Issues with the latest Whisper version v20240930

[workingwheel]

When I run instances locally on my 4090, I keep getting this error and it reverts back to using the CPU.

.venv\Lib\site-packages\whisper\model.py:124: UserWarning: 1Torch was not compiled with flash attention. (Triggered internally at C:\actions-runner_work\pytorch\pytorch\builder\windows\pytorch\aten\src\ATen\native\transformers\cuda\sdp_utils.cpp:555.) a = scaled_dot_product_attention(

I had to switch back to the older version: v20240927 from the latest v20240930
running on Cuda 12.4. I also have 11.8 and 12.6. All the latest drivers and modules. Python 3.12.

After reverting back to the previous version, I no longer get that error, it uses the GPU like it is supposed to. Just figure I share this in case anyone else is having problems.