search

ggerganov / whisper.cpp

Port of OpenAI's Whisper model in C/C++
MIT License
35.38k stars 3.61k forks source link

Support for ggml .bin format models #2524

Closed dagelf closed 2 days ago

dagelf commented 2 days ago

There are a bunch of FP8 .bin models with magic lmgg on huggingface: https://huggingface.co/ggerganov/whisper.cpp/tree/main.

As an exercise in getting to understand the codebase better, I'm going to try to add support for these, directly.

Is it really as simple as just extending whisper_model_load in whisper.cpp?

Would it be acceptable to refactor it to identify the model type, and then call the appropriate load function?

Is there any information that would need to be inferred that is in not contained in these

dagelf commented 2 days ago

Nevermind...... based on a misunderstanding 🤦 Someone asked about gguf files, support for which might be considered...

PS The models linked are blazingly fast.