Is there any way to use my AMD GPU for whisper AI model?

[AloofLibra]

Is there any way to use my AMD GPU for whisper AI model?
It seems like now it's using my CPU which is not very fast, so transcript generating process is slow and inaccurate.
Or maybe any tips on how to make transcript more accurate at least?

[elbarroca]

Yes—on Windows, the current source supports AMD Whisper acceleration through Vulkan.

From frontend/, the relevant pieces are:

• frontend/src-tauri/Cargo.toml: vulkan = ["whisper-rs/vulkan"]
• frontend/package.json: pnpm run tauri:build:vulkan
• frontend/build-gpu.ps1: the Windows helper that invokes that Vulkan build

For a source build, try:

cd frontend
pnpm install --frozen-lockfile
pnpm run tauri:build:vulkan

Use a current AMD driver with Vulkan support. The repo documents HIP/ROCm for Linux AMD; Windows uses Vulkan. If it still falls back to CPU, check the release/build version and the GPU auto-detection prerequisites (VULKAN_SDK and BLAS_INCLUDE_DIRS), because the detection script can fall back when they are absent. GPU acceleration should improve speed, not transcription accuracy.

References: Cargo features, build scripts, GPU detection