GitLab

Co-Authored-By: Stefan Schweter <stefan-it@users.noreply.github.com>

Co-Authored-By: HenrykBorzymowski <henrykborzymowski@users.noreply.github.com>

Co-Authored-By: Loreto Parisi <loretoparisi@gmail.com>
Co-Authored-By: Simone Francia <francia.simone1@gmail.com>

cc @julien-c

The FlauBERT configuration file inherits from XLMConfig, and is recognized as such when loading from AutoModels as the XLMConfig is checked before the FlaubertConfig.

Changing the order solves this problem, but a test should be added.

* [Umberto] model shortcuts

cc @loretoparisi @simonefrancia

see #2485

* Ensure that tokenizers will be correctly configured

cc @lysandrejik

* fill_mask helper

* [poc] FillMaskPipeline

* Revert "[poc] FillMaskPipeline"

This reverts commit 67eeea55b0f97b46c2b828de0f4ee97d87338335.

* Revert "fill_mask helper"

This reverts commit cacc17b884e14bb6b07989110ffe884ad9e36eaa.

* README: clarify that Pipelines can also do text-classification

cf. question at the AI&ML meetup last week, @mfuntowicz

* Fix test: test feature-extraction pipeline

* Test tweaks

* Slight refactor of existing pipeline (in preparation of new FillMaskPipeline)

* Extraneous doc

* More robust way of doing this

@mfuntowicz as we don't rely on the model name anymore (see AutoConfig)

* Also add RobertaConfig as a quickfix for wrong token_type_ids

* cs

* [BIG] FillMaskPipeline

Requesting pad_token_id would cause an error message when it is None. Use private _pad_token instead.

In batch_encode_plus we have to ensure that the tokenizer has a pad_token_id so that, when padding, no None values are added as padding. That would happen with gpt2, openai, transfoxl.

closes https://github.com/huggingface/transformers/issues/2640