Georgi Gerganov

Yes - very interesting. I'll definitely take a more detailed look at the paper and see if we can add this to the project. I am very hesitant about adding...

Hi! Glad to hear there is interest in this. Will definitely write something on the topic. Most likely will focus on the various optimisations that I have used -- the...

This feature is not implemented yet. I will add sometime in the future. In the meantime, maybe someone can give it a try - the reference code is here: https://github.com/openai/whisper/blob/main/whisper/decoding.py#L18-L69

The sampling strategy currently is not perfect - I have also seen it fail in the described way. I will improve it in the future. One thing you can try...

The options is now available. For example: `-on 24` will offset the index by `24` Also, I changed the `-o` argument to `-ot` or `--offset-t`

To convert all weights to F16, simply comment the following: https://github.com/ggerganov/whisper.cpp/blob/46a68fb9b5b19e322b2c7ee21550481798f0061c/models/convert-pt-to-ggml.py#L295-L302 You then have to change all `GGML_TYPE_F32` to `wtype` in `whisper_model_load()`. However, I think there are still some operations...

Thanks for the ideas - cosmopolitan sounds like an interesting option! Will give it a thought

Hi @chidiwilliams, I'm really bad with Windows and I don't even have a machine to try on, so I won't be able to help with the build. The original idea...

@CHERTS This page says that adding the `/arch:AVX2` compiler option should already define the preprocessor symbol `__AVX2__` - that is why I decided that it is not needed to define...

Oh my god! I just tested that and it seems to work o.O I reduced the audio context by half and the performance doubled. `jfk.wav` transcribed correctly! Very interesting... I...