Skip to content

Latest commit

 

History

History
187 lines (134 loc) · 7.47 KB

CHANGELOG.md

File metadata and controls

187 lines (134 loc) · 7.47 KB
Raw

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog and this project adheres to Semantic Versioning.

Unreleased

Added

  • New Neural Type System documentation. Also added decorator to generate docs for input/output ports. (PR #370) - @okuchaiev
  • New Neural Type System and its tests. (PR #307) - @okuchaiev
  • Named tensors tuple module's output for graph construction. (PR #268) - @stasbel
  • Introduced the deprecated decorator. (PR #298) - @tkornuta-nvidia
  • Implemented new mechanisms for importing and exporting of module configuration (init_params) to configuration (yml) files, along with unit tests, examples and tutorials (PR #339) - @tkornuta-nvidia
  • Speech Commands support. (PR #375) - @titu1994

Changed

  • Refactoring of nemo_nlp collections: (PR #368) - @VahidooX, @yzhang123, @ekmb

    • renaming and restructuring of files, folder, and functions in nemo_nlp
    • losses cleaned up. LossAggregatorNM moved to nemo/backends/pytorch/common/losses (PR #316) - @VahidooX, @yzhang123, @ekmb
    • renaming and restructuring of files, folder, and functions in nemo_nlp
    • Updated licenses

  • All collections changed to use New Neural Type System. (PR #307) - @okuchaiev

  • Additional Collections Repositories merged into core nemo_toolkit package. (PR #289) - @DEKHTIARJonathan

  • Refactor manifest files parsing and processing for re-using. (PR #284) - @stasbel

  • NeMo is not longer using pep8 code style rules. Code style rules are now enforced with isort and black incorporated into CI checks. (PR #286) - @stasbel

  • Major cleanup of Neural Module constructors (init), aiming at increasing the framework robustness: cleanup of NeuralModule initialization logic, refactor of trainer/actions (getting rid of local_params), fixes of several examples and unit tests, extraction and storing of intial parameters (init_params).
    (PR #309) - @tkornuta-nvidia

  • Updated nemo's use of the logging library. from nemo import logging is now the reccomended way of using the nemo logger. neural_factory.logger and all other instances of logger are now deprecated and planned for removal in the next version. Please see PR 267 for complete change information. (PR #267, PR #283, PR #305, PR #311) - @blisc

  • Changed Distributed Data Parallel from Apex to Torch (PR #336) - @blisc

  • Added TRADE (dialogue state tracking model) on MultiWOZ dataset (PR #322) - @chiphuyen, @VahidooX

  • Question answering: (PR #390) - @yzhang123

    • Changed question answering task to use Roberta and Albert as alternative backends to Bert
    • Added inference mode that does not require ground truth labels

Dependencies Update

  • Added dependency on wrapt (the new version of the deprecated warning) - @tkornuta-nvidia, @DEKHTIARJonathan

Deprecated

Fixed

  • Critical fix of the training action on CPU (PR #308) - @tkornuta-nvidia

Removed

  • gradient_predivide_factor arg of train() now has no effect (PR #336) - @blisc
  • Dropped support of the following ASR configs: jasper10x4.yaml, quartznet10x5.yaml, quartznet15x5_in.yaml, quartznet5x3.yaml, quartznet5x5.yaml, quartznet_an4.yaml. They are moved to experimental/configs and can still be used with v0.9 for use in replicating paper results (PR #354) - @blisc

Security

Contributors

0.9.0 - 2019-12-16

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 bug-fixes

0.8.2 - 2019-11-14

This is a quality improvement release for NeMo.

Highlights

  • Bugfixes
  • Support for Pytorch 1.3

0.8.1 - 2019-12-16

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

0.8.0 - 2019-12-16

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/