NVIDIA Neural Modules 1.0.0b1

Release 1.0.0b1

This release is a major re-design compared to previous version.
All NeMo models and modules are now compatible out-of-the box with Pytorch and Pytorch Lightning.
Every NeMo model is a LightningModule that comes equipped with all supporting infrastructure for training and reproducibility. Every NeMo model has an example configuration file and a corresponding script that contains all configurations needed for training. NeMo, Pytorch Lightning, and Hydra makes all NeMo models have the same look and feel so that it is easy to do Conversational AI research across multiple domains. New models such as Speaker verification and Megatron are added.

Highlights

• Pytorch Lightning based Core
• Hydra and Omegaconf configuration management
• All model's files tarred together as .nemo files make it easy for users to download models automatically from NGC
• NGC collections now includes a collection of all NeMo assets in one
• New Models & tutorials
  • ASR: SpeakerNet speaker verification model
  • NLP: Bio Megatron state of the art model trained on bio medical tasks
• ASR, NLP and TTS tutorials as interactive notebooks

Known Issues

Toolkit in an early version software. Breaking changes compared to previous version.

Resolved Issues

All models and modules can be used anywhere torch.nn.Module is expected.

NVIDIA Neural Modules v0.11.0

Release 0.11.0

This release improves ease of use and adds new features

Highlights

• Neural Graphs and NeMo models for ASR
• New models:
  • Voice activity detection
  • Speaker identification
  • Matchboxnet Speech commands
  • Megatron BERT trained on bio medical data
• Various improvements and bugfixes

NVIDIA Neural Modules v0.10.1

Release 0.10.1

This is a bug fix release

Highlights:

• Fixes an issue when distributed training would halt if Loss is NaN instead of skipping the batch
• Other minor fixes

NVIDIA Neural Modules v0.10.0

Release 0.10.0

This release improves overall stability of NeMo, revamps type system and adds new models

Highlights:

• Switch to a single package install (“nemo_toolkit”) with on-demand collections
• New, easier-to-use and more flexible neural type system
• Unified neural module import/export to configuration files
• Various bugfixes fixes and improvements
  • Codebase switched to black formatter
• New models:
  • Speech command recognition
  • TRADE dialog state tracking
  • Question answering
  • Roberta and Albert
  • FastSpeech

NVIDIA Neural Modules v0.9.0

Release 0.9.0

This release contains new features, new models and quality improvements for NeMo.

Highlights:

• Added "nemo_tts" - a Speech Synthesis collection with necessary modules for Tacotron2 and WaveGlow
• Added Mandarin support into nemo_asr and nemo_nlp
• Updated ASR and TTS checkpoints including Mandarin ASR
• Documentation now translated to Mandarin https://nvidia.github.io/NeMo/chinese/intro.html
• Export functionality for deployment
• General improvements and bugfixes

NVIDIA Neural Modules v0.8.2

Release 0.8.2

This is a quality improvement release for NeMo.

Highlights:

• Bugfixes
• Support for Pytorch 1.3

NVIDIA Neural Modules v0.8.1

Release 0.8.1

This is a quality improvement release for NeMo.

Highlights:

• Added introductory ASR tutorial explaining how to get started with deep learning for ASR
• Re-organization of NeMo NLP library
• More efficient BERT pre-training implementation
• General improvements and bugfixes
• Support for CPU-only scenario

Special thanks to our external contributors:

David Pollack (dhpollack), Harisankar Haridas (harisankarh), Dilshod Tadjibaev (antimora)

NVIDIA Neural Modules v0.8

The first public release of NVIDIA Neural Modules: NeMo.

This release also includes nemo_asr'' and nemo_nlp'' collections for Speech Recognition and Natural Language Processing.

Please refer to the documentation here: https://nvidia.github.io/NeMo/