NVIDIA / NeMo

A scalable generative AI framework built for researchers and developers working on Large Language Models, Multimodal, and Speech AI (Automatic Speech Recognition and Text-to-Speech)

https://docs.nvidia.com/nemo-framework/user-guide/latest/overview.html

Apache License 2.0

11.66k stars 2.44k forks source link

fix codec/asr/mos model loading for validation #10644

Closed ehsanvidia closed 2 days ago

ehsanvidia commented 3 days ago

What does this PR do ?

When using tensor parallelism, model assumes Nemo codec/ASR/MOS checkpoints are parallelized. This PR will fix loading checkpoints when tp is enabled

Who can review?

@zhehuaichen

ehosseiniasl commented 3 days ago

for using MOS model, a container with torchaudio should be used. otherwise, comment MOS model loading and inference