDA_PATH_V13_0 (cleared other CUDA_PATH_V* vars)
Persisted CUDA bin dir to user PATH
cuda C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.0\bin\nvcc.exe
CUDA_PATH = C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.0
CudaToolkitDir = C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.0
Node v22.17.1 / npm 10.9.2 too old.
installing Node.js LTS via winget...
Found Node.js (LTS) [OpenJS.NodeJS.LTS] Version 24.15.0
This application is licensed to you by its owner.
Microsoft is not responsible for, nor does it grant any licenses to, third-party packages.
Downloading https://nodejs.org/dist/v24.15.0/node-v24.15.0-x64.msi
██████████████████████████████ 30.9 MB / 30.9 MB
Successfully verified installer hash
Starting package install...
The installer will request to run as administrator. Expect a prompt.
Successfully installed
node v22.17.1 | npm 10.9.2
installing bun (faster frontend package installs)...
bun installed (1.3.12)
Python Python 3.13.13

system prerequisites ready

frontend up to date
installing OXC validator runtime...
oxc runtime installed

             setting up Python environment...
             Using C:\Users\whodis\AppData\Local\Programs\Python\Python313\python.exe (Python 3.13.13)
             reusing existing virtual environment at C:\Users\whodis\.unsloth\studio\unsloth_studio
             TORCHINDUCTOR_CACHE_DIR set to C:\tc (avoids MAX_PATH issues)
             installing PyTorch with CUDA support (cu130)...
             (This download is ~2.8 GB -- may take a few minutes)
             installing Triton for Windows...
             Triton for Windows installed (enables torch.compile)
             running ordered dependency installation...

deps [====================] 9/9 finalizing
deps installed

             pre-installing transformers 5.3.0 for newer model support...

transformers 5.3.0 pre-installed
pre-installing transformers 5.5.0 for Gemma 4 support...
transformers 5.5.0 pre-installed

             installing prebuilt llama.cpp bundle (preferred path)...

llama.cpp prebuilt install failed (continuing)
| python.exe : [llama-prebuilt] no existing llama.cpp install detected at C:\Users\whodis.unsloth\llama.cpp; performing
| fresh prebuilt install
| At C:\Users\whodis.unsloth\studio\unsloth_studio\Lib\site-packages\studio\setup.ps1:1933 char:35
| + $prebuiltOutput = & python @prebuiltArgs 2>&1 | Out-S ...
| + ~~~~~~~~~~~~~~~~~~~~~~~~~~~
| + CategoryInfo : NotSpecified: ([llama-prebuilt...rebuilt install:String) [], RemoteException
| + FullyQualifiedErrorId : NativeCommandError
| [llama-prebuilt] prebuilt install path failed; falling back to source build
| [llama-prebuilt] prebuilt fallback reason: failed to inspect published releases in ggml-org/llama.cpp: [WinError 5]
| Access is denied: 'C:\windows\system32\config\systemprofile\AppData\Local\Microsoft\WindowsApps'
| [llama-prebuilt] fatal helper error: [WinError 5] Access is denied:
| 'C:\windows\system32\config\systemprofile\AppData\Local\Microsoft\WindowsApps'
Prebuilt llama.cpp path unavailable or failed validation -- falling back to source build

             installing OpenSSL dev (for HTTPS in llama-server)...

Found OpenSSL [ShiningLight.OpenSSL.Dev] Version 3.6.2
This application is licensed to you by its owner.
Microsoft is not responsible for, nor does it grant any licenses to, third-party packages.
This package requires the following dependencies:

• Packages
  Microsoft.VCRedist.2015+.x64
  Downloading https://slproweb.com/download/Win64OpenSSL-3_6_2.msi
  ██████████████████████████████ 248 MB / 248 MB
  Successfully verified installer hash
  Starting package install...
  Successfully installed
  OpenSSL dev installed at C:\Program Files\OpenSSL-Win64

           building llama.cpp with CUDA support...
           This typically takes 5-10 minutes on first build.
  

Cloning llama.cpp @ b8816...

--- cmake configure ---
| -- The C compiler identification is unknown
| -- The CXX compiler identification is unknown
| cmake : CMake Error at CMakeLists.txt:2 (project):
| At C:\Users\whodis.unsloth\studio\unsloth_studio\Lib\site-packages\studio\setup.ps1:2380 char:24
| + $cmakeOutput = cmake @CmakeArgs 2>&1 | Out-String
| + ~~~~~~~~~~~~~~~~~~~~~
| + CategoryInfo : NotSpecified: (CMake Error at ...xt:2 (project)::String) [], RemoteException
| + FullyQualifiedErrorId : NativeCommandError
| No CMAKE_C_COMPILER could be found.
| CMake Error at CMakeLists.txt:2 (project):
| No CMAKE_CXX_COMPILER could be found.
| -- Configuring incomplete, errors occurred!
llama.cpp build failed at: cmake configure (0m 21.6s); continuing
To retry: delete C:\Users\whodis.unsloth\llama.cpp and re-run setup.
────────────────────────────────────────────────────
Unsloth Studio Setup Complete (limited: llama.cpp unavailable)
────────────────────────────────────────────────────
launch unsloth studio -H 0.0.0.0 -p 8888

[ERROR] unsloth studio setup failed (exit code 1)

[rolandtannous]

@LeoBorcherding also investigate this on windows. Get more information, try to reproduce. We need to understand why this doesn't happen with everyone.,

[LeoBorcherding]

Hello @runxc1, thank you for the detailed logs! To help us reproduce and investigate this, could you answer a few quick questions?

Did you run PowerShell as admin, or as a regular user? The installer requested UAC elevation mid-install for long paths, which can affect how some paths are resolved.

Which Visual Studio edition do you have installed, full VS 2022 or just Build Tools? And do you have the Desktop development with
C++ workload installed? VS was detected but cmake couldn't find any C/CXX compiler, which usually points to the C++ workload being missing.

Had you run the installer before this attempt? The log shows "reusing existing virtual environment" which suggests a previous partial install.

Was this a normal powershell?

The answers to these questions will help me reproduce the issue faster, thank you!

[runxc1]

Ran it as an admin. I did try twice. I have VS 2026 and an older VS 2022. Neither have C++ components but I may have installed something else that required it. I would say the sheer number of things it tried to have me install was jarring, probably wouldn't try again due to that.

[LeoBorcherding]

Thanks for the details @runxc1. The most likely culprit based on your logs is the missing C++ compiler workload in Visual Studio, but I'll need to reproduce and test the issue to confirm. If you're ever willing to retry, deleting C:\Users\whodis.unsloth\llama.cpp and re-running setup after adding the Desktop development with C++ workload in the VS Installer would be the thing to try. I'll let you know if I can reproduce your issue.

Appreciate the feedback on the install experience, will see what we can do to reduce the number of steps during install, however all of those packages are required. Soon there will also be an executable installer that is more simple.

[danielhanchen]

This should be fixed now - sorry on the delay - please update Unsloth Studio to the latest via curl -fsSL https://unsloth.ai/install.sh | sh for Linux, Mac, WSL or irm https://unsloth.ai/install.ps1 | iex for Windows - if it's not fixed yet, it'll be in the next Studio release - if it's still not working after that, please open another issue - appreciate it again!

The Windows installer was rewritten to use prebuilt binaries; re-run install.ps1 from unsloth.ai.