Currently I use the 'Metal' implementation and it's very fast on M1-4 Macs and most iOS devices, but on Intel Macs it falls back to CPU and is unusably slow.

1. What is the recommended ways of running whisper on older Intel Macs ?

2. Is there a 'runtime way' to find out whether the model will be run on CPU or GPU on a concrete device ? I am seeing iPhones which provide A11bionic, but at model initialisation they fall back to CPU. It would be interesting to get this information (beside a log notice) to issue a warning. How can I access that information ?

Any help is appreciated ! Thanks.