SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity
1,478 Commits 5 Branches 239 Tags
fede4ef45ddec6e7706548836b3ae2a7728fa93a
Commit Graph

265 Commits

Author SHA1 Message Date
thomwolf
fede4ef45d fixing #1133 2019-09-02 02:27:39 +02:00
Thomas Wolf
ff7368eb6b Merge pull request #1077 from huggingface/pruning-save-and-load 
Pruning changes so that deleted heads are kept on save/load
 2019-09-01 09:42:15 +02:00
LysandreJik
6ae0bb5291 XLM 100 different URLs 2019-08-31 14:46:31 -04:00
LysandreJik
819b468f70 Fixed XLM model url 2019-08-31 14:40:51 -04:00
LysandreJik
11600edc6e Rebase on master + DistilBERT head pruning patch 2019-08-31 00:37:41 -04:00
LysandreJik
b6992b7b47 Applied patch to OpenAI GPT, RoBERTa, TransfoL, XLM and XLNet 2019-08-31 00:33:50 -04:00
thomwolf
bdb4409ed8 updated pruning logic with sets - Bert and GPT-2 2019-08-31 00:33:50 -04:00
LysandreJik
0c8e823b03 Added patch to remaining models 2019-08-31 00:33:50 -04:00
LysandreJik
0cd283522a Attempt to fix head index 2019-08-31 00:33:50 -04:00
LysandreJik
c85b5db61a Conditional append/init + fixed warning 2019-08-31 00:33:50 -04:00
LysandreJik
5c2b94c82a Changed string so that Circle CI accepts the warning 2019-08-31 00:33:50 -04:00
LysandreJik
87747518e9 Blocks deletion from already deleted heads. Necessary integration test. 
Now raises a warning when a head to be deleted already has been deleted. An integration test verifying the total pipeline (-> from config -> save model -> load model -> additional head pruning) has been added.
 2019-08-31 00:33:50 -04:00
LysandreJik
719cb3738d Pruning for GPT and GPT-2 2019-08-31 00:33:50 -04:00
LysandreJik
fc1fbae45d XLM can be pruned 2019-08-31 00:33:50 -04:00
Lysandre
42e00cf9e1 Pruning saved to configuration first try 2019-08-31 00:33:50 -04:00
LysandreJik
d7a4c3252e Fixed filename 2019-08-31 00:08:56 -04:00
LysandreJik
7f006cdd87 Set seed for head_masking test 2019-08-30 23:58:49 -04:00
Julien Chaumond
1d438f15b3 [XLNet] Use pytorch's layernorm like in BERT 
See #1089

cc @thomwolf @lysandrejik

Also @dhpollack
 2019-08-30 20:20:15 -04:00
Julien Chaumond
574c5b3a72 [RoBERTa] LayerNorm's eps is not a nn.Parameter so there's no point setting it on the model 
Instead we correctly store it on the config

(regenerating the hosted config files)

cc @lysandrejik
 2019-08-30 20:09:24 -04:00
Thomas Wolf
206c35e9a4 Merge pull request #1154 from ziliwang/master 
fix: hard coding for max number
 2019-08-30 23:23:08 +02:00
Thomas Wolf
f3d18c71ec Merge pull request #1152 from epwalsh/fix-special-tokens 
fix adding special tokens
 2019-08-30 23:21:59 +02:00
Thomas Wolf
d483cd8e46 Merge pull request #1074 from huggingface/improved_testing 
Shortcut to special tokens' ids - fix GPT2 & RoBERTa tokenizers - improved testing for GPT/GPT-2
 2019-08-30 23:18:58 +02:00
Thomas Wolf
d2f21f08f5 Merge pull request #1092 from shijie-wu/xlm-tokenization 
Added cleaned configuration properties for tokenizer with serialization - improve tokenization of XLM
 2019-08-30 23:15:40 +02:00
thomwolf
bfe93a5a21 fix distilbert in auto tokenizer 2019-08-30 22:43:26 +02:00
thomwolf
256086bc69 clean up and simplify hubconf 2019-08-30 22:34:23 +02:00
thomwolf
80aa87d9a3 fix distilbert tokenizer 2019-08-30 22:24:23 +02:00
thomwolf
455a4c842c add distilbert tokenizer 2019-08-30 22:20:51 +02:00
LysandreJik
dec8f4d6fd Added DistilBERT models to all other AutoModels. 2019-08-30 13:52:18 -04:00
thomwolf
f35f612280 updating docstring for AutoModel 2019-08-30 12:48:55 -04:00
LysandreJik
25e8389439 Tests for added AutoModels 2019-08-30 12:48:55 -04:00
LysandreJik
dc43215c01 Added multiple AutoModel classes: AutoModelWithLMHead, AutoModelForQuestionAnswering and AutoModelForSequenceClassification 2019-08-30 12:48:55 -04:00
thomwolf
7044ed6b05 fix tokenizers serialization 2019-08-30 17:36:11 +02:00
Thomas Wolf
cd65c41a83 Merge branch 'master' into xlm-tokenization 2019-08-30 17:15:16 +02:00
thomwolf
69da972ace added test and debug tokenizer configuration serialization 2019-08-30 17:09:36 +02:00
thomwolf
88111de07c saving and reloading tokenizer configurations 2019-08-30 16:55:48 +02:00
thomwolf
3871b8a107 adding xlm 17 and 100 models and config on aws 2019-08-30 16:28:42 +02:00
thomwolf
8678ff8df5 adding 17 and 100 xlm models 2019-08-30 16:26:04 +02:00
thomwolf
82462c5cba Added option to setup pretrained tokenizer arguments 2019-08-30 15:30:41 +02:00
Thomas Wolf
41f35d0b3d Merge pull request #1089 from dhpollack/dhp/use_pytorch_layernorm 
change layernorm code to pytorch's native layer norm
 2019-08-30 14:49:08 +02:00
Thomas Wolf
50e615f43d Merge branch 'master' into improved_testing 2019-08-30 13:40:35 +02:00
thomwolf
f8aace6bcd update tokenizers to use self.XX_token_id instead of converting self.XX_token 2019-08-30 13:39:52 +02:00
thomwolf
8faf2e086b more doc on special tokens 2019-08-30 13:36:22 +02:00
thomwolf
ce5ef4b35d python2 doesn't spark joy 2019-08-30 13:22:43 +02:00
thomwolf
5dd7b677ad clean up all byte-level bpe tests 2019-08-30 12:43:08 +02:00
thomwolf
ca1a00a302 fix for python2 2019-08-30 12:29:31 +02:00
thomwolf
4e6a3172ce update roberta docstring as well 2019-08-30 12:23:37 +02:00
thomwolf
fd10d79b55 update GPT2 docstring 2019-08-30 12:23:12 +02:00
thomwolf
abe734ca1f fix GPT-2 and RoBERTa tests to be clean now 2019-08-30 12:20:18 +02:00
thomwolf
0f5a799456 fix GPT2DoubleHeadModel docstring 2019-08-30 11:49:23 +02:00
thomwolf
d51f72d5de adding shortcut to the ids of all the special tokens 2019-08-30 11:41:11 +02:00