SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity
3,996 Commits 5 Branches 239 Tags
fa2fbed3e5e05efeed721b52a74f4b40d6b382ff
Commit Graph

48 Commits

Author SHA1 Message Date
Sam Shleifer
9a687ebb77 [Marian Fixes] prevent predicting pad_token_id before softmax, support language codes, name multilingual models (#4290) 2020-05-13 17:29:41 -04:00
Julien Chaumond
4bf5042240 Fix BART tests on GPU (#4298) 2020-05-12 09:11:50 -04:00
Julien Chaumond
455c639093 CDN urls (#4030) 
* [file_utils] use_cdn + documentation

* Move to cdn. urls for weights

* [urls] Hotfix for bert-base-japanese
 2020-04-28 20:27:14 -04:00
Sam Shleifer
847e7f3379 MarianMTModel.from_pretrained('Helsinki-NLP/opus-marian-en-de') (#3908) 
Co-Authored-By: Stefan Schweter <stefan@schweter.it>
 2020-04-28 18:22:37 -04:00
Julien Chaumond
a946b6b51b [housekeeping] Upgrade # type Python 2 syntax 
cc @sshleifer
 2020-04-23 10:39:24 -04:00
Patrick von Platen
01c37dcdb5 [Config, Caching] Remove output_past everywhere and replace by use_cache argument (#3734) 
* remove output_past from pt

* make style

* add optional input length for gpt2

* add use cache to prepare input

* save memory in gpt2

* correct gpt2 test inputs

* make past input optional for gpt2

* finish use_cache for all models

* make style

* delete modeling_gpt2 change in test file

* correct docstring

* correct is true statements for gpt2
 2020-04-14 14:40:28 -04:00
Sam Shleifer
7a7fdf71f8 Multilingual BART - (#3602) 
- support mbart-en-ro weights
- add MBartTokenizer
 2020-04-10 11:25:39 -04:00
Sam Shleifer
715aa5b135 [Bart] Replace config.output_past with use_cache kwarg (#3632) 2020-04-07 19:08:26 -04:00
Patrick von Platen
390c128592 [Encoder-Decoder] Force models outputs to always have batch_size as their first dim (#3536) 
* solve conflicts

* improve comments
 2020-04-02 15:18:33 +02:00
dougian
1f72865726 [BART] Update encoder and decoder on set_input_embedding (#3501) 
Co-authored-by: Ioannis Douratsos <ioannisd@amazon.com>
 2020-03-30 12:20:37 -04:00
Patrick von Platen
5b44e0a31b [T5] Add training documenation (#3507) 
* Add clear description of how to train T5

* correct docstring in T5

* correct typo

* correct docstring format

* update t5 model docs

* implement collins feedback

* fix typo and add more explanation for sentinal tokens

* delete unnecessary todos
 2020-03-30 13:35:53 +02:00
Sam Shleifer
f6a23d1911 [BART] add bart-large-xsum weights (#3422) 2020-03-29 10:51:13 -04:00
Patrick von Platen
fa9af2468a Add T5 to docs (#3461) 
* add t5 docs basis

* improve docs

* add t5 docs

* improve t5 docstring

* add t5 tokenizer docstring

* finish docstring

* make style

* add pretrained models

* correct typo

* make examples work

* finalize docs
 2020-03-27 10:57:16 -04:00
Sam Shleifer
3ee431dd4c [Bart/Memory] Two separate, smaller decoder attention masks (#3371) 2020-03-26 21:34:15 -04:00
Sam Shleifer
63f4d8cad0 [Bart/Memory] SelfAttention only returns weights if config.outp… (#3369) 2020-03-26 18:42:39 -04:00
Sam Shleifer
2b2a2f8df2 [Bart] Fix: put dummy_inputs on correct device (#3398) 
* Dummy inputs to model.device

* Move self.device to ModuleUtilsMixin
 2020-03-26 18:42:09 -04:00
Sam Shleifer
1a5aefc95c [Seq2Seq Generation] Call encoder before expanding input_ids (#3370) 2020-03-26 18:41:19 -04:00
Sam Shleifer
39371ee454 [Bart/Memory] don't create lm_head (#3323) 
* delete lm_head, skips weight tying
* Fixed s3
 2020-03-26 18:40:39 -04:00
Patrick von Platen
95e00d0808 Clean special token init in modeling_....py (#3264) 
* make style

* fix conflicts
 2020-03-20 21:41:04 +01:00
Patrick von Platen
bbf26c4e61 Support T5 Generation (#3228) 
* fix conflicts

* update bart max length test

* correct spelling mistakes

* implemented model specific encode function

* fix merge conflicts

* better naming

* save intermediate state -> need to rethink strucuture a bit

* leave tf problem as it is for now

* current version

* add layers.pop

* remove ipdb

* make style

* clean return cut decoding

* remove ipdbs

* Fix restoring layers in the decoders that doesnt exists.

* push good intermediate solution for now

* fix conflicts

* always good to refuse to merge conflicts when rebasing

* fix small bug

* improve function calls

* remove unused file

* add correct scope behavior for t5_generate

Co-authored-by: Morgan Funtowicz <funtowiczmo@gmail.com>
 2020-03-19 23:18:23 +01:00
Sam Shleifer
4e4403c9b4 [BART] torch 1.0 compatibility (#3322) 
* config.activation_function
 2020-03-19 11:56:54 -04:00
Sam Shleifer
ad7233fc01 [BART] cleanup: remove redundant kwargs, improve docstrings (#3319) 2020-03-19 11:16:51 -04:00
Sam Shleifer
11573231c6 [BART] generation_mode as a kwarg not a class attribute (#3278) 2020-03-16 12:47:53 -04:00
Sam Shleifer
5ea8ba67b4 [BART] Remove unused kwargs (#3279) 
* Remove unused kwargs
* dont call forward in tests
 2020-03-15 23:00:44 -04:00
Sam Shleifer
2bd79e23de [BART] FP16 testing fixes (#3266) 2020-03-13 19:48:26 -04:00
Sam Shleifer
2e81b9d8d7 Bart: update example for #3140 compatibility (#3233) 
* Update bart example docs
 2020-03-12 10:36:37 -04:00
Patrick von Platen
a332cc9f7f finalize generation merge 2020-03-11 11:53:36 +01:00
patrickvonplaten
2acfe63964 best current version and make style 2020-03-11 11:06:56 +01:00
patrickvonplaten
c62444da39 fix conflicts 2020-03-11 11:06:56 +01:00
Patrick von Platen
7a11e925cf work in progress 2020-03-11 11:06:56 +01:00
Patrick von Platen
7cba11fb9b better naming 2020-03-11 11:06:56 +01:00
Patrick von Platen
ff648221bd fix conflicts 2020-03-11 11:06:56 +01:00
Patrick von Platen
d8e2b3c547 fix conflicts 2020-03-11 11:06:56 +01:00
Sam Shleifer
ed37f9fa4f [Bart] _prepare_decoder_inputs should use large negative (#3158) 2020-03-06 16:06:36 -05:00
Sam Shleifer
e58b3ec5df add imports to examples (#3160) 2020-03-06 11:15:33 -05:00
Thomas Wolf
6ffe03a0a1 Merge pull request #3137 from tomhosking/bart-refactor 
Refactor BartModel so that input checks are handled within enc/dec
 2020-03-06 13:06:34 +01:00
Sam Shleifer
857e0a0d3b Rename BartForMaskedLM -> BartForConditionalGeneration (#3114) 
* improved documentation
 2020-03-05 17:41:18 -05:00
sshleifer
14d40584b2 remove newline 2020-03-05 13:06:35 -05:00
sshleifer
1360dacaa3 cleanup deltas 2020-03-05 12:57:42 -05:00
sshleifer
810079de1f no ipdb 2020-03-05 12:48:14 -05:00
sshleifer
c36fdc88d4 tests pass 2020-03-05 12:33:08 -05:00
Tom Hosking
06a6cb6f36 Refactor BartModel so that input checks are handled within BartEncoder and BartDecoder 2020-03-05 13:45:41 +00:00
Sam Shleifer
e9e6efdc45 BartForSequenceClassification: fix num_labels, add test (#3110) 2020-03-03 15:54:29 -05:00
Sam Shleifer
5c5af879b6 [Bart] dont call .forward (#3094) 2020-03-03 15:14:12 -05:00
Sam Shleifer
b54ef78d0c Bart-CNN (#3059) 
`generate` code that produces 99% identical summarizations to fairseq on CNN test data, with caching.
 2020-03-02 10:35:53 -05:00
Julien Chaumond
9cda3620b6 Fix (non-slow) tests on GPU (torch) (#3024) 
* Fix tests on GPU (torch)

* Fix bart slow tests

Co-authored-by: Sam Shleifer <sshleifer@gmail.com>
 2020-02-26 11:59:25 -05:00
Sam Shleifer
92487a1dc0 Bart: fix layerdrop and cached decoder_input_ids for generation (#2969) 2020-02-22 16:25:04 -05:00
Sam Shleifer
53ce3854a1 New BartModel (#2745) 
* Results same as fairseq
* Wrote a ton of tests
* Struggled with api signatures
* added some docs
 2020-02-20 18:11:13 -05:00