Commits

Commits on Feb 24, 2020

Added check to model to ensure we take off the distributed wrapper before saving
sean.narenthiran
committedFeb 24, 2020

Commits on Feb 20, 2020

Set loss-scale to 1 to prevent using dynamic scaling
sean.narenthiran
committedFeb 20, 2020

Commits on Feb 11, 2020

Merge pull request SeanNaren#482 from pskrunner14/spec_augment
Sean Naren
committedFeb 11, 2020

Commits on Feb 4, 2020

added spec_augment impl
pskrunner14
committedFeb 4, 2020

Commits on Jan 22, 2020

Merge pull request SeanNaren#479 from marekrydlewski/master
Sean Naren
committedJan 22, 2020

Commits on Jan 16, 2020

Fix gradient clipping amp compatibility, save and load amp state
marekrydlewski
committedJan 16, 2020

Commits on Oct 1, 2019

Merge pull request SeanNaren#389 from edwardyoon/master
Sean Naren
committedOct 1, 2019
Merge pull request SeanNaren#458 from SeanNaren/feature/V2
Sean Naren
committedOct 1, 2019

Commits on Sep 30, 2019

Swap to using 3-gram ARPA file
sean.narenthiran
committedSep 30, 2019

Commits on Aug 26, 2019

Fixed bug in testing script, swap to BoolTensor for PyTorch 1.2 support
sean.narenthiran
committedAug 26, 2019

Commits on Aug 15, 2019

Fix testing
sean.narenthiran
committedAug 15, 2019
Merge branch 'master' into feature/V2
sean.narenthiran
committedAug 15, 2019
Merge pull request SeanNaren#449 from SeanNaren/feature/test
Sean Naren
committedAug 15, 2019

Commits on Aug 2, 2019

Use the same evaluation function in training/testing, ensure that we remove space characters before computing CER
sean.narenthiran
committedAug 2, 2019

Commits on Aug 1, 2019

Added instructions to tune LMs, added script to select params
sean.narenthiran
committedAug 1, 2019

Commits on Jul 29, 2019

Merge pull request SeanNaren#448 from Rasmusafj/master
Sean Naren
committedJul 29, 2019
Make lookahead rely on native pytorch Conv1d
Rasmusafj
committedJul 29, 2019

Commits on Jul 25, 2019

Removed need for torchaudio, added improvements for APEX mixed precision, improved tuning for LMs, updated Docs
sean.narenthiran
committedJul 25, 2019

Commits on Jul 2, 2019

Merge pull request SeanNaren#434 from SeanNaren/docker_fix
Sean Naren
committedJul 2, 2019
Updated PyTorch/CUDNN version to fix warp-ctc
Sean Naren
committedJul 2, 2019

Commits on May 20, 2019

Merge pull request SeanNaren#418 from jybaek/fixed-best_wer
Sean Naren
committedMay 20, 2019
Fixed issue where continue_from option unconditionally overwrites best_wer.
jybaek
committedMay 20, 2019

Commits on May 17, 2019

Merge pull request SeanNaren#417 from SeanNaren/feature/mixed-load
Sean Naren
committedMay 17, 2019
Fixed saving mixed-precision based models
sean.narenthiran
committedMay 17, 2019

Commits on May 10, 2019

Merge pull request SeanNaren#413 from jybaek/fixed-indent
Sean Naren
committedMay 10, 2019
Fixed Indent
jybaek
committedMay 10, 2019

Commits on Mar 6, 2019

Fixed issue in transcribe when casting to cuda tensor
Sean Naren
committedMar 6, 2019
Merge pull request SeanNaren#396 from Jogool/transcribe
Sean Naren
committedMar 6, 2019

Commits on Feb 27, 2019

removed call to removed static methods
Joseph Szurley
committedFeb 27, 2019

Commits on Feb 20, 2019

minor option description change.
edwardyoon
committedFeb 20, 2019

Commits on Feb 7, 2019

Merge pull request SeanNaren#382 from b-nroths/fix_continue_path
Sean Naren
committedFeb 7, 2019
update dockerfile
bnroths
committedFeb 7, 2019
fix continue path references
bnroths
committedFeb 7, 2019

Commits on Feb 3, 2019

Fixed input device to transcribe function
Sean Naren
committedFeb 3, 2019

Commits on Feb 1, 2019

Removed data time sync
Sean Naren
committedFeb 1, 2019