1580 Commits

This Branch

This Branch

All Branches

Author	SHA1	Message	Date
Myle Ott	3f9700868d	More flexible gradient normalization	2017-10-19 08:12:44 -07:00
Myle Ott	88a8bd42c8	Fix smoothed (sentence-level) BLEU calculation	2017-10-19 08:12:41 -07:00
Louis Martin	d92ce54c65	Ignore generated files for temporal convolution tbc	2017-10-19 08:12:39 -07:00
Louis Martin	84754894b9	Add attention matrix to output of SequenceGenerator	2017-10-19 08:12:27 -07:00
Myle Ott	376c265f35	Add support for NCCL v2	2017-10-19 08:12:24 -07:00
Myle Ott	8bafae2ee7	Better logging from criterions	2017-10-19 08:12:20 -07:00
Myle Ott	e432459b37	Add optimizer history to checkpoints (and rearrange criterions slightly)	2017-10-19 08:11:27 -07:00
Myle Ott	48631f7a3c	Allow --max-len-a to be a float	2017-10-19 08:06:43 -07:00
Myle Ott	813352e12c	Don't save/restore convolutional layers in incremental inference	2017-10-19 08:06:40 -07:00
Louis Martin	cb0d7b2ad1	Fix flake8 warnings	2017-10-19 08:06:27 -07:00
Myle Ott	7333d04d3f	Support configurable BPE symbol	2017-10-19 08:06:24 -07:00
Myle Ott	59d599a277	Move helper functions from generate.py to fairseq/dictionary.py	2017-10-19 08:06:20 -07:00
Sergey Edunov	af86c1ac3d	Update En2Fr model	2017-10-17 11:46:04 -07:00
Myle Ott	2b482f638a	Fix --no-progress-bar option in generate.py (#115)	2017-10-14 20:40:24 -07:00
Myle Ott	9c5fa5c544	Update README.md ... Don't suggest Miniconda (see #24)	2017-10-13 09:44:14 -04:00
Myle Ott	0ccd81d407	Update README.md ... Update PyTorch install instructions	2017-10-12 22:04:33 -04:00
Sergey Edunov	3f9b9838c3	Ignore invalid sentences in test and valid	2017-10-11 16:12:59 -07:00
Sergey Edunov	8f058ea0fb	Don't generate during training, add --quiet to generate.py	2017-10-11 16:11:28 -07:00
Sergey Edunov	a8260d52d6	BPE transformation for IWSLT	2017-10-11 16:10:09 -07:00
Sam Gross	ae0c05d920	Fix call ordering to ATen addmm and sum (#22)	2017-10-11 10:14:19 -04:00
Myle Ott	7aba60843f	Update progress_bar to be more robust to changes in tqdm (#21)	2017-10-11 10:05:29 -04:00
Myle Ott	2ad5888562	Refactor PaddingCollater	2017-10-02 10:08:17 -07:00
Myle Ott	4593ebfaf9	Fix handling of partially-empty initial batch (#11)	2017-09-28 09:32:10 -07:00
Myle Ott	03c4a71698	Fix generation when vocabulary is small relative to beam size (fixes #7)	2017-09-26 11:05:51 -07:00
Sergey Edunov	2d3161daa8	Issue #2, Checking size attribute of dst when dst is None	2017-09-24 11:16:47 -07:00
Myle Ott	a615533788	Better training support when GPUs are in "exclusive mode"	2017-09-19 10:53:39 -07:00
Sergey Edunov	a8bc4d0a6f	Small fix	2017-09-18 12:53:13 -07:00
Sergey Edunov	c6de2190f8	More fixes	2017-09-18 10:26:01 -07:00
Sergey Edunov	a15acdb062	Architecture settings and readme updates	2017-09-15 11:40:28 -07:00
Sergey Edunov	e734b0fa58	Initial commit	2017-09-14 17:22:43 -07:00