Windows says “unknown publisher”
The early-access installer is not code-signed. Verify the SHA-256 against the official GitHub release before deciding whether to run it. Never download from an unofficial mirror. Code signing is planned for a future release.
The model does not appear
- Confirm the file ends in
.ggufand finished downloading. - Open Settings → Storage and confirm the active Models folder.
- If the file is elsewhere, select its folder or move it into the active model folder.
- A D: drive is never required on a new installation.
The model fails to load
- Lower the context size.
- Try a smaller quantization or model.
- Set GPU layers to a smaller number if VRAM is limited.
- Try CPU backend to separate a GPU problem from a model problem.
- Check that the file is not incomplete or corrupted.
CUDA is not selected on an NVIDIA GPU
Set GPU backend to Auto-detect or CUDA, reload the model, then inspect Runtime info. Update the NVIDIA display driver if CUDA is unavailable. Older GTX cards or unsupported driver/runtime combinations may use Vulkan instead. The underlying runtime recommends automatic detection and prefers CUDA when it is supported.
Vulkan is unavailable or unstable
Update the GPU driver directly from AMD, NVIDIA, or Intel. Close other GPU-heavy programs to free VRAM. If several GPUs are present, Windows may expose more than one Vulkan device; use CPU temporarily if model loading remains unstable.
A download stays at zero
- Check the Downloads tab for detailed progress or an error.
- Confirm the selected drive is connected and has free space.
- Try a public model before using a gated repository.
- For a gated repository, add a valid Hugging Face access token in Settings.
- Retry after checking the Hugging Face model page in a browser.
Where to report a bug
Use the public issue tracker. Include app version, Windows version, CPU/GPU, backend, model filename, context size, and the exact error. Remove access tokens and personal data first.