Disable fp16 inference for Whisper on GTX 16xx
The GTX 16xx GPUs support CUDA but not fp16 inference. This adds a function that checks if the CUDA device contains "GTX 16xx" and disables fp16 if it does.
The GTX 16xx GPUs support CUDA but not fp16 inference. This adds a function that checks if the CUDA device contains "GTX 16xx" and disables fp16 if it does.