Commits

History for

Commits on Jul 12, 2018

Added proper cuda support for mask
sean.narenthiran
committedJul 12, 2018
Removed blacklist
sean.narenthiran
committedJul 12, 2018
Removed variables, added new masking layer, refactored training/loading of model
sean.narenthiran
committedJul 12, 2018

Commits on Jul 11, 2018

Changes to support variable length

SeanNaren
authored and
sean.narenthiran
committedJul 11, 2018

Commits on Jan 29, 2018

Removed the .tar extension from default model path
SeanNaren
committedJan 29, 2018

Commits on Jan 18, 2018

Switch cmd-line args to POSIX-style
ryanleary
committedJan 18, 2018

Commits on Jan 12, 2018

simplified batch softmax
Taras Sereda
committedJan 12, 2018

Commits on Dec 6, 2017

Added dim to softmax due to new pytorch release
SeanNaren
committedDec 6, 2017

Commits on Nov 20, 2017

Changes to support new ctcdecode package
SeanNaren
committedNov 20, 2017

Commits on Nov 10, 2017

Modified bidirectional loading for backwards compat
Sean Naren
committedNov 10, 2017
Merge pull request SeanNaren#179 from ryanleary/decoder-update
Sean Naren
committedNov 10, 2017

Commits on Nov 9, 2017

Add bi-directional check to model loading
SeanNaren
committedNov 9, 2017
Feedback changes
SeanNaren
committedNov 9, 2017
Add support for json output to transcribe
ryanleary
committedNov 9, 2017
add lookahead convolution layer for unidirectional rnn
geniki
committedNov 9, 2017

Commits on Nov 7, 2017

Update to newest decoder requirements
ryanleary
committedNov 7, 2017

Commits on Nov 6, 2017

Added separate function for package loading, modified continuing training
SeanNaren
committedNov 6, 2017

Commits on Oct 30, 2017

Merge remote-tracking branch 'origin/master'
SeanNaren
committedOct 30, 2017
Revert softmax deprecation fix till new release
SeanNaren
committedOct 30, 2017

Commits on Oct 27, 2017

Fix minor model issues
ryanleary
committedOct 27, 2017
Removed bucket sampler, use ordering of samples and batch sampler, fixed softmax deprecation error
SeanNaren
committedOct 27, 2017

Commits on Aug 30, 2017

Added flatten params
SeanNaren
committedAug 30, 2017

Commits on Jul 31, 2017

Refactor model output to reflect its type
dmitriy-serdyuk
committedJul 31, 2017

Commits on Jun 28, 2017

Add alternative softmax implementation
ryanleary
committedJun 28, 2017
Add softmax during evaluation
ryanleary
committedJun 28, 2017

Commits on Jun 12, 2017

Added parameter size count, various refactor of main training class
SeanNaren
committedJun 12, 2017
Small refactors in the model class
SeanNaren
committedJun 12, 2017

Commits on May 9, 2017

Return the current validation error/loss (SeanNaren#57 )

ryanleary
authored and
Sean Naren
committedMay 9, 2017

Commits on May 8, 2017

Refactor model serialization & deserialization for CPU/GPU portability (SeanNaren#54 )

ryanleary
authored and
Sean Naren
committedMay 8, 2017

Commits on May 6, 2017

Persist metadata, use in prediction, various fixes
SeanNaren
committedMay 6, 2017

Commits on May 5, 2017

Fix model saving for different RNN types
Ryan Leary
committedMay 5, 2017

Commits on Apr 27, 2017

Merge branch 'master' into rnn_choice
Sean Naren
committedApr 27, 2017
Refactored multiple RNN support, removed added skip_input linear layers as function has not been completed
SeanNaren
committedApr 27, 2017
passing only rnn class to the deepspeech module
EgorLakomkin
committedApr 27, 2017

Commits on Apr 26, 2017

added support of simple_rnn, gru and lstm for the deepspeech module
EgorLakomkin
committedApr 26, 2017