Commits

Branch selector

 9a95121

User selector

All users

Datepicker

All time

Commit History

Commits on Feb 9, 2018

• Adjust weight decay by the current learning rate to make it work correctly during annealing

  

  Sergey Edunov

  committed
  9a95121

  Copy full SHA for 9a95121

Commits on Jan 31, 2018

• Merge pull request facebookresearch#91 from facebookresearch/prepare_wmt

  Show description for e4c935a
  edunov

  authored
  e4c935a

  Copy full SHA for e4c935a

• spelling

  

  Sergey Edunov

  committed
  52b6119

  Copy full SHA for 52b6119

• Update README with new models

  

  Sergey Edunov

  committed
  2c18c27

  Copy full SHA for 2c18c27

• Merge pull request facebookresearch#95 from bastings/patch-1

  Show description for fb36614
  edunov

  authored
  fb36614

  Copy full SHA for fb36614

• Adding README and more parameters to En2De script

  

  Sergey Edunov

  committed
  971c2d6

  Copy full SHA for 971c2d6

Commits on Jan 29, 2018

• Ratio should be predlen/reflen not reflen/predlen

  Show description for 1ff3efc
  

  Joost Bastings

  authored
  1ff3efc

  Copy full SHA for 1ff3efc

Commits on Jan 27, 2018

• Merge branch 'master' of github.com:facebookresearch/fairseq-py into prepare_wmt

  

  Sergey Edunov

  committed
  d9f46c5

  Copy full SHA for d9f46c5

• Switch to news-commentary-v12

  

  Sergey Edunov

  committed
  4185d3e

  Copy full SHA for 4185d3e

Commits on Jan 22, 2018

• Fixed Weight Decay Regularization in Adam

  Show description for ee36a6f
  

  michaelauli

  authored and

  myleott

  committed
  ee36a6f

  Copy full SHA for ee36a6f

• Fix tests

  myleott

  committed
  66d9fcf

  Copy full SHA for 66d9fcf

• Output correct perplexity when training with --sentence-avg

  myleott

  committed
  f9362e8

  Copy full SHA for f9362e8

• Fix max_positions calculation in train.py

  myleott

  committed
  81ace09

  Copy full SHA for 81ace09

• Better warning message for inputs that are too long

  myleott

  committed
  3346943

  Copy full SHA for 3346943

• ATen Fix

  

  Michael Auli

  authored and

  myleott

  committed
  66314a6

  Copy full SHA for 66314a6

• Momentum correction

  

  Michael Auli

  authored and

  myleott

  committed
  173c577

  Copy full SHA for 173c577

• Report log likelihood for label smoothing

  

  edunov

  authored and

  myleott

  committed
  dd31fa9

  Copy full SHA for dd31fa9

• Share input/output embed

  

  edunov

  authored and

  myleott

  committed
  c537860

  Copy full SHA for c537860

• Better support for torch.no_grad (since volatile is deprecated)

  myleott

  committed
  907ca92

  Copy full SHA for 907ca92

• Fix training

  myleott

  committed
  0b84ab1

  Copy full SHA for 0b84ab1

• Save dictionary in model base classes

  myleott

  committed
  5eddda8

  Copy full SHA for 5eddda8

• Fix gradient clipping when --clip-norm=0

  myleott

  committed
  08a74a3

  Copy full SHA for 08a74a3

• Fix LearnedPositionalEmbedding

  myleott

  committed
  fd28c88

  Copy full SHA for fd28c88

• Move normalization of model output (e.g., via LSM) into model definition

  myleott

  committed
  4db6579

  Copy full SHA for 4db6579

• Move positional embeddings into LearnedPositionalEmbedding module

  myleott

  committed
  c21a6e2

  Copy full SHA for c21a6e2

• Fix warning about deprecated volatile kwarg for Variables

  myleott

  committed
  185a0df

  Copy full SHA for 185a0df

• Add option to SequenceGenerator to retain dropout

  myleott

  committed
  dccf790

  Copy full SHA for dccf790

• Add --max-sentences-valid to train.py

  myleott

  committed
  c542884

  Copy full SHA for c542884

• Streamline data formatting utils

  myleott

  committed
  eb005cd

  Copy full SHA for eb005cd

• Add reduce kwarg to criterions

  myleott

  committed
  6f6cb4a

  Copy full SHA for 6f6cb4a

• Raise FileNotFoundError if dictionary files don't exist

  myleott

  committed
  dcbf5e7

  Copy full SHA for dcbf5e7

• Output number of model parameters in train.py

  myleott

  committed
  fa50849

  Copy full SHA for fa50849

• Add explicit dimension to softmax calls

  myleott

  committed
  18a6d85

  Copy full SHA for 18a6d85

• Support deprecation of volatile Variables in latest PyTorch

  myleott

  committed
  7da4e06

  Copy full SHA for 7da4e06

• Minor fix for strip_pad functions

  myleott

  committed
  5637d54

  Copy full SHA for 5637d54

Pagination

Previous

Next