Commits · 4e1ec2d8833ff3a2f91443593f96cb6c178b4fc5 · OpenDAS / Fairseq

15 Jun, 2018 5 commits
- remove unused verbose option & make arguments to averaging script nicer · a3e4c4c3
  alexeib authored May 23, 2018
  
  a3e4c4c3
- ability to checkpoint when reaching certain number of updates · fc312d28
  Alexei Baevski authored May 23, 2018
  
  fc312d28
- allow specifying max_tokens for generation · 67af40c9
  Alexei Baevski authored May 15, 2018
  
  67af40c9
- Save and restore wall time in checkpoints · 0daba38e
  Myle Ott authored Apr 21, 2018
  
  0daba38e
- Simplify train.py (merge with singleprocess_train.py) · dc40ac58
  Myle Ott authored Apr 21, 2018
  
  dc40ac58
27 Feb, 2018 4 commits

Refactor incremental generation to be more explicit and less magical (#222) · 9438019f
Myle Ott authored Feb 24, 2018

9438019f
More unit test fixes · 0d90e35f
Myle Ott authored Feb 15, 2018

0d90e35f
Fix tests and flake8 · 29c82741
Myle Ott authored Feb 15, 2018

29c82741

fairseq-py goes distributed (#106) · 66415206

Myle Ott authored Feb 27, 2018

This PR includes breaking API changes to modularize fairseq-py and adds support for distributed training across multiple nodes.

Changes:
- c7033ef: add support for distributed training! See updated README for usage.
- e016299: modularize fairseq-py, adding support for register_model, register_criterion, register_optimizer, etc.
- 154e440: update LSTM implementation to use PackedSequence objects in the encoder, better following best practices and improving perf
- 90c2973 and 1da6265: improve unit test coverage

66415206

22 Jan, 2018 5 commits
- Fix max_positions calculation in train.py · 81ace092
  Myle Ott authored Jan 19, 2018
  
  81ace092
- Report log likelihood for label smoothing · dd31fa92
  Sergey Edunov authored Jan 16, 2018
  
  dd31fa92
- Add --max-sentences-valid to train.py · c542884d
  Myle Ott authored Jan 01, 2018
  
  c542884d
- Streamline data formatting utils · eb005cdb
  Myle Ott authored Jan 01, 2018
  
  eb005cdb
- Output number of model parameters in train.py · fa508492
  Myle Ott authored Dec 26, 2017
  
  fa508492
06 Dec, 2017 1 commit
- Save number of GPUs in args (and checkpoints) · 99493a85
  Myle Ott authored Dec 02, 2017
  
  99493a85
02 Dec, 2017 1 commit
- Fixed 2 typos (#75) · d74f200a
  toothlessdragon authored Dec 01, 2017
  
  d74f200a
13 Nov, 2017 1 commit
- Fallback to `--log-format=simple` for non-TTY terminals · 1b42c8c4
  Myle Ott authored Nov 12, 2017
  
  1b42c8c4
12 Nov, 2017 3 commits
- Fixes for `--log-format` · 83053f97
  Myle Ott authored Nov 11, 2017
  
  83053f97
- Fix max_positions_valid in train.py · 55a989e8
  Myle Ott authored Nov 11, 2017
  
  55a989e8
- Add `--log-format` option and JSON logger · c6d6256b
  Myle Ott authored Nov 11, 2017
  
  c6d6256b
08 Nov, 2017 7 commits
- Replace unk with original string · 42a0150c
  Louis Martin authored Nov 06, 2017
```
* Add <eos> for unk replacement
* Add IndexedRawTextDataset to load raw text files
* Replace unk with original string
* Add load_raw_text_dataset() and --output-format
* Move has_binary_files to data.py
```
  42a0150c
- Loop over evaluation dataloader in descending order · 7d44181d
  Myle Ott authored Nov 04, 2017
  
  7d44181d
- Add --max-sentence option for batching based on # sentences · f442f896
  Myle Ott authored Nov 04, 2017
  
  f442f896
- Improvements to data loader · 8f9dd964
  Myle Ott authored Oct 31, 2017
  
  8f9dd964
- Fix seed so that data is properly shuffled between epochs · 5ef59abd
  Myle Ott authored Oct 26, 2017
  
  5ef59abd
- Support different max_source_positions and max_target_positions · 2f781c5a
  Myle Ott authored Oct 25, 2017
  
  2f781c5a
- Add `--curriculum` option · 820f796f
  Myle Ott authored Oct 25, 2017
  
  820f796f
19 Oct, 2017 7 commits
- Set seed after each epoch to improve consistency when resuming · 104cead1
  Myle Ott authored Oct 19, 2017
  
  104cead1
- Prevent math overflow when loss is too high · 8b4c45a2
  Louis Martin authored Oct 19, 2017
  
  8b4c45a2
- Simplify deps of build_model to only depend on dict (instead of dataset) · 84b82dc6
  Myle Ott authored Oct 17, 2017
  
  84b82dc6
- Refactor model saving/loading to be more reusable · eea50f38
  Myle Ott authored Oct 12, 2017
  
  eea50f38
- Better logging from criterions · 8bafae2e
  Myle Ott authored Sep 28, 2017
  
  8bafae2e
- Add optimizer history to checkpoints (and rearrange criterions slightly) · e432459b
  Myle Ott authored Sep 27, 2017
  
  e432459b
- Fix flake8 warnings · cb0d7b2a
  Louis Martin authored Sep 25, 2017
  
  cb0d7b2a
11 Oct, 2017 3 commits
- Ignore invalid sentences in test and valid · 3f9b9838
  Sergey Edunov authored Oct 11, 2017
  
  3f9b9838
- Don't generate during training, add --quiet to generate.py · 8f058ea0
  Sergey Edunov authored Oct 11, 2017
  
  8f058ea0
- Update progress_bar to be more robust to changes in tqdm (#21) · 7aba6084
  Myle Ott authored Oct 11, 2017
  
  7aba6084
15 Sep, 2017 2 commits
- Architecture settings and readme updates · a15acdb0
  Sergey Edunov authored Sep 15, 2017
  
  a15acdb0
- Initial commit · e734b0fa
  Sergey Edunov authored Sep 14, 2017
  
  e734b0fa