GitLab

see #3359

cc @lysandrejik

cc @julien-c

hat/tip @bramvanroy

For more details, see https://pytorch.org/docs/stable/optim.html#how-to-adjust-learning-rate

The CONTRIBUTING file pins to a specific version of isort, so we might as well install that in `dev` . This makes it easier for contributors so they don't have to manually install the specific commit.

* make style

* fix conflicts

torch.cuda.empty_cache() was being called from a TF function (even when torch is unavailable)
not sure any replacement is needed if TF OOMs

I found there are two grammar errors or typo issues in the explanation of the encoding properties.

The original sentences:
If your was made of multiple \"parts\" such as (question, context), then this would be a vector with for each token the segment it belongs to
If your has been truncated into multiple subparts because of a length limit (for BERT for example the sequence length is limited to 512), this will contain all the remaining overflowing parts.

I think "input" should be inserted after the phrase "If your".

* 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>

Create card for BERT-Mini finetuned on SQuAD v2

- Only 17MB of Model weights!!

* Add a model card for FinBERT

This is a copy of https://github.com/TurkuNLP/FinBERT/blob/master/README.md.

* Added a file for uncased.

* Add metadata for cased.

* Added metadata for uncased.

For the tutorial of "How to generate text", the URL link was wrong (it was linked to the tutorial of "How to train a language model").

I fixed the URL.

* added return_token_type_ids argument for tokenizers which do not generate return_type_ids by default

* fixed styling

* Style
Co-authored-by: LysandreJik <lysandre.debut@reseau.eseo.fr>

* config.activation_function

* solving bug where for small epochs and large gradient_accumulation_steps we never train

* black formatting

* no need to change these files

`T5Tokenizer` instead of `XLNetTokenizer`

* fix issue 3289

* fix attention mask if input_ids None behavior

roberta_chinese_base card

albert_chinese_tiny card

* Create README.md

* model card

* add model card for cased

Remove hardcoded mask_token and use the value provided by the tokenizer.