Commit Graph
19 Commits
Author SHA1 Message Date
Julien Chaumond f9453d15e5 Fix broken link 2019-09-05 12:35:22 -04:00
Julien Chaumond f7ee2e5d20 [README] link to Write With Transformer 2019-09-05 12:33:46 -04:00
Julien Chaumond 2dcc5a1629 [doc] Add blurb about large-scale model downloads
cc @n1t0 @lysandrejik @thomwolf
2019-09-02 12:27:11 -04:00
Julien Chaumond 0fd0b674e6 [ci] legible output [skip ci] 2019-08-30 20:36:26 -04:00
Julien Chaumond b65a994f59 [ci] decrease parallelism to increase success prob 2019-08-30 20:33:16 -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
Julien Chaumond b0b9b8091b minor typo 2019-08-20 11:33:46 +02:00
Julien ChaumondandMyle Ott c4ef103447 [RoBERTa] First 4 authors
cf. https://github.com/huggingface/pytorch-transformers/pull/964#discussion_r313574354

Co-Authored-By: Myle Ott <myleott@fb.com>
2019-08-14 12:31:09 -04:00
Julien Chaumond baf08ca1d4 [RoBERTa] run_glue: correct pad_token + reorder labels 2019-08-13 12:51:15 -04:00
Julien Chaumond 912fdff899 [RoBERTa] Update run_glue for RoBERTa 2019-08-12 13:49:50 -04:00
Julien Chaumond b3d83d68db Fixup 9d0603148b 2019-08-12 12:28:55 -04:00
Julien Chaumond e367ac469c [RoBERTa] Re-apply 39d72bcc7b
cc @lysandrejik
2019-08-08 11:26:11 -04:00
Julien Chaumond 9d0603148b [RoBERTa] RobertaForSequenceClassification + conversion 2019-08-08 11:24:54 -04:00
Julien Chaumond cb9db101c7 Python 2 must DIE 2019-08-04 22:04:15 -04:00
Julien Chaumond 05c083520a [RoBERTa] model conversion, inference, tests 🔥 2019-08-04 21:39:21 -04:00
Julien Chaumond 44dd941efb link to swift-coreml-transformers 2019-08-01 09:50:30 -04:00
Julien Chaumond 8da280ebbe Setup CI 2018-12-20 16:33:39 -05:00
Julien Chaumond d57763f582 Fix typos 2018-12-18 19:23:22 -05:00