Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Apr 30, 2018

• feat(predict): add 'Predict' mode, inference input sentence with README description

  DongjunLee

  committed
  b6585fa

  Copy full SHA for b6585fa

• feat(r1.8): migrate codes r1.4 to r1.8

  DongjunLee

  committed
  f1b6524

  Copy full SHA for f1b6524

Commits on Mar 14, 2018

• bugfix(stacked): For the stacked layers, never get reset in loop at encoder and decoder (close #3)

  DongjunLee

  committed
  862c1ff

  Copy full SHA for 862c1ff

Commits on Jan 28, 2018

• update REAEME (add hb-research badge)

  DongjunLee

  committed
  aec8117

  Copy full SHA for aec8117

Commits on Jan 23, 2018

• update README (remove TODO's item)

  DongjunLee

  committed
  97a2289

  Copy full SHA for 97a2289

Commits on Jan 20, 2018

• refactor(model): move to eval, predict mode's graph logic to Graph class, define unified EstimatorSpec

  DongjunLee

  committed
  6e18ea6

  Copy full SHA for 6e18ea6

• refactor(hb-base): add utils (get_rev_vocab, send_message_slack) and apply to main, updated README

  DongjunLee

  committed
  d3fb58a

  Copy full SHA for d3fb58a

Commits on Dec 25, 2017

• update README (project structure - hb-base, usage - iwslt15-en-vi)

  DongjunLee

  committed
  c50cc1a

  Copy full SHA for c50cc1a

Commits on Dec 23, 2017

• set iwslt15-en-vi configs

  DongjunLee

  committed
  b891c93

  Copy full SHA for b891c93

• feat(batch_size): edit Config.model.batch_size -> batch_size by shape for dynamic

  DongjunLee

  committed
  2858ac0

  Copy full SHA for 2858ac0

• refactor: remove unused identity and print_verboase

  DongjunLee

  committed
  6d2b27c

  Copy full SHA for 6d2b27c

• feat(data): if max_seq_length existed using it, else set_max_seq_length with raw_data

  DongjunLee

  committed
  791f7c1

  Copy full SHA for 791f7c1

• feat: add sample data and decoder inputs notebook

  DongjunLee

  committed
  14de5cb

  Copy full SHA for 14de5cb

• feat(model): change output logic (remove flatted and using sequence-loss, slice and add next_token)

  DongjunLee

  committed
  0b12d45

  Copy full SHA for 0b12d45

• feat(attention): implements masked operation (blind future)

  DongjunLee

  committed
  80906a8

  Copy full SHA for 80906a8

Commits on Dec 22, 2017

• feat(data_loader): add progress bar to make_decoder_seq

  DongjunLee

  committed
  9381f14

  Copy full SHA for 9381f14

• feat(config): add iwslt15 english - vietnamese

  DongjunLee

  committed
  6f80ad2

  Copy full SHA for 6f80ad2

• feat(model): add dropout to encoded_inputs, output of sublayers

  DongjunLee

  committed
  4a4e566

  Copy full SHA for 4a4e566

Commits on Dec 20, 2017

• refactor: masked-self-attention set masked=False

  DongjunLee

  committed
  c4ddaaf

  Copy full SHA for c4ddaaf

• update README (Config, project_structure, tiny_kor_eng and tensorboard example) and images

  DongjunLee

  committed
  7e45e52

  Copy full SHA for 7e45e52

• refactor: add NotImplementedError instead of pass

  DongjunLee

  committed
  8af73ad

  Copy full SHA for 8af73ad

• feat(bleu): slice predictions after <eos> tokens

  DongjunLee

  committed
  d47ea73

  Copy full SHA for d47ea73

• feat(data): setting data_path and upload tiny_kor_eng data

  DongjunLee

  committed
  1211533

  Copy full SHA for 1211533

• feat(config): add print_verbose and debug options, and apply set_max_seq_len using make_train_and_test_set

  DongjunLee

  committed
  45bdee9

  Copy full SHA for 45bdee9

• bugfix(eval): change decoder_inputs name, using the previous one because of name_space scope

  DongjunLee

  committed
  95b0b2b

  Copy full SHA for 95b0b2b

• feat(eval): implements evaluate decoder loop, and fits sequence_loss, add _build_metric (BLEU score)

  DongjunLee

  committed
  10e1625

  Copy full SHA for 10e1625

• feat(batch_size): batch_size fixed in model config

  DongjunLee

  committed
  ac6215e

  Copy full SHA for ac6215e

Commits on Dec 19, 2017

• feat(eval): implements evaluate logic, inference output with loop [encoder_outputs, decoder_inputs (filled next_token), add reuse param to build functions

  DongjunLee

  committed
  7dffff2

  Copy full SHA for 7dffff2

• feat(test-data): test data do not need to process 'make_decoder_seq'

  DongjunLee

  committed
  7ff42f7

  Copy full SHA for 7ff42f7

• feat(multi-head): implements multi-head split -> scaled-dot-product -> concat logic

  DongjunLee

  committed
  e822c61

  Copy full SHA for e822c61

Commits on Dec 18, 2017

• remove duplicate Dataset text

  DongjunLee

  committed
  deffa5e

  Copy full SHA for deffa5e

• fixed project_structure formats and add Author

  DongjunLee

  committed
  64bc4b4

  Copy full SHA for 64bc4b4

• update README (title, project structure, experiments modes)

  DongjunLee

  committed
  b20f246

  Copy full SHA for b20f246

• refactor: merge dataset to data_loader and renaming get_dataset_batch

  DongjunLee

  committed
  d798f2f

  Copy full SHA for d798f2f

• refactor(transformer): make Encoder, Decoder class and positional_encoding move to attention

  DongjunLee

  committed
  50ba70a

  Copy full SHA for 50ba70a

Pagination

Previous

Next