fix #808

pytorch_transformers/modeling_bert.py

@@ -600,6 +600,9 @@ BERT_INPUTS_DOCSTRING = r"""
                 
                 ``token_type_ids:   0   0   0   0  0     0   0``
 
+            Bert is a model with absolute position embeddings so it's usually advised to pad the inputs on
+            the right rather than the left.
+
             Indices can be obtained using :class:`pytorch_transformers.BertTokenizer`.
             See :func:`pytorch_transformers.PreTrainedTokenizer.encode` and
             :func:`pytorch_transformers.PreTrainedTokenizer.convert_tokens_to_ids` for details.

pytorch_transformers/modeling_gpt2.py

@@ -390,6 +390,8 @@ GPT2_START_DOCSTRING = r"""    OpenAI GPT-2 model was proposed in
 GPT2_INPUTS_DOCSTRING = r"""    Inputs:
         **input_ids**: ``torch.LongTensor`` of shape ``(batch_size, sequence_length)``:
             Indices of input sequence tokens in the vocabulary.
+            GPT-2 is a model with absolute position embeddings so it's usually advised to pad the inputs on
+            the right rather than the left.
             Indices can be obtained using :class:`pytorch_transformers.BPT2Tokenizer`.
             See :func:`pytorch_transformers.PreTrainedTokenizer.encode` and
             :func:`pytorch_transformers.PreTrainedTokenizer.convert_tokens_to_ids` for details.

pytorch_transformers/modeling_openai.py

@@ -404,6 +404,8 @@ OPENAI_GPT_START_DOCSTRING = r"""    OpenAI GPT model was proposed in
 OPENAI_GPT_INPUTS_DOCSTRING = r"""    Inputs:
         **input_ids**: ``torch.LongTensor`` of shape ``(batch_size, sequence_length)``:
             Indices of input sequence tokens in the vocabulary.
+            GPT is a model with absolute position embeddings so it's usually advised to pad the inputs on
+            the right rather than the left.
             Indices can be obtained using :class:`pytorch_transformers.BPT2Tokenizer`.
             See :func:`pytorch_transformers.PreTrainedTokenizer.encode` and
             :func:`pytorch_transformers.PreTrainedTokenizer.convert_tokens_to_ids` for details.

pytorch_transformers/modeling_roberta.py

@@ -110,6 +110,10 @@ ROBERTA_INPUTS_DOCSTRING = r"""
 
             Fully encoded sequences or sequence pairs can be obtained using the RobertaTokenizer.encode function with 
             the ``add_special_tokens`` parameter set to ``True``.
+
+            RoBERTa is a model with absolute position embeddings so it's usually advised to pad the inputs on
+            the right rather than the left.
+
             See :func:`pytorch_transformers.PreTrainedTokenizer.encode` and
             :func:`pytorch_transformers.PreTrainedTokenizer.convert_tokens_to_ids` for details.
         **position_ids**: (`optional`) ``torch.LongTensor`` of shape ``(batch_size, sequence_length)``:

pytorch_transformers/modeling_transfo_xl.py

@@ -936,6 +936,8 @@ TRANSFO_XL_INPUTS_DOCSTRING = r"""
     Inputs:
         **input_ids**: ``torch.LongTensor`` of shape ``(batch_size, sequence_length)``:
             Indices of input sequence tokens in the vocabulary.
+            Transformer-XL is a model with relative position embeddings so you can either pad the inputs on
+            the right or on the left.
             Indices can be obtained using :class:`pytorch_transformers.TransfoXLTokenizer`.
             See :func:`pytorch_transformers.PreTrainedTokenizer.encode` and
             :func:`pytorch_transformers.PreTrainedTokenizer.convert_tokens_to_ids` for details.

pytorch_transformers/modeling_xlm.py

@@ -424,6 +424,10 @@ XLM_INPUTS_DOCSTRING = r"""
     Inputs:
         **input_ids**: ``torch.LongTensor`` of shape ``(batch_size, sequence_length)``:
             Indices of input sequence tokens in the vocabulary.
+
+            XLM is a model with absolute position embeddings so it's usually advised to pad the inputs on
+            the right rather than the left.
+
             Indices can be obtained using :class:`pytorch_transformers.XLMTokenizer`.
             See :func:`pytorch_transformers.PreTrainedTokenizer.encode` and
             :func:`pytorch_transformers.PreTrainedTokenizer.convert_tokens_to_ids` for details.

pytorch_transformers/modeling_xlnet.py

@@ -655,6 +655,8 @@ XLNET_INPUTS_DOCSTRING = r"""
     Inputs:
         **input_ids**: ``torch.LongTensor`` of shape ``(batch_size, sequence_length)``:
             Indices of input sequence tokens in the vocabulary.
+            XLNet is a model with relative position embeddings so you can either pad the inputs on
+            the right or on the left.
             Indices can be obtained using :class:`pytorch_transformers.XLNetTokenizer`.
             See :func:`pytorch_transformers.PreTrainedTokenizer.encode` and
             :func:`pytorch_transformers.PreTrainedTokenizer.convert_tokens_to_ids` for details.