Made an 8bit quantized whisper v3 for running via https://github.com/ggerganov/whisper.cpp
Recently converted and quantized from the full float32 model to avoid any errors or loss of precision.
sample command for transcribing a podcast .wav file:
./main -m wispher-v3-8bit-ggml.bin -t 4 -f podcast69episode.wav -of podcast69transcript -otxt -nt -np
Other quants included, the k quants are experimental.
q5_1 and q8 tested and work well.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support