Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Dec 14, 2017

• Implicitly save evaluation predictions when external evaluators are set

  guillaumekln

  committed
  e9a9068

  Copy full SHA for e9a9068

Commits on Dec 1, 2017

• Also apply e4f02aa to the Transformer

  guillaumekln

  committed
  3814300

  Copy full SHA for 3814300

Commits on Nov 30, 2017

• Add a identity tensor "length" for predict (#32)

  Show description for cfbea33
  

  alucardpj

  authored and

  guillaumekln

  committed
  cfbea33

  Copy full SHA for cfbea33

• Do not average loss in time by default (#33)

  Show description for e4f02aa
  guillaumekln

  authored
  e4f02aa

  Copy full SHA for e4f02aa

• Accept GloVe embedding files by default

  guillaumekln

  committed
  23ffc1d

  Copy full SHA for 23ffc1d

Commits on Nov 29, 2017

• Add option for uniform parameters initialization

  guillaumekln

  committed
  e809585

  Copy full SHA for e809585

Commits on Nov 28, 2017

• For attention, feed context and cell output into a dense layer

  guillaumekln

  committed
  825bd18

  Copy full SHA for 825bd18

• Increase default eval_delay value

  guillaumekln

  committed
  3339bf7

  Copy full SHA for 3339bf7

• Fix inference option name

  guillaumekln

  committed
  f4bd6a3

  Copy full SHA for f4bd6a3

• Separate sampling buffer size for more effective shuffling (#30)

  guillaumekln

  authored
  dd21aa0

  Copy full SHA for dd21aa0

Commits on Nov 27, 2017

• Register new documentation file

  guillaumekln

  committed
  ab82a66

  Copy full SHA for ab82a66

• Add new opennmt/utils/evaluator.py file to the documentation

  guillaumekln

  committed
  3a719be

  Copy full SHA for 3a719be

• Support external evaluators (#29)

  guillaumekln

  authored
  219b1cb

  Copy full SHA for 219b1cb

• Fix maximum_*_length options default value

  guillaumekln

  committed
  c2a0356

  Copy full SHA for c2a0356

Commits on Nov 24, 2017

• For classification, make the encoding vector selection configurable

  guillaumekln

  committed
  99dff5f

  Copy full SHA for 99dff5f

• Do not apply label smoothing when computing evaluation loss

  guillaumekln

  committed
  b7b06f4

  Copy full SHA for b7b06f4

Commits on Nov 23, 2017

• Allow embedding size to be different than the Transformer's model size (#28)

  guillaumekln

  authored
  968642c

  Copy full SHA for 968642c

• Add merge_config script

  guillaumekln

  committed
  c09b8f0

  Copy full SHA for c09b8f0

Commits on Nov 22, 2017

• Refactor position encoders

  guillaumekln

  committed
  996a8c1

  Copy full SHA for 996a8c1

• Update documentation badge

  guillaumekln

  committed
  daba8e6

  Copy full SHA for daba8e6

• Add badge to documentation

  guillaumekln

  committed
  567c83c

  Copy full SHA for 567c83c

• Allow models to set a global initializer

  guillaumekln

  committed
  44f8b02

  Copy full SHA for 44f8b02

• Add label smoothing (#27)

  guillaumekln

  authored
  742ae9a

  Copy full SHA for 742ae9a

• Fix missing padding in convolution over characters

  guillaumekln

  committed
  cac5c17

  Copy full SHA for cac5c17

Commits on Nov 21, 2017

• Mark some features as volatile (#26)

  Show description for a6a1e99
  guillaumekln

  authored
  a6a1e99

  Copy full SHA for a6a1e99

• Add a flag to control the final RNN decoder output

  guillaumekln

  committed
  85888f7

  Copy full SHA for 85888f7

• Merge pull request #25 from guillaumekln/optimize-transformer

  Show description for 86ab75c
  guillaumekln

  authored
  86ab75c

  Copy full SHA for 86ab75c

• Cache pre-projected keys and values

  guillaumekln

  committed
  c22b06f

  Copy full SHA for c22b06f

Commits on Nov 20, 2017

• Fuse all heads computation

  guillaumekln

  committed
  8735125

  Copy full SHA for 8735125

Commits on Nov 17, 2017

• Revise dataset processing order

  Show description for 6f3fe6e
  guillaumekln

  committed
  6f3fe6e

  Copy full SHA for 6f3fe6e

Commits on Nov 16, 2017

• Fix cluster definition of chief worker

  guillaumekln

  committed
  5a4d715

  Copy full SHA for 5a4d715

Commits on Nov 15, 2017

• Fix typo in comment

  guillaumekln

  authored
  4e7d6dd

  Copy full SHA for 4e7d6dd

Commits on Nov 14, 2017

• Support pretrained embeddings files with header (#23)

  guillaumekln

  authored
  201656e

  Copy full SHA for 201656e

• Fuse RNN decoder projection during training (#22)

  guillaumekln

  authored
  528ce14

  Copy full SHA for 528ce14

Commits on Nov 13, 2017

• Expose pad_in_time function

  guillaumekln

  committed
  100fc73

  Copy full SHA for 100fc73

Pagination

Previous

Next