Closed haozes closed 1 year ago
The watch surely doesn't meet the memory or processor requirements of these models.
Apple Watch 7 + series have 1GB+ RAM , It's enough to run whisper tiny model. and I tried it, it works. The only problem is the process, the transcribe time is too long (10 seconds wav need 6 sec on my Apple Watch Ultra) Maybe supporting ANE / CoreML can reduce the time to 2-3 seconds.
If you want to try, CoreML is supported. Though I don't think any of the watches have ANE.
If you want to try, CoreML is supported. Though I don't think any of the watches have ANE. As this document says: https://support.apple.com/en-il/HT213848
Apple Neural Engine Available for: Apple Watch Series 4 and later
Not so much information about ANE on watch, and I wil check it.
Hi, me again.
As I test by code: https://developer.apple.com/documentation/coreml/mlcomputedevice/4134993-allcomputedevices
MLComputeDevice.allComputeDevices
That's cool, print on my ultra watch, it has 1 neuralEngine , and 2 GPU.
I'll fix the watchOS build and you can feel free to try running it.
watchOS build working in latest 1.2.0 release. Feel free to try now if you want to make something with it.
Need watchOS support.
Best Regards