He Huang (Steve)

[Add] SLURP models and examples

18

The PR adds the new SOTA model we have for intent classification and slot filling for spoken language understanding. - [x] Move transformer modules from `nemo.collections.nlp.modules.common` to `nemo.collections.common.parts`, see also...

[Fix] schedulers with no max_steps param

3

All LR schedulers in PyTorch do not have the `max_steps` parameter, so we should not add `max_steps` to their `scheduler_args`. Previous code tackle the problem in case-by-case manner, while here...

move transformer from nlp to common

9

Since transformers are also used in some ASR related projects, we try to move the transformer package from nlp to common collection. Some files are left in the nlp collection...

fix canary chunk infer bug

3

# What does this PR do ? fix canary chunk infer bug, where the last segment might have mismatched shape with pre-allocated buffer.

Add SpeechLM to main

10

# What does this PR do ? Add SpeechLLM training/inference scripts to NeMo, along with dataset, model, examples and test. ### Main features - Model class for SALM-style architecture, which...

Hello, I'm trying to reproduce the speaker verification results in the WavLM paper using the ECAPA-TDNN baseline, but cannot get close to the scores in the paper. Could you please...

# What does this PR do ? Migrate SpeechLM to NeMo 2.0 **Collection**: [asr,llm,slm,motimodal] # Changelog - Add models, datamodules, training scripts