_scottas

@yoyolicoris @WindsorNguyen @PyTorch PyTorch's C++ API is definitely here to stay! This is only about what's best for the longterm health for torchaudio.

@WindsorNguyen @PyTorch I'm afraid the hidden message is PyTorch is gonna drop the C++ API completely...

github.com/pytorch/audio/… Torchaudio just announced it will be pure Python again. That means dropping efficient kernels like filter, RNN-Tranducer, etc. It's an unwise and disruptive decision tbh. If your work will be affected by this, please leave a comment there... @PyTorch

@yoyolicoris @unilightwf @PyTorch Issue poster here! All of the functionality critical functionality from torchaudio.io will be available in torchcodec: github.com/pytorch/torchc…. It just had a new release, with audio decoding! Encoding is being actively worked on.

@unilightwf @PyTorch Yeah I forgot to mention! Removing the whole io.* is crazy. Remember there used to be a benchmark that shows audio loading speed, and torchaudio performs quite well. Now years of efforts are gonna go to dust 😰

@keunwoochoi In our experience pre-training LLMs at scale, exact resume is a hard requirement. Have you looked into torchdata.nodes to see if it fits your needs? github.com/pytorch/data/b…