Refactor Code samples; Test code samples (#5036)

* Refactor code samples

* Test docstrings

* Style

* Tokenization examples

* Run rust of tests

* First step to testing source docs

* Style and BART comment

* Test the remainder of the code samples

* Style

* let to const

* Formatting fixes

* Ready for merge

* Fix fixture + Style

* Fix last tests

* Update docs/source/quicktour.rst
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* Addressing @sgugger's comments + Fix MobileBERT in TF
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

docs/source/_static/css/code-snippets.css

@@ -10,3 +10,7 @@
 .highlight .kn, .highlight .nv, .highlight .s2, .highlight .ow {
     color: #6670FF;
 }
+
+.highlight .gp {
+    color: #FB8D68;
+}
\ No newline at end of file

docs/source/_static/css/huggingface.css

@@ -44,6 +44,7 @@
     display: flex;
     flex-direction: row;
     justify-content: flex-end;
+    margin-right: 30px;
 }
 
 .framework-selector > button {
@@ -60,6 +61,12 @@
     padding: 5px;
 }
 
+/* Copy button */
+
+a.copybtn {
+    margin: 3px;
+}
+
 /* The literal code blocks */
 .rst-content tt.literal, .rst-content tt.literal, .rst-content code.literal {
     color: #6670FF;

docs/source/_static/js/custom.js

@@ -157,6 +157,8 @@ function platformToggle() {
     const codeBlocks = Array.from(document.getElementsByClassName("highlight"));
     const pytorchIdentifier = "## PYTORCH CODE";
     const tensorflowIdentifier = "## TENSORFLOW CODE";
+
+    const promptSpanIdentifier = `<span class="gp">&gt;&gt;&gt; </span>`
     const pytorchSpanIdentifier = `<span class="c1">${pytorchIdentifier}</span>`;
     const tensorflowSpanIdentifier = `<span class="c1">${tensorflowIdentifier}</span>`;
 
@@ -169,10 +171,22 @@ function platformToggle() {
         let tensorflowSpans;
 
         if(pytorchSpanPosition < tensorflowSpanPosition){
-            pytorchSpans = spans.slice(pytorchSpanPosition + pytorchSpanIdentifier.length + 1, tensorflowSpanPosition);
+            const isPrompt = spans.slice(
+                spans.indexOf(tensorflowSpanIdentifier) - promptSpanIdentifier.length,
+                spans.indexOf(tensorflowSpanIdentifier)
+            ) == promptSpanIdentifier;
+            const finalTensorflowSpanPosition = isPrompt ? tensorflowSpanPosition - promptSpanIdentifier.length : tensorflowSpanPosition;
+
+            pytorchSpans = spans.slice(pytorchSpanPosition + pytorchSpanIdentifier.length + 1, finalTensorflowSpanPosition);
             tensorflowSpans = spans.slice(tensorflowSpanPosition + tensorflowSpanIdentifier.length + 1, spans.length);
         }else{
-            tensorflowSpans = spans.slice(tensorflowSpanPosition + tensorflowSpanIdentifier.length + 1, pytorchSpanPosition);
+            const isPrompt = spans.slice(
+                spans.indexOf(pytorchSpanIdentifier) - promptSpanIdentifier.length,
+                spans.indexOf(pytorchSpanIdentifier)
+            ) == promptSpanIdentifier;
+            const finalPytorchSpanPosition = isPrompt ? pytorchSpanPosition - promptSpanIdentifier.length : pytorchSpanPosition;
+
+            tensorflowSpans = spans.slice(tensorflowSpanPosition + tensorflowSpanIdentifier.length + 1, finalPytorchSpanPosition);
             pytorchSpans = spans.slice(pytorchSpanPosition + pytorchSpanIdentifier.length + 1, spans.length);
         }
 

docs/source/conf.py

@@ -44,7 +44,8 @@ extensions = [
     'sphinx.ext.napoleon',
     'recommonmark',
     'sphinx.ext.viewcode',
-    'sphinx_markdown_tables'
+    'sphinx_markdown_tables',
+    'sphinx_copybutton'
 ]
 
 # Add any paths that contain templates here, relative to this directory.
@@ -74,6 +75,8 @@ exclude_patterns = [u'_build', 'Thumbs.db', '.DS_Store']
 # The name of the Pygments (syntax highlighting) style to use.
 pygments_style = None
 
+# Remove the prompt when copying examples
+copybutton_prompt_text = ">>> "
 
 # -- Options for HTML output -------------------------------------------------
 

docs/source/glossary.rst

@@ -45,17 +45,16 @@ tokenizer, which is a `WordPiece <https://arxiv.org/pdf/1609.08144.pdf>`__ token
 
 ::
 
-    from transformers import BertTokenizer
-    tokenizer = BertTokenizer.from_pretrained("bert-base-cased")
+    >>> from transformers import BertTokenizer
+    >>> tokenizer = BertTokenizer.from_pretrained("bert-base-cased")
 
-    sequence = "A Titan RTX has 24GB of VRAM"
+    >>> sequence = "A Titan RTX has 24GB of VRAM"
 
 The tokenizer takes care of splitting the sequence into tokens available in the tokenizer vocabulary.
 
 ::
 
-    tokenized_sequence = tokenizer.tokenize(sequence)
-    print(tokenized_sequence)
+    >>> tokenized_sequence = tokenizer.tokenize(sequence)
 
 The tokens are either words or subwords. Here for instance, "VRAM" wasn't in the model vocabulary, so it's been split
 in "V", "RA" and "M". To indicate those tokens are not separate words but parts of the same word, a double-dash is
@@ -63,6 +62,7 @@ added for "RA" and "M":
 
 ::
 
+    >>> print(tokenized_sequence)
     ['A', 'Titan', 'R', '##T', '##X', 'has', '24', '##GB', 'of', 'V', '##RA', '##M']
 
 These tokens can then be converted into IDs which are understandable by the model. This can be done by directly feeding
@@ -71,14 +71,14 @@ the sentence to the tokenizer, which leverages the Rust implementation of
 
 ::
 
-    encoded_sequence = tokenizer(sequence)["input_ids"]
-    print(encoded_sequence)
+    >>> encoded_sequence = tokenizer(sequence)["input_ids"]
 
 The tokenizer returns a dictionary with all the arguments necessary for its corresponding model to work properly. The
 token indices are under the key "input_ids":
 
 ::
 
+    >>> print(encoded_sequence)
     [101, 138, 18696, 155, 1942, 3190, 1144, 1572, 13745, 1104, 159, 9664, 2107, 102]
 
 Note that the tokenizer automatically adds "special tokens" (if the associated model rely on them) which are special
@@ -86,13 +86,14 @@ IDs the model sometimes uses. If we decode the previous sequence of ids,
 
 ::
 
-    tokenizer.decode(encoded_sequence)
+    >>> decoded_sequence = tokenizer.decode(encoded_sequence)
 
 we will see 
 
 ::
 
-    '[CLS] A Titan RTX has 24GB of VRAM [SEP]'
+    >>> print(decoded_sequence)
+    [CLS] A Titan RTX has 24GB of VRAM [SEP]
 
 because this is the way a :class:`~transformers.BertModel` is going to expect its inputs.
 
@@ -108,21 +109,20 @@ For example, consider these two sequences:
 
 ::
 
-    from transformers import BertTokenizer
-    tokenizer = BertTokenizer.from_pretrained("bert-base-cased")
+    >>> from transformers import BertTokenizer
+    >>> tokenizer = BertTokenizer.from_pretrained("bert-base-cased")
 
-    sequence_a = "This is a short sequence."
-    sequence_b = "This is a rather long sequence. It is at least longer than the sequence A."
+    >>> sequence_a = "This is a short sequence."
+    >>> sequence_b = "This is a rather long sequence. It is at least longer than the sequence A."
 
-    encoded_sequence_a = tokenizer(sequence_a)["input_ids"]
-    encoded_sequence_b = tokenizer(sequence_b)["input_ids"]
-    
-    len(encoded_sequence_a), len(encoded_sequence_b)
+    >>> encoded_sequence_a = tokenizer(sequence_a)["input_ids"]
+    >>> encoded_sequence_b = tokenizer(sequence_b)["input_ids"]
 
 The encoded versions have different lengths:
 
 ::
 
+    >>> len(encoded_sequence_a), len(encoded_sequence_b)
     (8, 19)
 
 Therefore, we can't be put then together in a same tensor as-is. The first sequence needs to be padded up to the length
@@ -133,15 +133,14 @@ it to pad like this:
 
 ::
 
-    padded_sequences = tokenizer([sequence_a, sequence_b], padding=True)
-    padded_sequences["input_ids"]
+    >>> padded_sequences = tokenizer([sequence_a, sequence_b], padding=True)
 
 We can see that 0s have been added on the right of the first sentence to make it the same length as the second one:
 
 ::
 
-    [[101, 1188, 1110, 170, 1603, 4954, 119, 102, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
-     [101, 1188, 1110, 170, 1897, 1263, 4954, 119, 1135, 1110, 1120, 1655, 2039, 1190, 1103, 4954, 138, 119, 102]]
+    >>> padded_sequences["input_ids"]
+    [[101, 1188, 1110, 170, 1603, 4954, 119, 102, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [101, 1188, 1110, 170, 1897, 1263, 4954, 119, 1135, 1110, 1120, 1655, 2039, 1190, 1103, 4954, 138, 119, 102]]
 
 This can then be converted into a tensor in PyTorch or TensorFlow. The attention mask is a binary tensor indicating
 the position of the padded indices so that the model does not attend to them. For the
@@ -150,14 +149,8 @@ a padded value. This attention mask is in the dictionary returned by the tokeniz
 
 ::
 
-    padded_sequences["attention_mask"]
-
-will give back
-
-::
-
-    [[1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
-     [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]]
+    >>> padded_sequences["attention_mask"]
+    [[1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]]
 
 .. _token-type-ids:
 
@@ -170,26 +163,27 @@ tokens. For example, the BERT model builds its two sequence input as such:
 
 ::
 
-   # [CLS] SEQUENCE_A [SEP] SEQUENCE_B [SEP]
+   >>> # [CLS] SEQUENCE_A [SEP] SEQUENCE_B [SEP]
 
 We can use our tokenizer to automatically generate such a sentence by passing the two sequences as two arguments (and
 not a list like before) like this:
 
 ::
 
-    from transformers import BertTokenizer
-    tokenizer = BertTokenizer.from_pretrained("bert-base-cased")
-    sequence_a = "HuggingFace is based in NYC"
-    sequence_b = "Where is HuggingFace based?"
+    >>> from transformers import BertTokenizer
+    >>> tokenizer = BertTokenizer.from_pretrained("bert-base-cased")
+    >>> sequence_a = "HuggingFace is based in NYC"
+    >>> sequence_b = "Where is HuggingFace based?"
 
-    encoded_dict = tokenizer(sequence_a, sequence_b)
-    tokenizer.decode(encoded_dict["input_ids"])
+    >>> encoded_dict = tokenizer(sequence_a, sequence_b)
+    >>> decoded = tokenizer.decode(encoded_dict["input_ids"])
 
 which will return:
 
 ::
 
-    "[CLS] HuggingFace is based in NYC [SEP] Where is HuggingFace based? [SEP]"
+    >>> print(decoded)
+    [CLS] HuggingFace is based in NYC [SEP] Where is HuggingFace based? [SEP]
 
 This is enough for some models to understand where one sequence ends and where another begins. However, other models
 such as BERT have an additional mechanism, which are the token type IDs (also called segment IDs). They are a binary
@@ -199,12 +193,7 @@ The tokenizer returns in the dictionary under the key "token_type_ids":
 
 ::
 
-    encoded_dict['token_type_ids']
-
-will return
-
-::
-
+    >>> encoded_dict['token_type_ids']
     [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1]
 
 The first sequence, the "context" used for the question, has all its tokens represented by :obj:`0`, whereas the

docs/source/multilingual.rst

@@ -36,10 +36,11 @@ Here is an example using the ``xlm-clm-enfr-1024`` checkpoint (Causal language m
 
 .. code-block::
 
-    import torch
-    from transformers import XLMTokenizer, XLMWithLMHeadModel
+    >>> import torch
+    >>> from transformers import XLMTokenizer, XLMWithLMHeadModel
 
-    tokenizer = XLMTokenizer.from_pretrained("xlm-clm-1024-enfr")
+    >>> tokenizer = XLMTokenizer.from_pretrained("xlm-clm-enfr-1024")
+    >>> model = XLMWithLMHeadModel.from_pretrained("xlm-clm-enfr-1024")
 
 
 The different languages this model/tokenizer handles, as well as the ids of these languages are visible using the
@@ -47,16 +48,15 @@ The different languages this model/tokenizer handles, as well as the ids of thes
 
 .. code-block::
 
-    # Continuation of the previous script
-    print(tokenizer.lang2id)  # {'en': 0, 'fr': 1}
+    >>> print(tokenizer.lang2id)
+    {'en': 0, 'fr': 1}
 
 
 These ids should be used when passing a language parameter during a model pass. Let's define our inputs:
 
 .. code-block::
 
-    # Continuation of the previous script
-    input_ids = torch.tensor([tokenizer.encode("Wikipedia was used to")]) # batch size of 1
+    >>> input_ids = torch.tensor([tokenizer.encode("Wikipedia was used to")]) # batch size of 1
 
 
 We should now define the language embedding by using the previously defined language id. We want to create a tensor
@@ -64,20 +64,18 @@ filled with the appropriate language ids, of the same size as input_ids. For eng
 
 .. code-block::
 
-    # Continuation of the previous script
-    language_id = tokenizer.lang2id['en']  # 0
-    langs = torch.tensor([language_id] * input_ids.shape[1])  # torch.tensor([0, 0, 0, ..., 0])
+    >>> language_id = tokenizer.lang2id['en']  # 0
+    >>> langs = torch.tensor([language_id] * input_ids.shape[1])  # torch.tensor([0, 0, 0, ..., 0])
 
-    # We reshape it to be of size (batch_size, sequence_length)
-    langs = langs.view(1, -1) # is now of shape [1, sequence_length] (we have a batch size of 1)
+    >>> # We reshape it to be of size (batch_size, sequence_length)
+    >>> langs = langs.view(1, -1) # is now of shape [1, sequence_length] (we have a batch size of 1)
 
 
 You can then feed it all as input to your model:
 
 .. code-block::
 
-    # Continuation of the previous script
-    outputs = model(input_ids, langs=langs)
+    >>> outputs = model(input_ids, langs=langs)
 
 
 The example `run_generation.py <https://github.com/huggingface/transformers/blob/master/examples/text-generation/run_generation.py>`__

docs/source/quicktour.rst

@@ -32,40 +32,33 @@ provides the following tasks out of the box:
 Let's see how this work for sentiment analysis (the other tasks are all covered in the
 :doc:`task summary </task_summary>`):
 
-::
+.. code-block::
 
-    from transformers import pipeline
-    classifier = pipeline('sentiment-analysis')
+    >>> from transformers import pipeline
+    >>> classifier = pipeline('sentiment-analysis')
 
 When typing this command for the first time, a pretrained model and its tokenizer are downloaded and cached. We will
 look at both later on, but as an introduction the tokenizer's job is to preprocess the text for the model, which is
 then responsible for making predictions. The pipeline groups all of that together, and post-process the predictions to
-make them readable. For instance
-
-::
+make them readable. For instance:
 
-    classifier('We are very happy to show you the 🤗 Transformers library.')
 
-will return something like this:
-
-::
+.. code-block::
 
+    >>> classifier('We are very happy to show you the 🤗 Transformers library.')
     [{'label': 'POSITIVE', 'score': 0.9997795224189758}]
 
 That's encouraging! You can use it on a list of sentences, which will be preprocessed then fed to the model as a
-`batch`:
-
-::
+`batch`, returning a list of dictionaries like this one:
 
-    classifier(["We are very happy to show you the 🤗 Transformers library.",
-                "We hope you don't hate it."])
+.. code-block::
 
-returning a list of dictionaries like this one:
-
-::
-
-    [{'label': 'POSITIVE', 'score': 0.9997795224189758},
-     {'label': 'NEGATIVE', 'score': 0.5308589935302734}]
+    >>> results = classifier(["We are very happy to show you the 🤗 Transformers library.",
+    ...            "We hope you don't hate it."])
+    >>> for result in results:
+    ...     print(f"label: {result['label']}, with score: {round(result['score'], 4)}")
+    label: POSITIVE, with score: 0.9998
+    label: NEGATIVE, with score: 0.5309
 
 You can see the second sentence has been classified as negative (it needs to be positive or negative) but its score is
 fairly neutral.
@@ -83,9 +76,9 @@ see how we can use it.
 
 You can directly pass the name of the model to use to :func:`~transformers.pipeline`:
 
-::
+.. code-block::
 
-    classifier = pipeline('sentiment-analysis', model="nlptown/bert-base-multilingual-uncased-sentiment")
+    >>> classifier = pipeline('sentiment-analysis', model="nlptown/bert-base-multilingual-uncased-sentiment")
 
 This classifier can now deal with texts in English, French, but also Dutch, German, Italian and Spanish! You can also
 replace that name by a local folder where you have saved a pretrained model (see below). You can also pass a model
@@ -98,29 +91,30 @@ tokenizer associated to the model we picked and instantiate it. The second is
 the model itself. Note that if we were using the library on an other task, the class of the model would change. The
 :doc:`task summary </task_summary>` tutorial summarizes which class is used for which task.
 
-::
+.. code-block::
 
-    ## PYTORCH CODE
-    from transformers import AutoTokenizer, AutoModelForSequenceClassification
-    ## TENSORFLOW CODE
-    from transformers import AutoTokenizer, TFAutoModelForSequenceClassification
+    >>> ## PYTORCH CODE
+    >>> from transformers import AutoTokenizer, AutoModelForSequenceClassification
+    >>> ## TENSORFLOW CODE
+    >>> from transformers import AutoTokenizer, TFAutoModelForSequenceClassification
 
 Now, to download the models and tokenizer we found previously, we just have to use the
 :func:`~transformers.AutoModelForSequenceClassification.from_pretrained` method (feel free to replace ``model_name`` by
 any other model from the model hub):
 
-::
+.. code-block::
 
-    ## PYTORCH CODE
-    model_name = "nlptown/bert-base-multilingual-uncased-sentiment"
-    model = AutoModelForSequenceClassification.from_pretrained(model_name)
-    tokenizer = AutoTokenizer.from_pretrained(model_name)
-    pipe = pipeline('sentiment-analysis', model=model, tokenizer=tokenizer)
-    ## TENSORFLOW CODE
-    model_name = "nlptown/bert-base-multilingual-uncased-sentiment"
-    model = TFAutoModelForSequenceClassification.from_pretrained(model_name)
-    tokenizer = AutoTokenizer.from_pretrained(model_name)
-    classifier = pipeline('sentiment-analysis', model=model, tokenizer=tokenizer)
+    >>> ## PYTORCH CODE
+    >>> model_name = "nlptown/bert-base-multilingual-uncased-sentiment"
+    >>> model = AutoModelForSequenceClassification.from_pretrained(model_name)
+    >>> tokenizer = AutoTokenizer.from_pretrained(model_name)
+    >>> pipe = pipeline('sentiment-analysis', model=model, tokenizer=tokenizer)
+    >>> ## TENSORFLOW CODE
+    >>> model_name = "nlptown/bert-base-multilingual-uncased-sentiment"
+    >>> # This model only exists in PyTorch, so we use the `from_pt` flag to import that model in TensorFlow.
+    >>> model = TFAutoModelForSequenceClassification.from_pretrained(model_name, from_pt=True) 
+    >>> tokenizer = AutoTokenizer.from_pretrained(model_name)
+    >>> classifier = pipeline('sentiment-analysis', model=model, tokenizer=tokenizer)
 
 If you don't find a model that has been pretrained on some data similar to yours, you will need to fine-tune a
 pretrained model on your data. We provide :doc:`example scripts </examples>` to do so. Once you're done, don't forget
@@ -136,16 +130,16 @@ using the :obj:`from_pretrained` method:
 
 ::
 
-    ## PYTORCH CODE
-    from transformers import AutoTokenizer, AutoModelForSequenceClassification
-    model_name = "distilbert-base-uncased-finetuned-sst-2-english"
-    model = AutoModelForSequenceClassification.from_pretrained(model_name)
-    tokenizer = AutoTokenizer.from_pretrained(model_name)
-    ## TENSORFLOW CODE
-    from transformers import AutoTokenizer, TFAutoModelForSequenceClassification
-    model_name = "distilbert-base-uncased-finetuned-sst-2-english"
-    model = TFAutoModelForSequenceClassification.from_pretrained(model_name)
-    tokenizer = AutoTokenizer.from_pretrained(model_name)
+    >>> ## PYTORCH CODE
+    >>> from transformers import AutoTokenizer, AutoModelForSequenceClassification
+    >>> model_name = "distilbert-base-uncased-finetuned-sst-2-english"
+    >>> pt_model = AutoModelForSequenceClassification.from_pretrained(model_name)
+    >>> tokenizer = AutoTokenizer.from_pretrained(model_name)
+    >>> ## TENSORFLOW CODE
+    >>> from transformers import AutoTokenizer, TFAutoModelForSequenceClassification
+    >>> model_name = "distilbert-base-uncased-finetuned-sst-2-english"
+    >>> tf_model = TFAutoModelForSequenceClassification.from_pretrained(model_name)
+    >>> tokenizer = AutoTokenizer.from_pretrained(model_name)
 
 Using the tokenizer
 ^^^^^^^^^^^^^^^^^^^
@@ -161,48 +155,56 @@ the model. To do this, the tokenizer has a `vocab`, which is the part we downloa
 
 To apply these steps on a given text, we can just feed it to our tokenizer:
 
-::
+.. code-block::
 
-    input = tokenizer("We are very happy to show you the 🤗 Transformers library.")
-    print(input)
+    >>> inputs = tokenizer("We are very happy to show you the 🤗 Transformers library.")
 
 This returns a dictionary string to list of ints. It contains the `ids of the tokens <glossary.html#input-ids>`__,
 as mentioned before, but also additional arguments that will be useful to the model. Here for instance, we also have an
 `attention mask <glossary.html#attention-mask>`__ that the model will use to have a better understanding of the sequence:
 
 
-::
-    {'input_ids': [101, 2057, 2024, 2200, 3407, 2000, 2265, 2017, 1996, 100, 19081, 3075, 1012, 102],
-     'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]}
+.. code-block::
+
+    >>> print(inputs)
+    {'input_ids': [101, 2057, 2024, 2200, 3407, 2000, 2265, 2017, 1996, 100, 19081, 3075, 1012, 102], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]}
 
 You can pass a list of sentences directly to your tokenizer. If your goal is to send them through your model as a
 batch, you probably want to pad them all to the same length, truncate them to the maximum length the model can accept
 and get tensors back. You can specify all of that to the tokenizer:
 
-::
-
-    ## PYTORCH CODE
-    batch = tokenizer(
-        ["We are very happy to show you the 🤗 Transformers library.",
-         "We hope you don't hate it."],
-        padding=True, truncation=True, return_tensors="pt")
-    print(batch)
-    ## TENSORFLOW CODE
-    batch = tokenizer(
-        ["We are very happy to show you the 🤗 Transformers library.",
-         "We hope you don't hate it."],
-        padding=True, truncation=True, return_tensors="tf")
-    print(batch)
+.. code-block::
+
+    >>> ## PYTORCH CODE
+    >>> pt_batch = tokenizer(
+    ...     ["We are very happy to show you the 🤗 Transformers library.", "We hope you don't hate it."],
+    ...     padding=True,
+    ...     truncation=True,
+    ...     return_tensors="pt"
+    ... )
+    >>> ## TENSORFLOW CODE
+    >>> tf_batch = tokenizer(
+    ...     ["We are very happy to show you the 🤗 Transformers library.", "We hope you don't hate it."],
+    ...     padding=True,
+    ...     truncation=True,
+    ...     return_tensors="tf"
+    ... )
 
 The padding is automatically applied on the side the model expect it (in this case, on the right), with the
 padding token the model was pretrained with. The attention mask is also adapted to take the padding into account:
 
-::
+.. code-block::
 
-    {'input_ids': tensor([[  101,  2057,  2024,  2200,  3407,  2000,  2265,  2017,  1996,   100, 19081,  3075,  1012,   102],
-                          [  101,  2057,  3246,  2017,  2123,  1005,  1056,  5223,  2009,  1012,   102,     0,     0,     0]]), 
-     'attention_mask': tensor([[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1],
-                               [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0]])}
+    >>> ## PYTORCH CODE
+    >>> for key, value in pt_batch.items():
+    ...     print(f"{key}: {value.numpy().tolist()}")
+    input_ids: [[101, 2057, 2024, 2200, 3407, 2000, 2265, 2017, 1996, 100, 19081, 3075, 1012, 102], [101, 2057, 3246, 2017, 2123, 1005, 1056, 5223, 2009, 1012, 102, 0, 0, 0]]
+    attention_mask: [[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0]]
+    >>> ## TENSORFLOW CODE
+    >>> for key, value in tf_batch.items():
+    ...     print(f"{key}: {value.numpy().tolist()}")
+    input_ids: [[101, 2057, 2024, 2200, 3407, 2000, 2265, 2017, 1996, 100, 19081, 3075, 1012, 102], [101, 2057, 3246, 2017, 2123, 1005, 1056, 5223, 2009, 1012, 102, 0, 0, 0]]
+    attention_mask: [[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0]]
 
 You can learn more about tokenizers :doc:`here <preprocessing>`.
 
@@ -213,20 +215,27 @@ Once your input has been preprocessed by the tokenizer, you can directly send it
 contain all the relevant information the model needs. If you're using a TensorFlow model, you can directly pass the
 dictionary keys to tensor, for a PyTorch model, you need to unpack the dictionary by adding :obj:`**`.
 
-::
+.. code-block::
 
-    ## PYTORCH CODE
-    outputs = model(**batch)
-    ## TENSORFLOW CODE
-    outputs = model(batch)
+    >>> ## PYTORCH CODE
+    >>> pt_outputs = pt_model(**pt_batch)
+    >>> ## TENSORFLOW CODE
+    >>> tf_outputs = tf_model(tf_batch)
 
 In 🤗 Transformers, all outputs are tuples (with only one element potentially). Here, we get a tuple with just the
 final activations of the model.
 
-::
+.. code-block::
 
-    (tensor([[-4.1329,  4.3811],
-             [ 0.0818, -0.0418]]),)
+    >>> ## PYTORCH CODE
+    >>> print(pt_outputs)
+    (tensor([[-4.0833,  4.3364],
+            [ 0.0818, -0.0418]], grad_fn=<AddmmBackward>),)
+    >>> ## TENSORFLOW CODE
+    >>> print(tf_outputs)
+    (<tf.Tensor: shape=(2, 2), dtype=float32, numpy=
+    array([[-4.0832963 ,  4.3364134 ],
+           [ 0.08181238, -0.04178794]], dtype=float32)>,)
 
 .. note::
 
@@ -235,33 +244,39 @@ final activations of the model.
 
 Let's apply the SoftMax activation to get predictions.
 
-::
+.. code-block::
 
-    ## PYTORCH CODE
-    import torch.nn.functional as F
-    predictions = F.softmax(outputs[0], dim=-1)
-    print(predictions)
-    ## TENSORFLOW CODE
-    predictions = tf.nn.softmax(outputs[0], axis=-1)
-    print(predictions)
+    >>> ## PYTORCH CODE
+    >>> import torch.nn.functional as F
+    >>> pt_predictions = F.softmax(pt_outputs[0], dim=-1)
+    >>> ## TENSORFLOW CODE
+    >>> import tensorflow as tf
+    >>> tf_predictions = tf.nn.softmax(tf_outputs[0], axis=-1)
 
 We can see we get the numbers from before:
 
-::
+.. code-block::
 
-    tensor([[2.0060e-04, 9.9980e-01],
-            [5.3086e-01, 4.6914e-01]])
+    >>> ## TENSORFLOW CODE
+    >>> print(tf_predictions)
+    tf.Tensor(
+    [[2.2042994e-04 9.9977952e-01]
+     [5.3086078e-01 4.6913919e-01]], shape=(2, 2), dtype=float32)
+    >>> ## PYTORCH CODE
+    >>> print(pt_predictions)
+    tensor([[2.2043e-04, 9.9978e-01],
+            [5.3086e-01, 4.6914e-01]], grad_fn=<SoftmaxBackward>)
 
 If you have labels, you can provide them to the model, it will return a tuple with the loss and the final activations.
 
-::
+.. code-block::
 
-    ## PYTORCH CODE
-    import torch
-    outputs = model(**batch, labels = torch.tensor([1, 0])
-    ## TENSORFLOW CODE
-    import tensorflow as tf
-    outputs = model(batch, labels = tf.constant([1, 0])
+    >>> ## PYTORCH CODE
+    >>> import torch
+    >>> pt_outputs = pt_model(**pt_batch, labels = torch.tensor([1, 0]))
+    >>> ## TENSORFLOW CODE
+    >>> import tensorflow as tf
+    >>> tf_outputs = tf_model(tf_batch, labels = tf.constant([1, 0]))
 
 Models are standard `torch.nn.Module <https://pytorch.org/docs/stable/nn.html#torch.nn.Module>`__ or
 `tf.keras.Model <https://www.tensorflow.org/api_docs/python/tf/keras/Model>`__ so you can use them in your usual
@@ -298,12 +313,12 @@ Lastly, you can also ask the model to return all hidden states and all attention
 
 ::
 
-    ## PYTORCH CODE
-    outputs = model(**batch, output_hidden_states=True, output_attentions=True)
-    all_hidden_states, all_attentions = outputs[-2:]
-    ## TENSORFLOW CODE
-    outputs = model(batch, output_hidden_states=True, output_attentions=True)
-    all_hidden_states, all_attentions = outputs[-2:]
+    >>> ## PYTORCH CODE
+    >>> pt_outputs = pt_model(**pt_batch, output_hidden_states=True, output_attentions=True)
+    >>> all_hidden_states, all_attentions = pt_outputs[-2:]
+    >>> ## TENSORFLOW CODE
+    >>> tf_outputs = tf_model(tf_batch, output_hidden_states=True, output_attentions=True)
+    >>> all_hidden_states, all_attentions = tf_outputs[-2:]
 
 Accessing the code
 ^^^^^^^^^^^^^^^^^^
@@ -318,18 +333,18 @@ using the :doc:`DistilBERT </model_doc/distilbert>` architecture. The model auto
 to that specific model, or browse the source code. This is how you would directly instantiate model and tokenizer
 without the auto magic:
 
-::
+.. code-block::
 
-    ## PYTORCH CODE
-    from transformers import DistilBertTokenizer, DistilBertForSequenceClassification
-    model_name = "distilbert-base-uncased-finetuned-sst-2-english"
-    model = DistilBertForSequenceClassification.from_pretrained(model_name)
-    tokenizer = DistilBertTokenizer.from_pretrained(model_name)
-    ## TENSORFLOW CODE
-    from transformers import DistilBertTokenizer, TFDistilBertForSequenceClassification
-    model_name = "distilbert-base-uncased-finetuned-sst-2-english"
-    model = TFDistilBertForSequenceClassification.from_pretrained(model_name)
-    tokenizer = DistilBertTokenizer.from_pretrained(model_name)
+    >>> ## PYTORCH CODE
+    >>> from transformers import DistilBertTokenizer, DistilBertForSequenceClassification
+    >>> model_name = "distilbert-base-uncased-finetuned-sst-2-english"
+    >>> model = DistilBertForSequenceClassification.from_pretrained(model_name)
+    >>> tokenizer = DistilBertTokenizer.from_pretrained(model_name)
+    >>> ## TENSORFLOW CODE
+    >>> from transformers import DistilBertTokenizer, TFDistilBertForSequenceClassification
+    >>> model_name = "distilbert-base-uncased-finetuned-sst-2-english"
+    >>> model = TFDistilBertForSequenceClassification.from_pretrained(model_name)
+    >>> tokenizer = DistilBertTokenizer.from_pretrained(model_name)
 
 Customizing the model
 ^^^^^^^^^^^^^^^^^^^^^
@@ -345,18 +360,18 @@ Here we use the predefined vocabulary of DistilBERT (hence load the tokenizer wi
 instantiate the model from the configuration instead of using the
 :func:`~transformers.DistilBertForSequenceClassification.from_pretrained` method).
 
-::
+.. code-block::
 
-    ## PYTORCH CODE
-    from transformers import DistilBertConfig, DistilBertTokenizer, DistilBertForSequenceClassification
-    config = DistilBertConfig(n_heads=8, dim=512, hidden_dim=4*512)
-    tokenizer = DistilBertTokenizer.from_pretrained('distilbert-base-uncased')
-    model = DistilBertForSequenceClassification(config)
-    ## TENSORFLOW CODE
-    from transformers import DistilBertConfig, DistilBertTokenizer, TFDistilBertForSequenceClassification
-    config = DistilBertConfig(n_heads=8, dim=512, hidden_dim=4*512)
-    tokenizer = DistilBertTokenizer.from_pretrained('distilbert-base-uncased')
-    model = TFDistilBertForSequenceClassification(config)
+    >>> ## PYTORCH CODE
+    >>> from transformers import DistilBertConfig, DistilBertTokenizer, DistilBertForSequenceClassification
+    >>> config = DistilBertConfig(n_heads=8, dim=512, hidden_dim=4*512)
+    >>> tokenizer = DistilBertTokenizer.from_pretrained('distilbert-base-uncased')
+    >>> model = DistilBertForSequenceClassification(config)
+    >>> ## TENSORFLOW CODE
+    >>> from transformers import DistilBertConfig, DistilBertTokenizer, TFDistilBertForSequenceClassification
+    >>> config = DistilBertConfig(n_heads=8, dim=512, hidden_dim=4*512)
+    >>> tokenizer = DistilBertTokenizer.from_pretrained('distilbert-base-uncased')
+    >>> model = TFDistilBertForSequenceClassification(config)
 
 For something that only changes the head of the model (for instance, the number of labels), you can still use a
 pretrained model for the body. For instance, let's define a classifier for 10 different labels using a pretrained body.
@@ -364,15 +379,15 @@ We could create a configuration with all the default values and just change the
 can directly pass any argument a configuration would take to the :func:`from_pretrained` method and it will update the
 default configuration with it:
 
-::
-
-    ## PYTORCH CODE
-    from transformers import DistilBertConfig, DistilBertTokenizer, DistilBertForSequenceClassification
-    model_name = "distilbert-base-uncased"
-    model = DistilBertForSequenceClassification.from_pretrained(model_name, num_labels=10)
-    tokenizer = DistilBertTokenizer.from_pretrained(model_name)
-    ## TENSORFLOW CODE
-    from transformers import DistilBertConfig, DistilBertTokenizer, TFDistilBertForSequenceClassification
-    model_name = "distilbert-base-uncased"
-    model = TFDistilBertForSequenceClassification.from_pretrained(model_name, num_labels=10)
-    tokenizer = DistilBertTokenizer.from_pretrained(model_name)
+.. code-block::
+
+    >>> ## PYTORCH CODE
+    >>> from transformers import DistilBertConfig, DistilBertTokenizer, DistilBertForSequenceClassification
+    >>> model_name = "distilbert-base-uncased"
+    >>> model = DistilBertForSequenceClassification.from_pretrained(model_name, num_labels=10)
+    >>> tokenizer = DistilBertTokenizer.from_pretrained(model_name)
+    >>> ## TENSORFLOW CODE
+    >>> from transformers import DistilBertConfig, DistilBertTokenizer, TFDistilBertForSequenceClassification
+    >>> model_name = "distilbert-base-uncased"
+    >>> model = TFDistilBertForSequenceClassification.from_pretrained(model_name, num_labels=10)
+    >>> tokenizer = DistilBertTokenizer.from_pretrained(model_name)

docs/source/task_summary.rst

@@ -50,21 +50,21 @@ a model on a GLUE sequence classification task, you may leverage the
 Here is an example using the pipelines do to sentiment analysis: identifying if a sequence is positive or negative.
 It leverages a fine-tuned model on sst2, which is a GLUE task.
 
-::
-
-    from transformers import pipeline
+This returns a label ("POSITIVE" or "NEGATIVE") alongside a score, as follows:
 
-    nlp = pipeline("sentiment-analysis")
+.. code-block::
 
-    print(nlp("I hate you"))
-    print(nlp("I love you"))
+    >>> from transformers import pipeline
 
-This returns a label ("POSITIVE" or "NEGATIVE") alongside a score, as follows:
+    >>> nlp = pipeline("sentiment-analysis")
 
-::
+    >>> result = nlp("I hate you")[0]
+    >>> print(f"label: {result['label']}, with score: {round(result['score'], 4)}")
+    label: NEGATIVE, with score: 0.9991
 
-    [{'label': 'NEGATIVE', 'score': 0.9991129}]
-    [{'label': 'POSITIVE', 'score': 0.99986565}]
+    >>> result = nlp("I love you")[0]
+    >>> print(f"label: {result['label']}, with score: {round(result['score'], 4)}")
+    label: POSITIVE, with score: 0.9999
 
 
 Here is an example of doing a sequence classification using a model to determine if two sequences are paraphrases
@@ -80,76 +80,72 @@ of each other. The process is the following:
 - Compute the softmax of the result to get probabilities over the classes
 - Print the results
 
-::
-
-    ## PYTORCH CODE
-    from transformers import AutoTokenizer, AutoModelForSequenceClassification
-    import torch
-
-    tokenizer = AutoTokenizer.from_pretrained("bert-base-cased-finetuned-mrpc")
-    model = AutoModelForSequenceClassification.from_pretrained("bert-base-cased-finetuned-mrpc")
-
-    classes = ["not paraphrase", "is paraphrase"]
+.. code-block::
 
-    sequence_0 = "The company HuggingFace is based in New York City"
-    sequence_1 = "Apples are especially bad for your health"
-    sequence_2 = "HuggingFace's headquarters are situated in Manhattan"
+    >>> ## PYTORCH CODE
+    >>> from transformers import AutoTokenizer, AutoModelForSequenceClassification
+    >>> import torch
 
-    paraphrase = tokenizer.encode_plus(sequence_0, sequence_2, return_tensors="pt")
-    not_paraphrase = tokenizer.encode_plus(sequence_0, sequence_1, return_tensors="pt")
+    >>> tokenizer = AutoTokenizer.from_pretrained("bert-base-cased-finetuned-mrpc")
+    >>> model = AutoModelForSequenceClassification.from_pretrained("bert-base-cased-finetuned-mrpc")
 
-    paraphrase_classification_logits = model(**paraphrase)[0]
-    not_paraphrase_classification_logits = model(**not_paraphrase)[0]
+    >>> classes = ["not paraphrase", "is paraphrase"]
 
-    paraphrase_results = torch.softmax(paraphrase_classification_logits, dim=1).tolist()[0]
-    not_paraphrase_results = torch.softmax(not_paraphrase_classification_logits, dim=1).tolist()[0]
+    >>> sequence_0 = "The company HuggingFace is based in New York City"
+    >>> sequence_1 = "Apples are especially bad for your health"
+    >>> sequence_2 = "HuggingFace's headquarters are situated in Manhattan"
 
-    print("Should be paraphrase")
-    for i in range(len(classes)):
-        print(f"{classes[i]}: {round(paraphrase_results[i] * 100)}%")
+    >>> paraphrase = tokenizer.encode_plus(sequence_0, sequence_2, return_tensors="pt")
+    >>> not_paraphrase = tokenizer.encode_plus(sequence_0, sequence_1, return_tensors="pt")
 
-    print("\nShould not be paraphrase")
-    for i in range(len(classes)):
-        print(f"{classes[i]}: {round(not_paraphrase_results[i] * 100)}%")
-    ## TENSORFLOW CODE
-    from transformers import AutoTokenizer, TFAutoModelForSequenceClassification
-    import tensorflow as tf
+    >>> paraphrase_classification_logits = model(**paraphrase)[0]
+    >>> not_paraphrase_classification_logits = model(**not_paraphrase)[0]
 
-    tokenizer = AutoTokenizer.from_pretrained("bert-base-cased-finetuned-mrpc")
-    model = TFAutoModelForSequenceClassification.from_pretrained("bert-base-cased-finetuned-mrpc")
+    >>> paraphrase_results = torch.softmax(paraphrase_classification_logits, dim=1).tolist()[0]
+    >>> not_paraphrase_results = torch.softmax(not_paraphrase_classification_logits, dim=1).tolist()[0]
 
-    classes = ["not paraphrase", "is paraphrase"]
-
-    sequence_0 = "The company HuggingFace is based in New York City"
-    sequence_1 = "Apples are especially bad for your health"
-    sequence_2 = "HuggingFace's headquarters are situated in Manhattan"
+    >>> # Should be paraphrase
+    >>> for i in range(len(classes)):
+    ...     print(f"{classes[i]}: {int(round(paraphrase_results[i] * 100))}%")
+    not paraphrase: 10%
+    is paraphrase: 90%
 
-    paraphrase = tokenizer.encode_plus(sequence_0, sequence_2, return_tensors="tf")
-    not_paraphrase = tokenizer.encode_plus(sequence_0, sequence_1, return_tensors="tf")
+    >>> # Should not be paraphrase
+    >>> for i in range(len(classes)):
+    ...     print(f"{classes[i]}: {int(round(not_paraphrase_results[i] * 100))}%")
+    not paraphrase: 94%
+    is paraphrase: 6%
+    >>> ## TENSORFLOW CODE
+    >>> from transformers import AutoTokenizer, TFAutoModelForSequenceClassification
+    >>> import tensorflow as tf
 
-    paraphrase_classification_logits = model(paraphrase)[0]
-    not_paraphrase_classification_logits = model(not_paraphrase)[0]
+    >>> tokenizer = AutoTokenizer.from_pretrained("bert-base-cased-finetuned-mrpc")
+    >>> model = TFAutoModelForSequenceClassification.from_pretrained("bert-base-cased-finetuned-mrpc")
 
-    paraphrase_results = tf.nn.softmax(paraphrase_classification_logits, axis=1).numpy()[0]
-    not_paraphrase_results = tf.nn.softmax(not_paraphrase_classification_logits, axis=1).numpy()[0]
+    >>> classes = ["not paraphrase", "is paraphrase"]
 
-    print("Should be paraphrase")
-    for i in range(len(classes)):
-        print(f"{classes[i]}: {round(paraphrase_results[i] * 100)}%")
+    >>> sequence_0 = "The company HuggingFace is based in New York City"
+    >>> sequence_1 = "Apples are especially bad for your health"
+    >>> sequence_2 = "HuggingFace's headquarters are situated in Manhattan"
 
-    print("\nShould not be paraphrase")
-    for i in range(len(classes)):
-        print(f"{classes[i]}: {round(not_paraphrase_results[i] * 100)}%")
+    >>> paraphrase = tokenizer.encode_plus(sequence_0, sequence_2, return_tensors="tf")
+    >>> not_paraphrase = tokenizer.encode_plus(sequence_0, sequence_1, return_tensors="tf")
 
-This outputs the following results:
+    >>> paraphrase_classification_logits = model(paraphrase)[0]
+    >>> not_paraphrase_classification_logits = model(not_paraphrase)[0]
 
-::
+    >>> paraphrase_results = tf.nn.softmax(paraphrase_classification_logits, axis=1).numpy()[0]
+    >>> not_paraphrase_results = tf.nn.softmax(not_paraphrase_classification_logits, axis=1).numpy()[0]
 
-    Should be paraphrase
+    >>> # Should be paraphrase
+    >>> for i in range(len(classes)):
+    ...     print(f"{classes[i]}: {int(round(paraphrase_results[i] * 100))}%")
     not paraphrase: 10%
     is paraphrase: 90%
 
-    Should not be paraphrase
+    >>> # Should not be paraphrase
+    >>> for i in range(len(classes)):
+    ...     print(f"{classes[i]}: {int(round(not_paraphrase_results[i] * 100))}%")
     not paraphrase: 94%
     is paraphrase: 6%
 
@@ -163,28 +159,30 @@ a model on a SQuAD task, you may leverage the `run_squad.py`.
 Here is an example using the pipelines do to question answering: extracting an answer from a text given a question.
 It leverages a fine-tuned model on SQuAD.
 
-::
-
-    from transformers import pipeline
+.. code-block::
 
-    nlp = pipeline("question-answering")
+    >>> from transformers import pipeline
 
-    context = r"""
-    Extractive Question Answering is the task of extracting an answer from a text given a question. An example of a
-    question answering dataset is the SQuAD dataset, which is entirely based on that task. If you would like to fine-tune
-    a model on a SQuAD task, you may leverage the `run_squad.py`.
-    """
+    >>> nlp = pipeline("question-answering")
 
-    print(nlp(question="What is extractive question answering?", context=context))
-    print(nlp(question="What is a good example of a question answering dataset?", context=context))
+    >>> context = r"""
+    ... Extractive Question Answering is the task of extracting an answer from a text given a question. An example of a
+    ... question answering dataset is the SQuAD dataset, which is entirely based on that task. If you would like to fine-tune
+    ... a model on a SQuAD task, you may leverage the examples/question-answering/run_squad.py script.
+    ... """
 
 This returns an answer extracted from the text, a confidence score, alongside "start" and "end" values which
 are the positions of the extracted answer in the text.
 
-::
+.. code-block::
 
-    {'score': 0.622232091629833, 'start': 34, 'end': 96, 'answer': 'the task of extracting an answer from a text given a question.'}
-    {'score': 0.5115299158662765, 'start': 147, 'end': 161, 'answer': 'SQuAD dataset,'}
+    >>> result = nlp(question="What is extractive question answering?", context=context)
+    >>> print(f"Answer: '{result['answer']}', score: {round(result['score'], 4)}, start: {result['start']}, end: {result['end']}")
+    Answer: 'the task of extracting an answer from a text given a question.', score: 0.6226, start: 34, end: 96
+
+    >>> result = nlp(question="What is a good example of a question answering dataset?", context=context)
+    >>> print(f"Answer: '{result['answer']}', score: {round(result['score'], 4)}, start: {result['start']}, end: {result['end']}")
+    Answer: 'SQuAD dataset,', score: 0.5053, start: 147, end: 161
 
 
 Here is an example of question answering using a model and a tokenizer. The process is the following:
@@ -200,92 +198,91 @@ Here is an example of question answering using a model and a tokenizer. The proc
 - Fetch the tokens from the identified start and stop values, convert those tokens to a string.
 - Print the results
 
-::
-
-    ## PYTORCH CODE
-    from transformers import AutoTokenizer, AutoModelForQuestionAnswering
-    import torch
-
-    tokenizer = AutoTokenizer.from_pretrained("bert-large-uncased-whole-word-masking-finetuned-squad")
-    model = AutoModelForQuestionAnswering.from_pretrained("bert-large-uncased-whole-word-masking-finetuned-squad")
-
-    text = r"""
-    🤗 Transformers (formerly known as pytorch-transformers and pytorch-pretrained-bert) provides general-purpose
-    architectures (BERT, GPT-2, RoBERTa, XLM, DistilBert, XLNet…) for Natural Language Understanding (NLU) and Natural
-    Language Generation (NLG) with over 32+ pretrained models in 100+ languages and deep interoperability between
-    TensorFlow 2.0 and PyTorch.
-    """
-
-    questions = [
-        "How many pretrained models are available in 🤗 Transformers?",
-        "What does 🤗 Transformers provide?",
-        "🤗 Transformers provides interoperability between which frameworks?",
-    ]
-
-    for question in questions:
-        inputs = tokenizer.encode_plus(question, text, add_special_tokens=True, return_tensors="pt")
-        input_ids = inputs["input_ids"].tolist()[0]
-
-        text_tokens = tokenizer.convert_ids_to_tokens(input_ids)
-        answer_start_scores, answer_end_scores = model(**inputs)
-
-        answer_start = torch.argmax(
-            answer_start_scores
-        )  # Get the most likely beginning of answer with the argmax of the score
-        answer_end = torch.argmax(answer_end_scores) + 1  # Get the most likely end of answer with the argmax of the score
-
-        answer = tokenizer.convert_tokens_to_string(tokenizer.convert_ids_to_tokens(input_ids[answer_start:answer_end]))
-
-        print(f"Question: {question}")
-        print(f"Answer: {answer}\n")
-    ## TENSORFLOW CODE
-    from transformers import AutoTokenizer, TFAutoModelForQuestionAnswering
-    import tensorflow as tf
-
-    tokenizer = AutoTokenizer.from_pretrained("bert-large-uncased-whole-word-masking-finetuned-squad")
-    model = TFAutoModelForQuestionAnswering.from_pretrained("bert-large-uncased-whole-word-masking-finetuned-squad")
-
-    text = r"""
-    🤗 Transformers (formerly known as pytorch-transformers and pytorch-pretrained-bert) provides general-purpose
-    architectures (BERT, GPT-2, RoBERTa, XLM, DistilBert, XLNet…) for Natural Language Understanding (NLU) and Natural
-    Language Generation (NLG) with over 32+ pretrained models in 100+ languages and deep interoperability between
-    TensorFlow 2.0 and PyTorch.
-    """
-
-    questions = [
-        "How many pretrained models are available in 🤗 Transformers?",
-        "What does 🤗 Transformers provide?",
-        "🤗 Transformers provides interoperability between which frameworks?",
-    ]
-
-    for question in questions:
-        inputs = tokenizer.encode_plus(question, text, add_special_tokens=True, return_tensors="tf")
-        input_ids = inputs["input_ids"].numpy()[0]
-
-        text_tokens = tokenizer.convert_ids_to_tokens(input_ids)
-        answer_start_scores, answer_end_scores = model(inputs)
-
-        answer_start = tf.argmax(
-            answer_start_scores, axis=1
-        ).numpy()[0]  # Get the most likely beginning of answer with the argmax of the score
-        answer_end = (
-            tf.argmax(answer_end_scores, axis=1) + 1
-        ).numpy()[0]  # Get the most likely end of answer with the argmax of the score
-        answer = tokenizer.convert_tokens_to_string(tokenizer.convert_ids_to_tokens(input_ids[answer_start:answer_end]))
-
-        print(f"Question: {question}")
-        print(f"Answer: {answer}\n")
-
-This outputs the questions followed by the predicted answers:
-
-::
-
+.. code-block::
+
+    >>> ## PYTORCH CODE
+    >>> from transformers import AutoTokenizer, AutoModelForQuestionAnswering
+    >>> import torch
+
+    >>> tokenizer = AutoTokenizer.from_pretrained("bert-large-uncased-whole-word-masking-finetuned-squad")
+    >>> model = AutoModelForQuestionAnswering.from_pretrained("bert-large-uncased-whole-word-masking-finetuned-squad")
+
+    >>> text = r"""
+    ... 🤗 Transformers (formerly known as pytorch-transformers and pytorch-pretrained-bert) provides general-purpose
+    ... architectures (BERT, GPT-2, RoBERTa, XLM, DistilBert, XLNet…) for Natural Language Understanding (NLU) and Natural
+    ... Language Generation (NLG) with over 32+ pretrained models in 100+ languages and deep interoperability between
+    ... TensorFlow 2.0 and PyTorch.
+    ... """
+
+    >>> questions = [
+    ...     "How many pretrained models are available in 🤗 Transformers?",
+    ...     "What does 🤗 Transformers provide?",
+    ...     "🤗 Transformers provides interoperability between which frameworks?",
+    ... ]
+
+    >>> for question in questions:
+    ...     inputs = tokenizer.encode_plus(question, text, add_special_tokens=True, return_tensors="pt")
+    ...     input_ids = inputs["input_ids"].tolist()[0]
+    ...
+    ...     text_tokens = tokenizer.convert_ids_to_tokens(input_ids)
+    ...     answer_start_scores, answer_end_scores = model(**inputs)
+    ...
+    ...     answer_start = torch.argmax(
+    ...         answer_start_scores
+    ...     )  # Get the most likely beginning of answer with the argmax of the score
+    ...     answer_end = torch.argmax(answer_end_scores) + 1  # Get the most likely end of answer with the argmax of the score
+    ...
+    ...     answer = tokenizer.convert_tokens_to_string(tokenizer.convert_ids_to_tokens(input_ids[answer_start:answer_end]))
+    ...
+    ...     print(f"Question: {question}")
+    ...     print(f"Answer: {answer}")
+    Question: How many pretrained models are available in 🤗 Transformers?
+    Answer: over 32 +
+    Question: What does 🤗 Transformers provide?
+    Answer: general - purpose architectures
+    Question: 🤗 Transformers provides interoperability between which frameworks?
+    Answer: tensorflow 2 . 0 and pytorch
+    >>> ## TENSORFLOW CODE
+    >>> from transformers import AutoTokenizer, TFAutoModelForQuestionAnswering
+    >>> import tensorflow as tf
+
+    >>> tokenizer = AutoTokenizer.from_pretrained("bert-large-uncased-whole-word-masking-finetuned-squad")
+    >>> model = TFAutoModelForQuestionAnswering.from_pretrained("bert-large-uncased-whole-word-masking-finetuned-squad")
+
+    >>> text = r"""
+    ... 🤗 Transformers (formerly known as pytorch-transformers and pytorch-pretrained-bert) provides general-purpose
+    ... architectures (BERT, GPT-2, RoBERTa, XLM, DistilBert, XLNet…) for Natural Language Understanding (NLU) and Natural
+    ... Language Generation (NLG) with over 32+ pretrained models in 100+ languages and deep interoperability between
+    ... TensorFlow 2.0 and PyTorch.
+    ... """
+
+    >>> questions = [
+    ...     "How many pretrained models are available in 🤗 Transformers?",
+    ...     "What does 🤗 Transformers provide?",
+    ...     "🤗 Transformers provides interoperability between which frameworks?",
+    ... ]
+
+    >>> for question in questions:
+    ...     inputs = tokenizer.encode_plus(question, text, add_special_tokens=True, return_tensors="tf")
+    ...     input_ids = inputs["input_ids"].numpy()[0]
+    ...
+    ...     text_tokens = tokenizer.convert_ids_to_tokens(input_ids)
+    ...     answer_start_scores, answer_end_scores = model(inputs)
+    ...
+    ...     answer_start = tf.argmax(
+    ...         answer_start_scores, axis=1
+    ...     ).numpy()[0]  # Get the most likely beginning of answer with the argmax of the score
+    ...     answer_end = (
+    ...         tf.argmax(answer_end_scores, axis=1) + 1
+    ...     ).numpy()[0]  # Get the most likely end of answer with the argmax of the score
+    ...     answer = tokenizer.convert_tokens_to_string(tokenizer.convert_ids_to_tokens(input_ids[answer_start:answer_end]))
+    ...
+    ...     print(f"Question: {question}")
+    ...     print(f"Answer: {answer}")
     Question: How many pretrained models are available in 🤗 Transformers?
     Answer: over 32 +
-
     Question: What does 🤗 Transformers provide?
     Answer: general - purpose architectures
-
     Question: 🤗 Transformers provides interoperability between which frameworks?
     Answer: tensorflow 2 . 0 and pytorch
 
@@ -313,25 +310,44 @@ see `Lewis, Lui, Goyal et al. <https://arxiv.org/abs/1910.13461>`__, part 4.2).
 
 Here is an example of using pipelines to replace a mask from a sequence:
 
-::
+.. code-block::
 
-    from transformers import pipeline
+    >>> from transformers import pipeline
 
-    nlp = pipeline("fill-mask")
-    print(nlp(f"HuggingFace is creating a {nlp.tokenizer.mask_token} that the community uses to solve NLP tasks."))
+    >>> nlp = pipeline("fill-mask")
 
 This outputs the sequences with the mask filled, the confidence score as well as the token id in the tokenizer
 vocabulary:
 
-::
-
-    [
-        {'sequence': '<s> HuggingFace is creating a tool that the community uses to solve NLP tasks.</s>', 'score': 0.15627853572368622, 'token': 3944},
-        {'sequence': '<s> HuggingFace is creating a framework that the community uses to solve NLP tasks.</s>', 'score': 0.11690319329500198, 'token': 7208},
-        {'sequence': '<s> HuggingFace is creating a library that the community uses to solve NLP tasks.</s>', 'score': 0.058063216507434845, 'token': 5560},
-        {'sequence': '<s> HuggingFace is creating a database that the community uses to solve NLP tasks.</s>', 'score': 0.04211743175983429, 'token': 8503},
-        {'sequence': '<s> HuggingFace is creating a prototype that the community uses to solve NLP tasks.</s>', 'score': 0.024718601256608963, 'token': 17715}
-    ]
+.. code-block::
+
+    >>> from pprint import pprint
+    >>> pprint(nlp(f"HuggingFace is creating a {nlp.tokenizer.mask_token} that the community uses to solve NLP tasks."))
+    [{'score': 0.1792745739221573,
+      'sequence': '<s>HuggingFace is creating a tool that the community uses to '
+                  'solve NLP tasks.</s>',
+      'token': 3944,
+      'token_str': 'Ġtool'},
+     {'score': 0.11349421739578247,
+      'sequence': '<s>HuggingFace is creating a framework that the community uses '
+                  'to solve NLP tasks.</s>',
+      'token': 7208,
+      'token_str': 'Ġframework'},
+     {'score': 0.05243554711341858,
+      'sequence': '<s>HuggingFace is creating a library that the community uses to '
+                  'solve NLP tasks.</s>',
+      'token': 5560,
+      'token_str': 'Ġlibrary'},
+     {'score': 0.03493533283472061,
+      'sequence': '<s>HuggingFace is creating a database that the community uses '
+                  'to solve NLP tasks.</s>',
+      'token': 8503,
+      'token_str': 'Ġdatabase'},
+     {'score': 0.02860250137746334,
+      'sequence': '<s>HuggingFace is creating a prototype that the community uses '
+                  'to solve NLP tasks.</s>',
+      'token': 17715,
+      'token_str': 'Ġprototype'}]
 
 Here is an example doing masked language modeling using a model and a tokenizer. The process is the following:
 
@@ -345,51 +361,48 @@ Here is an example doing masked language modeling using a model and a tokenizer.
 - Retrieve the top 5 tokens using the PyTorch :obj:`topk` or TensorFlow :obj:`top_k` methods.
 - Replace the mask token by the tokens and print the results
 
-::
-
-    ## PYTORCH CODE
-    from transformers import AutoModelWithLMHead, AutoTokenizer
-    import torch
+.. code-block::
 
-    tokenizer = AutoTokenizer.from_pretrained("distilbert-base-cased")
-    model = AutoModelWithLMHead.from_pretrained("distilbert-base-cased")
+    >>> ## PYTORCH CODE
+    >>> from transformers import AutoModelWithLMHead, AutoTokenizer
+    >>> import torch
 
-    sequence = f"Distilled models are smaller than the models they mimic. Using them instead of the large versions would help {tokenizer.mask_token} our carbon footprint."
+    >>> tokenizer = AutoTokenizer.from_pretrained("distilbert-base-cased")
+    >>> model = AutoModelWithLMHead.from_pretrained("distilbert-base-cased")
 
-    input = tokenizer.encode(sequence, return_tensors="pt")
-    mask_token_index = torch.where(input == tokenizer.mask_token_id)[1]
+    >>> sequence = f"Distilled models are smaller than the models they mimic. Using them instead of the large versions would help {tokenizer.mask_token} our carbon footprint."
 
-    token_logits = model(input)[0]
-    mask_token_logits = token_logits[0, mask_token_index, :]
+    >>> input = tokenizer.encode(sequence, return_tensors="pt")
+    >>> mask_token_index = torch.where(input == tokenizer.mask_token_id)[1]
 
-    top_5_tokens = torch.topk(mask_token_logits, 5, dim=1).indices[0].tolist()
+    >>> token_logits = model(input)[0]
+    >>> mask_token_logits = token_logits[0, mask_token_index, :]
 
-    for token in top_5_tokens:
-        print(sequence.replace(tokenizer.mask_token, tokenizer.decode([token])))
-    ## TENSORFLOW CODE
-    from transformers import TFAutoModelWithLMHead, AutoTokenizer
-    import tensorflow as tf
+    >>> top_5_tokens = torch.topk(mask_token_logits, 5, dim=1).indices[0].tolist()
+    >>> ## TENSORFLOW CODE
+    >>> from transformers import TFAutoModelWithLMHead, AutoTokenizer
+    >>> import tensorflow as tf
 
-    tokenizer = AutoTokenizer.from_pretrained("distilbert-base-cased")
-    model = TFAutoModelWithLMHead.from_pretrained("distilbert-base-cased")
+    >>> tokenizer = AutoTokenizer.from_pretrained("distilbert-base-cased")
+    >>> model = TFAutoModelWithLMHead.from_pretrained("distilbert-base-cased")
 
-    sequence = f"Distilled models are smaller than the models they mimic. Using them instead of the large versions would help {tokenizer.mask_token} our carbon footprint."
+    >>> sequence = f"Distilled models are smaller than the models they mimic. Using them instead of the large versions would help {tokenizer.mask_token} our carbon footprint."
 
-    input = tokenizer.encode(sequence, return_tensors="tf")
-    mask_token_index = tf.where(input == tokenizer.mask_token_id)[0, 1]
+    >>> input = tokenizer.encode(sequence, return_tensors="tf")
+    >>> mask_token_index = tf.where(input == tokenizer.mask_token_id)[0, 1]
 
-    token_logits = model(input)[0]
-    mask_token_logits = token_logits[0, mask_token_index, :]
+    >>> token_logits = model(input)[0]
+    >>> mask_token_logits = token_logits[0, mask_token_index, :]
 
-    top_5_tokens = tf.math.top_k(mask_token_logits, 5).indices.numpy()
+    >>> top_5_tokens = tf.math.top_k(mask_token_logits, 5).indices.numpy()
 
-    for token in top_5_tokens:
-        print(sequence.replace(tokenizer.mask_token, tokenizer.decode([token])))
 
 This prints five sequences, with the top 5 tokens predicted by the model:
 
-::
+.. code-block::
 
+    >>> for token in top_5_tokens:
+    ...     print(sequence.replace(tokenizer.mask_token, tokenizer.decode([token])))
     Distilled models are smaller than the models they mimic. Using them instead of the large versions would help reduce our carbon footprint.
     Distilled models are smaller than the models they mimic. Using them instead of the large versions would help increase our carbon footprint.
     Distilled models are smaller than the models they mimic. Using them instead of the large versions would help decrease our carbon footprint.
@@ -408,65 +421,63 @@ Usually, the next token is predicted by sampling from the logits of the last hid
 
 Here is an example using the tokenizer and model and leveraging the :func:`~transformers.PreTrainedModel.top_k_top_p_filtering` method to sample the next token following an input sequence of tokens.
 
-::
-
-    ## PYTORCH CODE
-    from transformers import AutoModelWithLMHead, AutoTokenizer, top_k_top_p_filtering
-    import torch
-    from torch.nn import functional as F
+.. code-block::
 
+    >>> ## PYTORCH CODE
+    >>> from transformers import AutoModelWithLMHead, AutoTokenizer, top_k_top_p_filtering
+    >>> import torch
+    >>> from torch.nn import functional as F
 
-    tokenizer = AutoTokenizer.from_pretrained("gpt2")
-    model = AutoModelWithLMHead.from_pretrained("gpt2")
+    >>> tokenizer = AutoTokenizer.from_pretrained("gpt2")
+    >>> model = AutoModelWithLMHead.from_pretrained("gpt2")
 
-    sequence = f"Hugging Face is based in DUMBO, New York City, and "
+    >>> sequence = f"Hugging Face is based in DUMBO, New York City, and "
 
-    input_ids = tokenizer.encode(sequence, return_tensors="pt")
+    >>> input_ids = tokenizer.encode(sequence, return_tensors="pt")
 
-    # get logits of last hidden state
-    next_token_logits = model(input_ids)[0][:, -1, :]
+    >>> # get logits of last hidden state
+    >>> next_token_logits = model(input_ids)[0][:, -1, :]
 
-    # filter
-    filtered_next_token_logits = top_k_top_p_filtering(next_token_logits, top_k=50, top_p=1.0)
+    >>> # filter
+    >>> filtered_next_token_logits = top_k_top_p_filtering(next_token_logits, top_k=50, top_p=1.0)
 
-    # sample
-    probs = F.softmax(filtered_next_token_logits, dim=-1)
-    next_token = torch.multinomial(probs, num_samples=1)
+    >>> # sample
+    >>> probs = F.softmax(filtered_next_token_logits, dim=-1)
+    >>> next_token = torch.multinomial(probs, num_samples=1)
 
-    generated = torch.cat([input_ids, next_token], dim=-1)
+    >>> generated = torch.cat([input_ids, next_token], dim=-1)
 
-    resulting_string = tokenizer.decode(generated.tolist()[0])
-    print(resulting_string)
-    ## TENSORFLOW CODE
-    from transformers import TFAutoModelWithLMHead, AutoTokenizer, tf_top_k_top_p_filtering
-    import tensorflow as tf
+    >>> resulting_string = tokenizer.decode(generated.tolist()[0])
+    >>> ## TENSORFLOW CODE
+    >>> from transformers import TFAutoModelWithLMHead, AutoTokenizer, tf_top_k_top_p_filtering
+    >>> import tensorflow as tf
 
-    tokenizer = AutoTokenizer.from_pretrained("gpt2")
-    model = TFAutoModelWithLMHead.from_pretrained("gpt2")
+    >>> tokenizer = AutoTokenizer.from_pretrained("gpt2")
+    >>> model = TFAutoModelWithLMHead.from_pretrained("gpt2")
 
-    sequence = f"Hugging Face is based in DUMBO, New York City, and "
+    >>> sequence = f"Hugging Face is based in DUMBO, New York City, and "
 
-    input_ids = tokenizer.encode(sequence, return_tensors="tf")
+    >>> input_ids = tokenizer.encode(sequence, return_tensors="tf")
 
-    # get logits of last hidden state
-    next_token_logits = model(input_ids)[0][:, -1, :]
+    >>> # get logits of last hidden state
+    >>> next_token_logits = model(input_ids)[0][:, -1, :]
 
-    # filter
-    filtered_next_token_logits = tf_top_k_top_p_filtering(next_token_logits, top_k=50, top_p=1.0)
+    >>> # filter
+    >>> filtered_next_token_logits = tf_top_k_top_p_filtering(next_token_logits, top_k=50, top_p=1.0)
 
-    # sample
-    next_token = tf.random.categorical(filtered_next_token_logits, dtype=tf.int32, num_samples=1)
+    >>> # sample
+    >>> next_token = tf.random.categorical(filtered_next_token_logits, dtype=tf.int32, num_samples=1)
 
-    generated = tf.concat([input_ids, next_token], axis=1)
+    >>> generated = tf.concat([input_ids, next_token], axis=1)
 
-    resulting_string = tokenizer.decode(generated.numpy().tolist()[0])
-    print(resulting_string)
+    >>> resulting_string = tokenizer.decode(generated.numpy().tolist()[0])
 
 
 This outputs a (hopefully) coherent next token following the original sequence, which is in our case is the word *has*:
 
-::
+.. code-block::
 
+    print(resulting_string)
     Hugging Face is based in DUMBO, New York City, and has
 
 In the next section, we show how this functionality is leveraged in :func:`~transformers.PreTrainedModel.generate` to generate multiple tokens up to a user-defined length.
@@ -476,12 +487,14 @@ Text Generation
 
 In text generation (*a.k.a* *open-ended text generation*) the goal is to create a coherent portion of text that is a continuation from the given context. As an example, is it shown how *GPT-2* can be used in pipelines to generate text. As a default all models apply *Top-K* sampling when used in pipelines as configured in their respective configurations (see `gpt-2 config <https://s3.amazonaws.com/models.huggingface.co/bert/gpt2-config.json>`_ for example).
 
-::
+.. code-block::
+
+    >>> from transformers import pipeline
 
-    from transformers import pipeline
+    >>> text_generator = pipeline("text-generation")
+    >>> print(text_generator("As far as I am concerned, I will", max_length=50, do_sample=False))
+    [{'generated_text': 'As far as I am concerned, I will be the first to admit that I am not a fan of the idea of a "free market." I think that the idea of a free market is a bit of a stretch. I think that the idea'}]
 
-    text_generator = pipeline("text-generation")
-    print(text_generator("As far as I am concerned, I will", max_length=50))
 
 
 Here the model generates a random text with a total maximal length of *50* tokens from context *"As far as I am concerned, I will"*.
@@ -489,58 +502,59 @@ The default arguments of ``PreTrainedModel.generate()`` can directly be override
 
 Here is an example for text generation using XLNet and its tokenzier. 
 
-::
+.. code-block::
 
-    ## PYTORCH CODE
-    from transformers import AutoModelWithLMHead, AutoTokenizer
+    >>> ## PYTORCH CODE
+    >>> from transformers import AutoModelWithLMHead, AutoTokenizer
 
-    model = AutoModelWithLMHead.from_pretrained("xlnet-base-cased")
-    tokenizer = AutoTokenizer.from_pretrained("xlnet-base-cased")
+    >>> model = AutoModelWithLMHead.from_pretrained("xlnet-base-cased")
+    >>> tokenizer = AutoTokenizer.from_pretrained("xlnet-base-cased")
 
-    # Padding text helps XLNet with short prompts - proposed by Aman Rusia in https://github.com/rusiaaman/XLNet-gen#methodology
-    PADDING_TEXT = """In 1991, the remains of Russian Tsar Nicholas II and his family
-    (except for Alexei and Maria) are discovered.
-    The voice of Nicholas's young son, Tsarevich Alexei Nikolaevich, narrates the
-    remainder of the story. 1883 Western Siberia,
-    a young Grigori Rasputin is asked by his father and a group of men to perform magic.
-    Rasputin has a vision and denounces one of the men as a horse thief. Although his
-    father initially slaps him for making such an accusation, Rasputin watches as the
-    man is chased outside and beaten. Twenty years later, Rasputin sees a vision of
-    the Virgin Mary, prompting him to become a priest. Rasputin quickly becomes famous,
-    with people, even a bishop, begging for his blessing. <eod> </s> <eos>""" 
+    >>> # Padding text helps XLNet with short prompts - proposed by Aman Rusia in https://github.com/rusiaaman/XLNet-gen#methodology
+    >>> PADDING_TEXT = """In 1991, the remains of Russian Tsar Nicholas II and his family
+    ... (except for Alexei and Maria) are discovered.
+    ... The voice of Nicholas's young son, Tsarevich Alexei Nikolaevich, narrates the
+    ... remainder of the story. 1883 Western Siberia,
+    ... a young Grigori Rasputin is asked by his father and a group of men to perform magic.
+    ... Rasputin has a vision and denounces one of the men as a horse thief. Although his
+    ... father initially slaps him for making such an accusation, Rasputin watches as the
+    ... man is chased outside and beaten. Twenty years later, Rasputin sees a vision of
+    ... the Virgin Mary, prompting him to become a priest. Rasputin quickly becomes famous,
+    ... with people, even a bishop, begging for his blessing. <eod> </s> <eos>"""
 
-    prompt = "Today the weather is really nice and I am planning on "
-    inputs = tokenizer.encode(PADDING_TEXT + prompt, add_special_tokens=False, return_tensors="pt")
+    >>> prompt = "Today the weather is really nice and I am planning on "
+    >>> inputs = tokenizer.encode(PADDING_TEXT + prompt, add_special_tokens=False, return_tensors="pt")
 
-    prompt_length = len(tokenizer.decode(inputs[0], skip_special_tokens=True, clean_up_tokenization_spaces=True))
-    outputs = model.generate(inputs, max_length=250, do_sample=True, top_p=0.95, top_k=60)
-    generated = prompt + tokenizer.decode(outputs[0])[prompt_length:]
+    >>> prompt_length = len(tokenizer.decode(inputs[0], skip_special_tokens=True, clean_up_tokenization_spaces=True))
+    >>> outputs = model.generate(inputs, max_length=250, do_sample=True, top_p=0.95, top_k=60)
+    >>> generated = prompt + tokenizer.decode(outputs[0])[prompt_length:]
 
-    print(generated)
-    ## TENSORFLOW CODE
-    from transformers import TFAutoModelWithLMHead, AutoTokenizer
-
-    model = TFAutoModelWithLMHead.from_pretrained("xlnet-base-cased")
-    tokenizer = AutoTokenizer.from_pretrained("xlnet-base-cased")
-
-    # Padding text helps XLNet with short prompts - proposed by Aman Rusia in https://github.com/rusiaaman/XLNet-gen#methodology
-    PADDING_TEXT = """In 1991, the remains of Russian Tsar Nicholas II and his family
-    (except for Alexei and Maria) are discovered.
-    The voice of Nicholas's young son, Tsarevich Alexei Nikolaevich, narrates the
-    remainder of the story. 1883 Western Siberia,
-    a young Grigori Rasputin is asked by his father and a group of men to perform magic.
-    Rasputin has a vision and denounces one of the men as a horse thief. Although his
-    father initially slaps him for making such an accusation, Rasputin watches as the
-    man is chased outside and beaten. Twenty years later, Rasputin sees a vision of
-    the Virgin Mary, prompting him to become a priest. Rasputin quickly becomes famous,
-    with people, even a bishop, begging for his blessing. <eod> </s> <eos>""" 
-
-    prompt = "Today the weather is really nice and I am planning on "
-    inputs = tokenizer.encode(PADDING_TEXT + prompt, add_special_tokens=False, return_tensors="tf")
-
-    prompt_length = len(tokenizer.decode(inputs[0], skip_special_tokens=True, clean_up_tokenization_spaces=True))
-    outputs = model.generate(inputs, max_length=250, do_sample=True, top_p=0.95, top_k=60)
-    generated = prompt + tokenizer.decode(outputs[0])[prompt_length:]
+    >>> ## TENSORFLOW CODE
+    >>> from transformers import TFAutoModelWithLMHead, AutoTokenizer
+
+    >>> model = TFAutoModelWithLMHead.from_pretrained("xlnet-base-cased")
+    >>> tokenizer = AutoTokenizer.from_pretrained("xlnet-base-cased")
+
+    >>> # Padding text helps XLNet with short prompts - proposed by Aman Rusia in https://github.com/rusiaaman/XLNet-gen#methodology
+    >>> PADDING_TEXT = """In 1991, the remains of Russian Tsar Nicholas II and his family
+    ... (except for Alexei and Maria) are discovered.
+    ... The voice of Nicholas's young son, Tsarevich Alexei Nikolaevich, narrates the
+    ... remainder of the story. 1883 Western Siberia,
+    ... a young Grigori Rasputin is asked by his father and a group of men to perform magic.
+    ... Rasputin has a vision and denounces one of the men as a horse thief. Although his
+    ... father initially slaps him for making such an accusation, Rasputin watches as the
+    ... man is chased outside and beaten. Twenty years later, Rasputin sees a vision of
+    ... the Virgin Mary, prompting him to become a priest. Rasputin quickly becomes famous,
+    ... with people, even a bishop, begging for his blessing. <eod> </s> <eos>"""
+
+    >>> prompt = "Today the weather is really nice and I am planning on "
+    >>> inputs = tokenizer.encode(PADDING_TEXT + prompt, add_special_tokens=False, return_tensors="tf")
+
+    >>> prompt_length = len(tokenizer.decode(inputs[0], skip_special_tokens=True, clean_up_tokenization_spaces=True))
+    >>> outputs = model.generate(inputs, max_length=250, do_sample=True, top_p=0.95, top_k=60)
+    >>> generated = prompt + tokenizer.decode(outputs[0])[prompt_length:]
+
+.. code-block::
 
     print(generated)
 
@@ -575,21 +589,22 @@ of 9 classes:
 It leverages a fine-tuned model on CoNLL-2003, fine-tuned by `@stefan-it <https://github.com/stefan-it>`__ from
 `dbmdz <https://github.com/dbmdz>`__.
 
-::
+.. code-block::
 
-    from transformers import pipeline
+    >>> from transformers import pipeline
 
-    nlp = pipeline("ner")
+    >>> nlp = pipeline("ner")
 
-    sequence = "Hugging Face Inc. is a company based in New York City. Its headquarters are in DUMBO, therefore very" \
-               "close to the Manhattan Bridge which is visible from the window."
+    >>> sequence = "Hugging Face Inc. is a company based in New York City. Its headquarters are in DUMBO, therefore very"
+    ...            "close to the Manhattan Bridge which is visible from the window."
 
-    print(nlp(sequence))
 
 This outputs a list of all words that have been identified as an entity from the 9 classes defined above. Here is the
 expected results:
 
-::
+.. code-block::
+
+    print(nlp(sequence))
 
     [
         {'word': 'Hu', 'score': 0.9995632767677307, 'entity': 'I-ORG'},
@@ -623,75 +638,73 @@ Here is an example doing named entity recognition using a model and a tokenizer.
   for each token.
 - Zip together each token with its prediction and print it.
 
-::
-
-    ## PYTORCH CODE
-    from transformers import AutoModelForTokenClassification, AutoTokenizer
-    import torch
-
-    model = AutoModelForTokenClassification.from_pretrained("dbmdz/bert-large-cased-finetuned-conll03-english")
-    tokenizer = AutoTokenizer.from_pretrained("bert-base-cased")
-
-    label_list = [
-        "O",       # Outside of a named entity
-        "B-MISC",  # Beginning of a miscellaneous entity right after another miscellaneous entity
-        "I-MISC",  # Miscellaneous entity
-        "B-PER",   # Beginning of a person's name right after another person's name
-        "I-PER",   # Person's name
-        "B-ORG",   # Beginning of an organisation right after another organisation
-        "I-ORG",   # Organisation
-        "B-LOC",   # Beginning of a location right after another location
-        "I-LOC"    # Location
-    ]
-
-    sequence = "Hugging Face Inc. is a company based in New York City. Its headquarters are in DUMBO, therefore very" \
-               "close to the Manhattan Bridge."
-
-    # Bit of a hack to get the tokens with the special tokens
-    tokens = tokenizer.tokenize(tokenizer.decode(tokenizer.encode(sequence)))
-    inputs = tokenizer.encode(sequence, return_tensors="pt")
-
-    outputs = model(inputs)[0]
-    predictions = torch.argmax(outputs, dim=2)
-
-    print([(token, label_list[prediction]) for token, prediction in zip(tokens, predictions[0].tolist())])
-    ## TENSORFLOW CODE
-    from transformers import TFAutoModelForTokenClassification, AutoTokenizer
-    import tensorflow as tf
-
-    model = TFAutoModelForTokenClassification.from_pretrained("dbmdz/bert-large-cased-finetuned-conll03-english")
-    tokenizer = AutoTokenizer.from_pretrained("bert-base-cased")
-
-    label_list = [
-        "O",       # Outside of a named entity
-        "B-MISC",  # Beginning of a miscellaneous entity right after another miscellaneous entity
-        "I-MISC",  # Miscellaneous entity
-        "B-PER",   # Beginning of a person's name right after another person's name
-        "I-PER",   # Person's name
-        "B-ORG",   # Beginning of an organisation right after another organisation
-        "I-ORG",   # Organisation
-        "B-LOC",   # Beginning of a location right after another location
-        "I-LOC"    # Location
-    ]
-
-    sequence = "Hugging Face Inc. is a company based in New York City. Its headquarters are in DUMBO, therefore very" \
-               "close to the Manhattan Bridge."
+.. code-block::
+
+    >>> ## PYTORCH CODE
+    >>> from transformers import AutoModelForTokenClassification, AutoTokenizer
+    >>> import torch
+
+    >>> model = AutoModelForTokenClassification.from_pretrained("dbmdz/bert-large-cased-finetuned-conll03-english")
+    >>> tokenizer = AutoTokenizer.from_pretrained("bert-base-cased")
+
+    >>> label_list = [
+    ...     "O",       # Outside of a named entity
+    ...     "B-MISC",  # Beginning of a miscellaneous entity right after another miscellaneous entity
+    ...     "I-MISC",  # Miscellaneous entity
+    ...     "B-PER",   # Beginning of a person's name right after another person's name
+    ...     "I-PER",   # Person's name
+    ...     "B-ORG",   # Beginning of an organisation right after another organisation
+    ...     "I-ORG",   # Organisation
+    ...     "B-LOC",   # Beginning of a location right after another location
+    ...     "I-LOC"    # Location
+    ... ]
+
+    >>> sequence = "Hugging Face Inc. is a company based in New York City. Its headquarters are in DUMBO, therefore very" \
+    ...            "close to the Manhattan Bridge."
+
+    >>> # Bit of a hack to get the tokens with the special tokens
+    >>> tokens = tokenizer.tokenize(tokenizer.decode(tokenizer.encode(sequence)))
+    >>> inputs = tokenizer.encode(sequence, return_tensors="pt")
+
+    >>> outputs = model(inputs)[0]
+    >>> predictions = torch.argmax(outputs, dim=2)
+    >>> ## TENSORFLOW CODE
+    >>> from transformers import TFAutoModelForTokenClassification, AutoTokenizer
+    >>> import tensorflow as tf
+
+    >>> model = TFAutoModelForTokenClassification.from_pretrained("dbmdz/bert-large-cased-finetuned-conll03-english")
+    >>> tokenizer = AutoTokenizer.from_pretrained("bert-base-cased")
+
+    >>> label_list = [
+    ...     "O",       # Outside of a named entity
+    ...     "B-MISC",  # Beginning of a miscellaneous entity right after another miscellaneous entity
+    ...     "I-MISC",  # Miscellaneous entity
+    ...     "B-PER",   # Beginning of a person's name right after another person's name
+    ...     "I-PER",   # Person's name
+    ...     "B-ORG",   # Beginning of an organisation right after another organisation
+    ...     "I-ORG",   # Organisation
+    ...     "B-LOC",   # Beginning of a location right after another location
+    ...     "I-LOC"    # Location
+    ... ]
+
+    >>> sequence = "Hugging Face Inc. is a company based in New York City. Its headquarters are in DUMBO, therefore very" \
+    ...            "close to the Manhattan Bridge."
+
+    >>> # Bit of a hack to get the tokens with the special tokens
+    >>> tokens = tokenizer.tokenize(tokenizer.decode(tokenizer.encode(sequence)))
+    >>> inputs = tokenizer.encode(sequence, return_tensors="tf")
+
+    >>> outputs = model(inputs)[0]
+    >>> predictions = tf.argmax(outputs, axis=2)
 
-    # Bit of a hack to get the tokens with the special tokens
-    tokens = tokenizer.tokenize(tokenizer.decode(tokenizer.encode(sequence)))
-    inputs = tokenizer.encode(sequence, return_tensors="tf")
-
-    outputs = model(inputs)[0]
-    predictions = tf.argmax(outputs, axis=2)
-
-    print([(token, label_list[prediction]) for token, prediction in zip(tokens, predictions[0].numpy())])
 
 This outputs a list of each token mapped to their prediction. Differently from the pipeline, here every token has
 a prediction as we didn't remove the "0" class which means that no particular entity was found on that token. The
 following array should be the output:
 
-::
+.. code-block::
 
+    >>> print([(token, label_list[prediction]) for token, prediction in zip(tokens, predictions[0].numpy())])
     [('[CLS]', 'O'), ('Hu', 'I-ORG'), ('##gging', 'I-ORG'), ('Face', 'I-ORG'), ('Inc', 'I-ORG'), ('.', 'O'), ('is', 'O'), ('a', 'O'), ('company', 'O'), ('based', 'O'), ('in', 'O'), ('New', 'I-LOC'), ('York', 'I-LOC'), ('City', 'I-LOC'), ('.', 'O'), ('Its', 'O'), ('headquarters', 'O'), ('are', 'O'), ('in', 'O'), ('D', 'I-LOC'), ('##UM', 'I-LOC'), ('##BO', 'I-LOC'), (',', 'O'), ('therefore', 'O'), ('very', 'O'), ('##c', 'O'), ('##lose', 'O'), ('to', 'O'), ('the', 'O'), ('Manhattan', 'I-LOC'), ('Bridge', 'I-LOC'), ('.', 'O'), ('[SEP]', 'O')]
 
 Summarization
@@ -705,40 +718,39 @@ If you would like to fine-tune a model on a summarization task, you may leverage
 Here is an example using the pipelines do to summarization. 
 It leverages a Bart model that was fine-tuned on the CNN / Daily Mail data set.
 
-::
-
-    from transformers import pipeline
-
-    summarizer = pipeline("summarization")
-
-    ARTICLE = """ New York (CNN)When Liana Barrientos was 23 years old, she got married in Westchester County, New York. 
-    A year later, she got married again in Westchester County, but to a different man and without divorcing her first husband. 
-    Only 18 days after that marriage, she got hitched yet again. Then, Barrientos declared "I do" five more times, sometimes only within two weeks of each other. 
-    In 2010, she married once more, this time in the Bronx. In an application for a marriage license, she stated it was her "first and only" marriage. 
-    Barrientos, now 39, is facing two criminal counts of "offering a false instrument for filing in the first degree," referring to her false statements on the 
-    2010 marriage license application, according to court documents. 
-    Prosecutors said the marriages were part of an immigration scam. 
-    On Friday, she pleaded not guilty at State Supreme Court in the Bronx, according to her attorney, Christopher Wright, who declined to comment further. 
-    After leaving court, Barrientos was arrested and charged with theft of service and criminal trespass for allegedly sneaking into the New York subway through an emergency exit, said Detective 
-    Annette Markowski, a police spokeswoman. In total, Barrientos has been married 10 times, with nine of her marriages occurring between 1999 and 2002. 
-    All occurred either in Westchester County, Long Island, New Jersey or the Bronx. She is believed to still be married to four men, and at one time, she was married to eight men at once, prosecutors say. 
-    Prosecutors said the immigration scam involved some of her husbands, who filed for permanent residence status shortly after the marriages. 
-    Any divorces happened only after such filings were approved. It was unclear whether any of the men will be prosecuted. 
-    The case was referred to the Bronx District Attorney\'s Office by Immigration and Customs Enforcement and the Department of Homeland Security\'s 
-    Investigation Division. Seven of the men are from so-called "red-flagged" countries, including Egypt, Turkey, Georgia, Pakistan and Mali. 
-    Her eighth husband, Rashid Rajput, was deported in 2006 to his native Pakistan after an investigation by the Joint Terrorism Task Force. 
-    If convicted, Barrientos faces up to four years in prison.  Her next court appearance is scheduled for May 18.
-    """
-    
-    print(summarizer(ARTICLE, max_length=130, min_length=30))
+.. code-block::
+
+    >>> from transformers import pipeline
+
+    >>> summarizer = pipeline("summarization")
+
+    >>> ARTICLE = """ New York (CNN)When Liana Barrientos was 23 years old, she got married in Westchester County, New York.
+    ... A year later, she got married again in Westchester County, but to a different man and without divorcing her first husband.
+    ... Only 18 days after that marriage, she got hitched yet again. Then, Barrientos declared "I do" five more times, sometimes only within two weeks of each other.
+    ... In 2010, she married once more, this time in the Bronx. In an application for a marriage license, she stated it was her "first and only" marriage.
+    ... Barrientos, now 39, is facing two criminal counts of "offering a false instrument for filing in the first degree," referring to her false statements on the
+    ... 2010 marriage license application, according to court documents.
+    ... Prosecutors said the marriages were part of an immigration scam.
+    ... On Friday, she pleaded not guilty at State Supreme Court in the Bronx, according to her attorney, Christopher Wright, who declined to comment further.
+    ... After leaving court, Barrientos was arrested and charged with theft of service and criminal trespass for allegedly sneaking into the New York subway through an emergency exit, said Detective
+    ... Annette Markowski, a police spokeswoman. In total, Barrientos has been married 10 times, with nine of her marriages occurring between 1999 and 2002.
+    ... All occurred either in Westchester County, Long Island, New Jersey or the Bronx. She is believed to still be married to four men, and at one time, she was married to eight men at once, prosecutors say.
+    ... Prosecutors said the immigration scam involved some of her husbands, who filed for permanent residence status shortly after the marriages.
+    ... Any divorces happened only after such filings were approved. It was unclear whether any of the men will be prosecuted.
+    ... The case was referred to the Bronx District Attorney\'s Office by Immigration and Customs Enforcement and the Department of Homeland Security\'s
+    ... Investigation Division. Seven of the men are from so-called "red-flagged" countries, including Egypt, Turkey, Georgia, Pakistan and Mali.
+    ... Her eighth husband, Rashid Rajput, was deported in 2006 to his native Pakistan after an investigation by the Joint Terrorism Task Force.
+    ... If convicted, Barrientos faces up to four years in prison.  Her next court appearance is scheduled for May 18.
+    ... """
 
 Because the summarization pipeline depends on the ``PretrainedModel.generate()`` method, we can override the default arguments 
 of ``PretrainedModel.generate()`` directly in the pipeline as is shown for ``max_length`` and ``min_length`` above.
 This outputs the following summary:
 
-::
+.. code-block::
 
-  Liana Barrientos has been married 10 times, sometimes within two weeks of each other. Prosecutors say the marriages were part of an immigration scam. She pleaded not guilty at State Supreme Court in the Bronx on Friday.
+    >>> print(summarizer(ARTICLE, max_length=130, min_length=30, do_sample=False))
+    [{'summary_text': 'Liana Barrientos, 39, is charged with two counts of "offering a false instrument for filing in the first degree" In total, she has been married 10 times, with nine of her marriages occurring between 1999 and 2002. She is believed to still be married to four men.'}]
 
 Here is an example doing summarization using a model and a tokenizer. The process is the following:
 
@@ -748,29 +760,26 @@ Here is an example doing summarization using a model and a tokenizer. The proces
 - Add the T5 specific prefix "summarize: ".
 
 Here Google`s T5 model is used that was only pre-trained on a multi-task mixed data set (including CNN / Daily Mail), but nevertheless yields very good results.
-::
-
-    ## PYTORCH CODE
-    from transformers import AutoModelWithLMHead, AutoTokenizer
+.. code-block::
 
-    model = AutoModelWithLMHead.from_pretrained("t5-base")
-    tokenizer = AutoTokenizer.from_pretrained("t5-base")
+    >>> ## PYTORCH CODE
+    >>> from transformers import AutoModelWithLMHead, AutoTokenizer
 
-    # T5 uses a max_length of 512 so we cut the article to 512 tokens.
-    inputs = tokenizer.encode("summarize: " + ARTICLE, return_tensors="pt", max_length=512)
-    outputs = model.generate(inputs, max_length=150, min_length=40, length_penalty=2.0, num_beams=4, early_stopping=True)
-    print(outputs)
+    >>> model = AutoModelWithLMHead.from_pretrained("t5-base")
+    >>> tokenizer = AutoTokenizer.from_pretrained("t5-base")
 
-    ## TENSORFLOW CODE
-    from transformers import TFAutoModelWithLMHead, AutoTokenizer
+    >>> # T5 uses a max_length of 512 so we cut the article to 512 tokens.
+    >>> inputs = tokenizer.encode("summarize: " + ARTICLE, return_tensors="pt", max_length=512)
+    >>> outputs = model.generate(inputs, max_length=150, min_length=40, length_penalty=2.0, num_beams=4, early_stopping=True)
+    >>> ## TENSORFLOW CODE
+    >>> from transformers import TFAutoModelWithLMHead, AutoTokenizer
 
-    model = TFAutoModelWithLMHead.from_pretrained("t5-base")
-    tokenizer = AutoTokenizer.from_pretrained("t5-base")
+    >>> model = TFAutoModelWithLMHead.from_pretrained("t5-base")
+    >>> tokenizer = AutoTokenizer.from_pretrained("t5-base")
 
-    # T5 uses a max_length of 512 so we cut the article to 512 tokens.
-    inputs = tokenizer.encode("summarize: " + ARTICLE, return_tensors="tf", max_length=512)
-    outputs = model.generate(inputs, max_length=150, min_length=40, length_penalty=2.0, num_beams=4, early_stopping=True)
-    print(outputs)
+    >>> # T5 uses a max_length of 512 so we cut the article to 512 tokens.
+    >>> inputs = tokenizer.encode("summarize: " + ARTICLE, return_tensors="tf", max_length=512)
+    >>> outputs = model.generate(inputs, max_length=150, min_length=40, length_penalty=2.0, num_beams=4, early_stopping=True)
 
 Translation
 ----------------------------------------------------
@@ -784,12 +793,13 @@ Here is an example using the pipelines do to translation.
 It leverages a T5 model that was only pre-trained on a multi-task mixture dataset (including WMT), but yields impressive 
 translation results nevertheless.
 
-::
+.. code-block::
 
-    from transformers import pipeline
+    >>> from transformers import pipeline
 
-    translator = pipeline("translation_en_to_de")
-    print(translator("Hugging Face is a technology company based in New York and Paris", max_length=40))
+    >>> translator = pipeline("translation_en_to_de")
+    >>> print(translator("Hugging Face is a technology company based in New York and Paris", max_length=40))
+    [{'translation_text': 'Hugging Face ist ein Technologieunternehmen mit Sitz in New York und Paris.'}]
 
 Because the translation pipeline depends on the ``PretrainedModel.generate()`` method, we can override the default arguments 
 of ``PretrainedModel.generate()`` directly in the pipeline as is shown for ``max_length`` above.
@@ -806,26 +816,30 @@ Here is an example doing translation using a model and a tokenizer. The process
 - Leverage the ``PretrainedModel.generate()`` method.
 - Add the T5 specific prefix "translate English to German: "
 
-::
-
-    ## PYTORCH CODE
-    from transformers import AutoModelWithLMHead, AutoTokenizer
+.. code-block::
 
-    model = AutoModelWithLMHead.from_pretrained("t5-base")
-    tokenizer = AutoTokenizer.from_pretrained("t5-base")
+    >>> ## PYTORCH CODE
+    >>> from transformers import AutoModelWithLMHead, AutoTokenizer
 
-    inputs = tokenizer.encode("translate English to German: Hugging Face is a technology company based in New York and Paris", return_tensors="pt")
-    outputs = model.generate(inputs, max_length=40, num_beams=4, early_stopping=True)
+    >>> model = AutoModelWithLMHead.from_pretrained("t5-base")
+    >>> tokenizer = AutoTokenizer.from_pretrained("t5-base")
 
-    print(outputs)
+    >>> inputs = tokenizer.encode("translate English to German: Hugging Face is a technology company based in New York and Paris", return_tensors="pt")
+    >>> outputs = model.generate(inputs, max_length=40, num_beams=4, early_stopping=True)
 
-    ## TENSORFLOW CODE
-    from transformers import TFAutoModelWithLMHead, AutoTokenizer
+    >>> print(outputs)
+    tensor([[    0, 11560,  3896,  8881,   229,   236,     3, 14366, 15377,   181,
+             11216,    16,   368,  1060,    64,  1919,     5]])
+    >>> ## TENSORFLOW CODE
+    >>> from transformers import TFAutoModelWithLMHead, AutoTokenizer
 
-    model = TFAutoModelWithLMHead.from_pretrained("t5-base")
-    tokenizer = AutoTokenizer.from_pretrained("t5-base")
+    >>> model = TFAutoModelWithLMHead.from_pretrained("t5-base")
+    >>> tokenizer = AutoTokenizer.from_pretrained("t5-base")
 
-    inputs = tokenizer.encode("translate English to German: Hugging Face is a technology company based in New York and Paris", return_tensors="tf")
-    outputs = model.generate(inputs, max_length=40, num_beams=4, early_stopping=True)
+    >>> inputs = tokenizer.encode("translate English to German: Hugging Face is a technology company based in New York and Paris", return_tensors="tf")
+    >>> outputs = model.generate(inputs, max_length=40, num_beams=4, early_stopping=True)
 
-    print(outputs)
+    >>> print(outputs)
+    tf.Tensor(
+    [[    0 11560  3896  8881   229   236     3 14366 15377   181 11216    16
+        368  1060    64  1919     5]], shape=(1, 17), dtype=int32)

setup.py

@@ -86,7 +86,7 @@ extras["all"] = extras["serving"] + ["tensorflow", "torch"]
 
 extras["testing"] = ["pytest", "pytest-xdist", "timeout-decorator", "psutil"]
 # sphinx-rtd-theme==0.5.0 introduced big changes in the style.
-extras["docs"] = ["recommonmark", "sphinx", "sphinx-markdown-tables", "sphinx-rtd-theme==0.4.3"]
+extras["docs"] = ["recommonmark", "sphinx", "sphinx-markdown-tables", "sphinx-rtd-theme==0.4.3", "sphinx-copybutton"]
 extras["quality"] = [
     "black",
     "isort @ git+git://github.com/timothycrosley/isort.git@e63ae06ec7d70b06df9e528357650281a3d3ec22#egg=isort",

src/transformers/configuration_albert.py

@@ -81,22 +81,22 @@ class AlbertConfig(PretrainedConfig):
 
         Example::
 
-            from transformers import AlbertConfig, AlbertModel
-            # Initializing an ALBERT-xxlarge style configuration
-            albert_xxlarge_configuration = AlbertConfig()
-
-            # Initializing an ALBERT-base style configuration
-            albert_base_configuration = AlbertConfig(
-                hidden_size=768,
-                num_attention_heads=12,
-                intermediate_size=3072,
-            )
-
-            # Initializing a model from the ALBERT-base style configuration
-            model = AlbertModel(albert_xxlarge_configuration)
-
-            # Accessing the model configuration
-            configuration = model.config
+            >>> from transformers import AlbertConfig, AlbertModel
+            >>> # Initializing an ALBERT-xxlarge style configuration
+            >>> albert_xxlarge_configuration = AlbertConfig()
+
+            >>> # Initializing an ALBERT-base style configuration
+            >>> albert_base_configuration = AlbertConfig(
+            ...      hidden_size=768,
+            ...      num_attention_heads=12,
+            ...      intermediate_size=3072,
+            ...  )
+
+            >>> # Initializing a model from the ALBERT-base style configuration
+            >>> model = AlbertModel(albert_xxlarge_configuration)
+
+            >>> # Accessing the model configuration
+            >>> configuration = model.config
     """
 
     model_type = "albert"

src/transformers/configuration_bart.py

@@ -73,9 +73,13 @@ class BartConfig(PretrainedConfig):
     ):
         r"""
             :class:`~transformers.BartConfig` is the configuration class for `BartModel`.
-            Examples:
-                config = BartConfig.from_pretrained('bart-large')
-                model = BartModel(config)
+
+            Examples::
+
+                >>> from transformers import BartConfig, BartModel
+
+                >>> config = BartConfig.from_pretrained('facebook/bart-large')
+                >>> model = BartModel(config)
         """
         if "hidden_size" in common_kwargs:
             raise ValueError("hidden size is called d_model")

src/transformers/configuration_bert.py

@@ -95,16 +95,16 @@ class BertConfig(PretrainedConfig):
 
         Example::
 
-            from transformers import BertModel, BertConfig
+            >>> from transformers import BertModel, BertConfig
 
-            # Initializing a BERT bert-base-uncased style configuration
-            configuration = BertConfig()
+            >>> # Initializing a BERT bert-base-uncased style configuration
+            >>> configuration = BertConfig()
 
-            # Initializing a model from the bert-base-uncased style configuration
-            model = BertModel(configuration)
+            >>> # Initializing a model from the bert-base-uncased style configuration
+            >>> model = BertModel(configuration)
 
-            # Accessing the model configuration
-            configuration = model.config
+            >>> # Accessing the model configuration
+            >>> configuration = model.config
     """
     model_type = "bert"
 

src/transformers/configuration_ctrl.py

@@ -66,16 +66,16 @@ class CTRLConfig(PretrainedConfig):
 
         Example::
 
-            from transformers import CTRLModel, CTRLConfig
+            >>> from transformers import CTRLModel, CTRLConfig
 
-            # Initializing a CTRL configuration
-            configuration = CTRLConfig()
+            >>> # Initializing a CTRL configuration
+            >>> configuration = CTRLConfig()
 
-            # Initializing a model from the configuration
-            model = CTRLModel(configuration)
+            >>> # Initializing a model from the configuration
+            >>> model = CTRLModel(configuration)
 
-            # Accessing the model configuration
-            configuration = model.config
+            >>> # Accessing the model configuration
+            >>> configuration = model.config
     """
 
     model_type = "ctrl"

src/transformers/configuration_distilbert.py

@@ -80,16 +80,16 @@ class DistilBertConfig(PretrainedConfig):
 
         Example::
 
-            from transformers import DistilBertModel, DistilBertConfig
+            >>> from transformers import DistilBertModel, DistilBertConfig
 
-            # Initializing a DistilBERT configuration
-            configuration = DistilBertConfig()
+            >>> # Initializing a DistilBERT configuration
+            >>> configuration = DistilBertConfig()
 
-            # Initializing a model from the configuration
-            model = DistilBertModel(configuration)
+            >>> # Initializing a model from the configuration
+            >>> model = DistilBertModel(configuration)
 
-            # Accessing the model configuration
-            configuration = model.config
+            >>> # Accessing the model configuration
+            >>> configuration = model.config
     """
     model_type = "distilbert"
 

src/transformers/configuration_electra.py

@@ -101,16 +101,16 @@ class ElectraConfig(PretrainedConfig):
 
         Example::
 
-            from transformers import ElectraModel, ElectraConfig
+            >>> from transformers import ElectraModel, ElectraConfig
 
-            # Initializing a ELECTRA electra-base-uncased style configuration
-            configuration = ElectraConfig()
+            >>> # Initializing a ELECTRA electra-base-uncased style configuration
+            >>> configuration = ElectraConfig()
 
-            # Initializing a model from the electra-base-uncased style configuration
-            model = ElectraModel(configuration)
+            >>> # Initializing a model from the electra-base-uncased style configuration
+            >>> model = ElectraModel(configuration)
 
-            # Accessing the model configuration
-            configuration = model.config
+            >>> # Accessing the model configuration
+            >>> configuration = model.config
     """
     model_type = "electra"
 

src/transformers/configuration_encoder_decoder.py

@@ -42,20 +42,20 @@ class EncoderDecoderConfig(PretrainedConfig):
 
         Example::
 
-            from transformers import BertConfig, EncoderDecoderConfig, EncoderDecoderModel
+            >>> from transformers import BertConfig, EncoderDecoderConfig, EncoderDecoderModel
 
-            # Initializing a BERT bert-base-uncased style configuration
-            config_encoder = BertConfig()
-            config_decoder = BertConfig()
+            >>> # Initializing a BERT bert-base-uncased style configuration
+            >>> config_encoder = BertConfig()
+            >>> config_decoder = BertConfig()
 
-            config = EncoderDecoderConfig.from_encoder_decoder_configs(config_encoder, config_decoder)
+            >>> config = EncoderDecoderConfig.from_encoder_decoder_configs(config_encoder, config_decoder)
 
-            # Initializing a Bert2Bert model from the bert-base-uncased style configurations
-            model = EncoderDecoderModel(config=config)
+            >>> # Initializing a Bert2Bert model from the bert-base-uncased style configurations
+            >>> model = EncoderDecoderModel(config=config)
 
-            # Accessing the model configuration
-            config_encoder = model.config.encoder
-            config_decoder  = model.config.decoder
+            >>> # Accessing the model configuration
+            >>> config_encoder = model.config.encoder
+            >>> config_decoder  = model.config.decoder
     """
     model_type = "encoder_decoder"
 

src/transformers/configuration_gpt2.py

@@ -100,16 +100,16 @@ class GPT2Config(PretrainedConfig):
 
         Example::
 
-            from transformers import GPT2Model, GPT2Config
+            >>> from transformers import GPT2Model, GPT2Config
 
-            # Initializing a GPT2 configuration
-            configuration = GPT2Config()
+            >>> # Initializing a GPT2 configuration
+            >>> configuration = GPT2Config()
 
-            # Initializing a model from the configuration
-            model = GPT2Model(configuration)
+            >>> # Initializing a model from the configuration
+            >>> model = GPT2Model(configuration)
 
-            # Accessing the model configuration
-            configuration = model.config
+            >>> # Accessing the model configuration
+            >>> configuration = model.config
     """
 
     model_type = "gpt2"

src/transformers/configuration_longformer.py

@@ -49,16 +49,16 @@ class LongformerConfig(RobertaConfig):
 
         Example::
 
-            from transformers import LongformerConfig, LongformerModel
+            >>> from transformers import LongformerConfig, LongformerModel
 
-            # Initializing a Longformer configuration
-            configuration = LongformerConfig()
+            >>> # Initializing a Longformer configuration
+            >>> configuration = LongformerConfig()
 
-            # Initializing a model from the configuration
-            model = LongformerModel(configuration)
+            >>> # Initializing a model from the configuration
+            >>> model = LongformerModel(configuration)
 
-            # Accessing the model configuration
-            configuration = model.config
+            >>> # Accessing the model configuration
+            >>> configuration = model.config
     """
     model_type = "longformer"
 

src/transformers/configuration_mobilebert.py

@@ -85,16 +85,16 @@ class MobileBertConfig(PretrainedConfig):
 
         Example:
 
-            from transformers import MobileBertModel, MobileBertConfig
+            >>> from transformers import MobileBertModel, MobileBertConfig
 
-            # Initializing a MobileBERT configuration
-            configuration = MobileBertConfig()
+            >>> # Initializing a MobileBERT configuration
+            >>> configuration = MobileBertConfig()
 
-            # Initializing a model from the configuration above
-            model = MobileBertModel(configuration)
+            >>> # Initializing a model from the configuration above
+            >>> model = MobileBertModel(configuration)
 
-            # Accessing the model configuration
-            configuration = model.config
+            >>> # Accessing the model configuration
+            >>> configuration = model.config
 
         Attributes:
             pretrained_config_archive_map (Dict[str, str]):

src/transformers/configuration_openai.py

@@ -98,16 +98,16 @@ class OpenAIGPTConfig(PretrainedConfig):
 
         Example::
 
-            from transformers import OpenAIGPTConfig, OpenAIGPTModel
+            >>> from transformers import OpenAIGPTConfig, OpenAIGPTModel
 
-            # Initializing a GPT configuration
-            configuration = OpenAIGPTConfig()
+            >>> # Initializing a GPT configuration
+            >>> configuration = OpenAIGPTConfig()
 
-            # Initializing a model from the configuration
-            model = OpenAIGPTModel(configuration)
+            >>> # Initializing a model from the configuration
+            >>> model = OpenAIGPTModel(configuration)
 
-            # Accessing the model configuration
-            configuration = model.config
+            >>> # Accessing the model configuration
+            >>> configuration = model.config
     """
 
     model_type = "openai-gpt"