This repository has been archived by the owner on Feb 12, 2022. It is now read-only.

Commits

Commits on Feb 12, 2022

Updated/Added CODEOWNERS with ECCN
svc-scm
committedFeb 12, 2022

Commits on Jun 19, 2018

Updating some typos in README
keskarnitish
committedJun 19, 2018

Commits on Jun 13, 2018

removing optional dependency
keskarnitish
committedJun 13, 2018
Fixing name typo
keskarnitish
committedJun 13, 2018
PyTorch 0.4 compatible
Nitish Keskar
committedJun 13, 2018

Commits on Apr 25, 2018

Merge pull request #31 from huggingface/master
keskarnitish
committedApr 25, 2018

Commits on Apr 9, 2018

Fixing bugs in the typed out WT103 reproducibility command
Smerity
committedApr 9, 2018

Commits on Apr 3, 2018

Bugfix: parameters of both the model as well as the criterion (Adaptive Softmax) must be optimized
Smerity
committedApr 3, 2018

Commits on Mar 29, 2018

rm duplicate
julien-c
committedMar 29, 2018

Commits on Mar 23, 2018

Merge pull request #25 from Smerity/master
Smerity
committedMar 23, 2018
Update README.md
Smerity
committedMar 23, 2018

Commits on Mar 20, 2018

Bugfix: evaluation did not use adaptive softmax criterion correctly
Smerity
committedMar 20, 2018

Commits on Mar 19, 2018

Main: add tied adaptive softmax, Minor: on resume, set hyperparams
Smerity
committedMar 19, 2018
Script to process and prepare enwik8 for dataset use
Smerity
committedMar 19, 2018

Commits on Mar 18, 2018

--when to trigger lr/10 and full state model save/load
Smerity
committedMar 18, 2018
Update dataset collection script (WT2, WT103, enwik8, PTB, PTBC)
Smerity
committedMar 18, 2018

Commits on Mar 14, 2018

Cache data, resume models, SGD + ADAM, no clip/alpha/beta overhead, add BPC
Smerity
committedMar 14, 2018

Commits on Dec 12, 2017

Merge pull request #16 from salesforce/keskarnitish-patch-1
Smerity
committedDec 12, 2017
Fixing the PTB data set 404 issue
keskarnitish
committedDec 12, 2017

Commits on Nov 26, 2017

Generate was broken for QRNN as QRNN didn't flush their previously stored X, resulting in diff batch sizes, hence angry concat dimensions
Smerity
committedNov 26, 2017

Commits on Nov 6, 2017

Fix: Previous commit added layer norm QRNN accidentally
Smerity
committedNov 6, 2017

Commits on Nov 5, 2017

Allow for untied weights with different embedding and softmax sizes
Smerity
committedNov 5, 2017

Commits on Oct 23, 2017

Add support for GRU as base RNN type
Smerity
committedOct 23, 2017

Commits on Oct 10, 2017

Silly line of code fix: True if x else True
Smerity
committedOct 10, 2017

Commits on Oct 9, 2017

Merge pull request #10 from Smerity/master
Smerity
committedOct 9, 2017
Update README to have WT2 QRNN results
Smerity
committedOct 9, 2017

Commits on Oct 5, 2017

Fixing minor QRNN issues in README
Smerity
committedOct 5, 2017
Update README to describe training QRNN model
Smerity
committedOct 5, 2017
Add QRNN as an optional RNN model to use - multiple times faster than cuDNN LSTM with same or better results
Smerity
committedOct 5, 2017

Commits on Oct 3, 2017

Merge pull request #9 from keskarnitish/master
Smerity
committedOct 3, 2017
New Hyperparameters
keskarnitish
committedOct 3, 2017

Commits on Sep 27, 2017

Fix default data paths for and
Smerity
committedSep 27, 2017

Commits on Sep 22, 2017

Numpy random seed should be changed with Torch CPU and GPU random seed
Smerity
committedSep 22, 2017
Fix the double dropout bug
Smerity
committedSep 22, 2017

Commits on Sep 7, 2017

Update README to feature PyTorch 0.2
Smerity
committedSep 7, 2017