Pull master

README.md

@@ -164,8 +164,9 @@ At some point in the future, you'll be able to seamlessly move from pre-training
 14. **[MMBT](https://github.com/facebookresearch/mmbt/)** (from Facebook), released together with the paper a [Supervised Multimodal Bitransformers for Classifying Images and Text](https://arxiv.org/pdf/1909.02950.pdf) by Douwe Kiela, Suvrat Bhooshan, Hamed Firooz, Davide Testuggine.
 15. **[FlauBERT](https://github.com/getalp/Flaubert)** (from CNRS) released with the paper [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) by Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab.
 16. **[BART](https://github.com/pytorch/fairseq/tree/master/examples/bart)** (from Facebook) released with the paper [BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension](https://arxiv.org/pdf/1910.13461.pdf) by Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov and Luke Zettlemoyer.
-17. **[Other community models](https://huggingface.co/models)**, contributed by the [community](https://huggingface.co/users).
-18. Want to contribute a new model? We have added a **detailed guide and templates** to guide you in the process of adding a new model. You can find them in the [`templates`](./templates) folder of the repository. Be sure to check the [contributing guidelines](./CONTRIBUTING.md) and contact the maintainers or open an issue to collect feedbacks before starting your PR.
+17. **[ELECTRA](https://github.com/google-research/electra)** (from Google Research/Stanford University) released with the paper [ELECTRA: Pre-training text encoders as discriminators rather than generators](https://arxiv.org/abs/2003.10555) by Kevin Clark, Minh-Thang Luong, Quoc V. Le, Christopher D. Manning.
+18. **[Other community models](https://huggingface.co/models)**, contributed by the [community](https://huggingface.co/users).
+19. Want to contribute a new model? We have added a **detailed guide and templates** to guide you in the process of adding a new model. You can find them in the [`templates`](./templates) folder of the repository. Be sure to check the [contributing guidelines](./CONTRIBUTING.md) and contact the maintainers or open an issue to collect feedbacks before starting your PR.
 
 These implementations have been tested on several datasets (see the example scripts) and should match the performances of the original implementations (e.g. ~93 F1 on SQuAD for BERT Whole-Word-Masking, ~88 F1 on RocStories for OpenAI GPT, ~18.3 perplexity on WikiText 103 for Transformer-XL, ~0.916 Peason R coefficient on STS-B for XLNet). You can find more details on the performances in the Examples section of the [documentation](https://huggingface.co/transformers/examples.html).
 

docs/README.md

@@ -47,6 +47,8 @@ Once you have setup `sphinx`, you can build the documentation by running the fol
 make html
 ```
 
+A folder called ``_build/html`` should have been created. You can now open the file ``_build/html/index.html`` in your browser. 
+
 ---
 **NOTE**
 

docs/source/conf.py

@@ -26,7 +26,7 @@
 # The short X.Y version
 version = u''
 # The full version, including alpha/beta/rc tags
-release = u'2.6.0'
+release = u'2.8.0'
 
 
 # -- General configuration ---------------------------------------------------

docs/source/index.rst

@@ -104,3 +104,4 @@ The library currently contains PyTorch and Tensorflow implementations, pre-train
     model_doc/flaubert
     model_doc/bart
     model_doc/t5
+    model_doc/electra

docs/source/migration.md

@@ -27,7 +27,7 @@ loss = outputs[0]
 # In transformers you can also have access to the logits:
 loss, logits = outputs[:2]
 
-# And even the attention weigths if you configure the model to output them (and other outputs too, see the docstrings and documentation)
+# And even the attention weights if you configure the model to output them (and other outputs too, see the docstrings and documentation)
 model = BertForSequenceClassification.from_pretrained('bert-base-uncased', output_attentions=True)
 outputs = model(input_ids, labels=labels)
 loss, logits, attentions = outputs

docs/source/model_doc/electra.rst

@@ -0,0 +1,115 @@
+ELECTRA
+----------------------------------------------------
+
+The ELECTRA model was proposed in the paper.
+`ELECTRA: Pre-training Text Encoders as Discriminators Rather Than Generators <https://openreview.net/pdf?id=r1xMH1BtvB>`__.
+ELECTRA is a new pre-training approach which trains two transformer models: the generator and the discriminator. The
+generator's role is to replace tokens in a sequence, and is therefore trained as a masked language model. The discriminator,
+which is the model we're interested in, tries to identify which tokens were replaced by the generator in the sequence.
+
+The abstract from the paper is the following:
+
+*Masked language modeling (MLM) pre-training methods such as BERT corrupt
+the input by replacing some tokens with [MASK] and then train a model to
+reconstruct the original tokens. While they produce good results when transferred
+to downstream NLP tasks, they generally require large amounts of compute to be
+effective. As an alternative, we propose a more sample-efficient pre-training task
+called replaced token detection. Instead of masking the input, our approach
+corrupts it by replacing some tokens with plausible alternatives sampled from a small
+generator network. Then, instead of training a model that predicts the original
+identities of the corrupted tokens, we train a discriminative model that predicts
+whether each token in the corrupted input was replaced by a generator sample
+or not. Thorough experiments demonstrate this new pre-training task is more
+efficient than MLM because the task is defined over all input tokens rather than
+just the small subset that was masked out. As a result, the contextual representations
+learned by our approach substantially outperform the ones learned by BERT
+given the same model size, data, and compute. The gains are particularly strong
+for small models; for example, we train a model on one GPU for 4 days that
+outperforms GPT (trained using 30x more compute) on the GLUE natural language
+understanding benchmark. Our approach also works well at scale, where it
+performs comparably to RoBERTa and XLNet while using less than 1/4 of their
+compute and outperforms them when using the same amount of compute.*
+
+Tips:
+
+- ELECTRA is the pre-training approach, therefore there is nearly no changes done to the underlying model: BERT. The
+  only change is the separation of the embedding size and the hidden size -> The embedding size is generally smaller,
+  while the hidden size is larger. An additional projection layer (linear) is used to project the embeddings from
+  their embedding size to the hidden size. In the case where the embedding size is the same as the hidden size, no
+  projection layer is used.
+- The ELECTRA checkpoints saved using `Google Research's implementation <https://github.com/google-research/electra>`__
+  contain both the generator and discriminator. The conversion script requires the user to name which model to export
+  into the correct architecture. Once converted to the HuggingFace format, these checkpoints may be loaded into all
+  available ELECTRA models, however. This means that the discriminator may be loaded in the `ElectraForMaskedLM` model,
+  and the generator may be loaded in the `ElectraForPreTraining` model (the classification head will be randomly
+  initialized as it doesn't exist in the generator).
+
+
+ElectraConfig
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.ElectraConfig
+    :members:
+
+
+ElectraTokenizer
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.ElectraTokenizer
+    :members:
+
+
+ElectraModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.ElectraModel
+    :members:
+
+
+ElectraForPreTraining
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.ElectraForPreTraining
+    :members:
+
+
+ElectraForMaskedLM
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.ElectraForMaskedLM
+    :members:
+
+
+ElectraForTokenClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.ElectraForTokenClassification
+    :members:
+
+
+TFElectraModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TFElectraModel
+    :members:
+
+
+TFElectraForPreTraining
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TFElectraForPreTraining
+    :members:
+
+
+TFElectraForMaskedLM
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TFElectraForMaskedLM
+    :members:
+
+
+TFElectraForTokenClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TFElectraForTokenClassification
+    :members:

docs/source/model_doc/t5.rst

@@ -16,13 +16,45 @@ To facilitate future work on transfer learning for NLP, we release our dataset,
 
 The Authors' code can be found `here <https://github.com/google-research/text-to-text-transfer-transformer>`_ .
 
+Training
+~~~~~~~~~~~~~~~~~~~~
+T5 is an encoder-decoder model and converts all NLP problems into a text-to-text format. It is trained using teacher forcing.
+This means that for training we always need an input sequence and a target sequence. 
+The input sequence is fed to the model using ``input_ids``. The target sequence is shifted to the right, *i.e.* perprended by a start-sequence token and fed to the decoder using the `decoder_input_ids`. In teacher-forcing style, the target sequence is then appended by the EOS token and corresponds to the ``lm_labels``. The PAD token is hereby used as the start-sequence token.
+T5 can be trained / fine-tuned both in a supervised and unsupervised fashion.
+
+- Unsupervised denoising training
+  In this setup spans of the input sequence are masked by so-called sentinel tokens (*a.k.a* unique mask tokens) 
+  and the output sequence is formed as a concatenation of the same sentinel tokens and the *real* masked tokens. 
+  Each sentinel tokens represents a unique mask token for this sentence and should start with ``<extra_id_1>``, ``<extrac_id_2>``, ... up to ``<extra_id_100>``. As a default 100 sentinel tokens are available in ``T5Tokenizer``.
+  *E.g.* the sentence "The cute dog walks in the park" with the masks put on "cute dog" and "the" should be processed as follows: 
+
+::
+
+  input_ids = tokenizer.encode('The <extra_id_1> walks in <extra_id_2> park', return_tensors='pt')
+  lm_labels = tokenizer.encode('<extra_id_1> cute dog <extra_id_2> the <extra_id_3> </s>', return_tensors='pt')
+  # the forward function automatically creates the correct decoder_input_ids
+  model(input_ids=input_ids, lm_labels=lm_labels)
+
+- Supervised training
+  In this setup the input sequence and output sequence are standard sequence to sequence input output mapping.
+  In translation, *e.g.* the input sequence "The house is wonderful." and output sequence "Das Haus ist wunderbar." should 
+  be processed as follows:
+
+::
+
+  input_ids = tokenizer.encode('translate English to German: The house is wonderful. </s>', return_tensors='pt')
+  lm_labels = tokenizer.encode('Das Haus ist wunderbar. </s>', return_tensors='pt')
+  # the forward function automatically creates the correct decoder_input_ids
+  model(input_ids=input_ids, lm_labels=lm_labels)
+
 Tips
 ~~~~~~~~~~~~~~~~~~~~
 - T5 is an encoder-decoder model pre-trained on a multi-task mixture of unsupervised 
-  and supervised tasks and which each task is cast as a sequence to sequence task. 
-  Therefore T5 works well on a variety of tasks out-of-the-box by prepending a different prefix to the input corresponding to each task, e.g.: for translation: *translate English to German: ..., summarize: ...*.
-  For more information about the which prefix to use, it is easiest to look into Appendix D of the `paper <https://arxiv.org/pdf/1910.10683.pdf>`_ .
-- For sequence to sequence generation, it is recommended to use ``T5ForConditionalGeneration.generate()``. The method takes care of feeding the encoded input via cross-attention layers to the decoder and auto-regressively generating the decoder output.
+  and supervised tasks and for which each task is converted into a text-to-text format.
+  T5 works well on a variety of tasks out-of-the-box by prepending a different prefix to the input corresponding to each task, e.g.: for translation: *translate English to German: ..., summarize: ...*.
+  For more information about which prefix to use, it is easiest to look into Appendix D of the `paper <https://arxiv.org/pdf/1910.10683.pdf>`_ .
+- For sequence to sequence generation, it is recommended to use ``T5ForConditionalGeneration.generate()``. The method takes care of feeding the encoded input via cross-attention layers to the decoder and auto-regressively generates the decoder output.
 - T5 uses relative scalar embeddings. Encoder input padding can be done on the left and on the right.
 
 

docs/source/notebooks.md

@@ -0,0 +1 @@
+../../notebooks/README.md