whisper working

Got it — if Option 2 didn’t work, we’ll go with Option 1 and reinstall PyTorch in CPU-only mode so Whisper stops throwing the could not create a primitive error.

Here’s exactly what you should run inside your venv:

pip uninstall torch -y
pip cache purge
pip install torch==2.4.1+cpu torchvision==0.19.1+cpu torchaudio==2.4.1+cpu --index-url https://download.pytorch.org/whl/cpu

Then test Whisper again with:

whisper fung.ogg --model tiny --language zh --device cpu --fp16 False

This will:

1. Remove the old Torch build that was trying to use oneDNN/MKL.
2. Install a clean CPU-only build that works fine on Windows without special CPU instructions.
3. Use the tiny model so it downloads quickly and runs faster.

Do you want me to also tweak your transcribe_zh.bat so it always uses this tiny model for Chinese and avoids FP16 automatically?