Jeremy Howard#90
what a wonderful project: parakeet.cpp
https://github.com/mudler/parakeet.cpp
GGML based parakeet inference pipeline that's 2x faster than my ONNX parakeet pipeline on Apple Silicon! (Needed a few local patches to get it going)
No Digg Deeper questions have been answered for this story yet.
thanks @mudler_it great stuff!
@badlogicgames nice, just benched against mine https://github.com/hyperpuncher/chough (uses sherpa-onnx)
really fast with short files, but get's slower with longer duration
15sec: chough - 2s (1s if using remote) parakeet.cpp - 1s
5min: chough - 16s 1.8gb (lower mem using chunking) parakeet.cpp - 43s 5.8gb
@badlogicgames Thank you! really appreciated coming from you!
@badlogicgames 2x faster on apple silicon is actually wild for a ggml pipeline
what patches were the trickiest to get right?
@badlogicgames c++ 母语者开心的哭了
@badlogicgames is it for the robot you're working on? Parakeet models are great value , no doubt about that
@hyper_puncher ahh good to know!
@mudler_it @badlogicgames does it matter if on macos or windows?