Core ML model does not seem to be working in iOS app

[ahsanzia341]

I was able to integrate the coreml model in the app and the log says, "Core ML Model Loaded" (please see the attached screenshot for reference) but when I am running the transcription using the function whisper_full, I am not seeing any difference in speed as I am seeing on my terminal. I have transcribed for a 40 secs audio. Below are the runs I gave for the transcriptions.

This screenshot is when the Core ML Model got loaded.

Any help would be appreciated.

[ggerganov]

How much is the encode time, ms per run without WHISPER_COREML? Check only the first run, as in this screenshot:

[ahsanzia341]

Ok I just checked the difference in speed is there. It is taking 160 secs to encode this audio without coreml in the app but when I ran the same audio in the terminal, the difference is far greater. So, I want to know why my app is not performing as such. Here are the results.

First run:

Second run:

[yousifa]

How did you enable coreml? Is it a build flag?

[ahsanzia341]

How did you enable coreml? Is it a build flag?

Yes, I had to add it in the .h file to get it working in the iOS app.

[yousifa]

How did you enable coreml? Is it a build flag?

Yes, I had to add it in the .h file to get it working in the iOS app.

Yep, figured it out. Have to include the files under /coreml in the app as well!

[andforce]

@ahsanzia341 Hi, could you show some code that how used CorlML on iOS? thanks very munch~