Is it possible to run turbo model on M1?

argmaxinc / WhisperKit

On-device Speech Recognition for Apple Silicon

https://takeargmax.com/blog/whisperkit

MIT License

3.17k stars 267 forks source link

Is it possible to run turbo model on M1? #149

Closed yujinqiu closed 4 months ago

yujinqiu commented 4 months ago

    #if os(macOS)
    if deviceName.hasPrefix("arm64") {
        if Process.processor.contains("Apple M1") {
            // Disable turbo variants for M1

From the code, I know turbo model is disabled on M1. Is it possible to run turbo model on M1? why can't we run the turbo model ?

atiorh commented 4 months ago

The models are JIT compiled on the first load (Core ML and WhisperKit call this phase "Specialization"). Compilation time and peak memory are prohibitively high for turbo+M1 combination. If you do try the combination, it would be great if you documented your experience here for others.