transformers/transformers/modeling_t5.py

# coding=utf-8
# Copyright 2018 Mesh TensorFlow authors, T5 Authors and HuggingFace Inc. team.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
""" PyTorch T5 model. """

from __future__ import absolute_import, division, print_function, unicode_literals

import json
import logging
import math
import os
import sys
import copy
import itertools
from io import open

import torch
from torch import nn
import torch.nn.functional as F
from torch.nn import CrossEntropyLoss, MSELoss

from .modeling_utils import PreTrainedModel, prune_linear_layer
from .configuration_t5 import T5Config
from .file_utils import add_start_docstrings, DUMMY_INPUTS, DUMMY_MASK

logger = logging.getLogger(__name__)

####################################################
# This dict contrains shortcut names and associated url
# for the pretrained weights provided with the models
####################################################
T5_PRETRAINED_MODEL_ARCHIVE_MAP = {
    't5-small': "https://s3.amazonaws.com/models.huggingface.co/bert/t5-small-pytorch_model.bin",
    't5-base': "https://s3.amazonaws.com/models.huggingface.co/bert/t5-base-pytorch_model.bin",
    't5-large': "https://s3.amazonaws.com/models.huggingface.co/bert/t5-large-pytorch_model.bin",
    't5-3b': "https://s3.amazonaws.com/models.huggingface.co/bert/t5-3b-pytorch_model.bin",
    't5-11b': "https://s3.amazonaws.com/models.huggingface.co/bert/t5-11b-pytorch_model.bin",
}

####################################################
# This is a conversion method from TF 1.0 to PyTorch
# More details: https://medium.com/huggingface/from-tensorflow-to-pytorch-265f40ef2a28
####################################################
def load_tf_weights_in_t5(model, config, tf_checkpoint_path):
    """ Load tf checkpoints in a pytorch model.
    """
    try:
        import re
        import numpy as np
        import tensorflow as tf
    except ImportError:
        logger.error("Loading a TensorFlow model in PyTorch, requires TensorFlow to be installed. Please see "
            "https://www.tensorflow.org/install/ for installation instructions.")
        raise
    tf_path = os.path.abspath(tf_checkpoint_path)
    logger.info("Converting TensorFlow checkpoint from {}".format(tf_path))
    # Load weights from TF model
    init_vars = tf.train.list_variables(tf_path)
    names = []
    tf_weights = {}
    for name, shape in init_vars:
        logger.info("Loading TF weight {} with shape {}".format(name, shape))
        array = tf.train.load_variable(tf_path, name)
        names.append(name)
        tf_weights[name] = array

    for txt_name in names:
        name = txt_name.split('/')
        # adam_v and adam_m are variables used in AdamWeightDecayOptimizer to calculated m and v
        # which are not required for using pretrained model
        if any(n in ["adam_v", "adam_m", "global_step"] for n in name):
            logger.info("Skipping {}".format("/".join(name)))
            tf_weights.pop(txt_name, None)
            continue
        if '_slot_' in name[-1]:
            logger.info("Skipping {}".format("/".join(name)))
            tf_weights.pop(txt_name, None)
            continue
        pointer = model
        array = tf_weights[txt_name]
        for m_name in name:
            if re.fullmatch(r'[A-Za-z]+_\d+', m_name):
                l = re.split(r'_(\d+)', m_name)
            else:
                l = [m_name]
            if l[0] in ['kernel', 'scale', 'embedding']:
                pointer = getattr(pointer, 'weight')
            # elif l[0] == 'scale':
            #     pointer = getattr(pointer, 'weight')
            # elif l[0] == 'output_bias' or l[0] == 'beta':
            #     pointer = getattr(pointer, 'bias')
            # elif l[0] == 'squad':
            #     pointer = getattr(pointer, 'classifier')
            else:
                try:
                    pointer = getattr(pointer, l[0])
                except AttributeError:
                    logger.info("Skipping {}".format("/".join(name)))
                    continue
            if len(l) >= 2:
                num = int(l[1])
                pointer = pointer[num]
        if l[0] not in ['kernel', 'scale', 'embedding']:
            pointer = getattr(pointer, 'weight')
        if l[0] != 'embedding':
            logger.info("Transposing numpy weight of shape {} for {}".format(array.shape, name))
            array = np.transpose(array)
        try:
            assert pointer.shape == array.shape
        except AssertionError as e:
            e.args += (pointer.shape, array.shape)
            raise
        logger.info("Initialize PyTorch weight {}".format(name))
        pointer.data = torch.from_numpy(array.astype(np.float32))
        tf_weights.pop(txt_name, None)

    logger.info("Weights not copied to PyTorch model: {}".format(', '.join(tf_weights.keys())))
    # logger.info("Weights not copied to PyTorch model: {}".format(', '.join(tf_weights.keys())))
    return model


####################################################
# PyTorch Models are constructed by sub-classing
# - torch.nn.Module for the layers and
# - PreTrainedModel for the models (it-self a sub-class of torch.nn.Module)
####################################################

class T5LayerNorm(nn.Module):
    def __init__(self, hidden_size, eps=1e-6):
        """ Construct a layernorm module in the T5 style
            No bias and no substraction of mean.
        """
        super(T5LayerNorm, self).__init__()
        self.weight = nn.Parameter(torch.ones(hidden_size))
        self.variance_epsilon = eps

    def forward(self, x):
        variance = x.pow(2).mean(-1, keepdim=True)
        x = x / torch.sqrt(variance + self.variance_epsilon)
        return self.weight * x


class T5DenseReluDense(nn.Module):
    def __init__(self, config):
        super(T5DenseReluDense, self).__init__()
        self.wi = nn.Linear(config.d_model, config.d_ff, bias=False)
        self.wo = nn.Linear(config.d_ff, config.d_model, bias=False)
        self.dropout = nn.Dropout(config.dropout_rate)

    def forward(self, hidden_states):
        h = self.wi(hidden_states)
        h = F.relu(h)
        h = self.dropout(h)
        h = self.wo(h)
        return h


class T5LayerFF(nn.Module):
    def __init__(self, config):
        super(T5LayerFF, self).__init__()
        self.DenseReluDense = T5DenseReluDense(config)
        self.layer_norm = T5LayerNorm(config.d_model, eps=config.layer_norm_epsilon)
        self.dropout = nn.Dropout(config.dropout_rate)

    def forward(self, hidden_states):
        norm_x = self.layer_norm(hidden_states)
        y = self.DenseReluDense(norm_x)
        layer_output = hidden_states + self.dropout(y)
        return layer_output


class T5Attention(nn.Module):
    NEW_ID = itertools.count()

    def __init__(self, config, has_relative_attention_bias=False):
        super(T5Attention, self).__init__()
        self.layer_id = next(T5Attention.NEW_ID)
        self.is_decoder = config.is_decoder
        self.has_relative_attention_bias = has_relative_attention_bias

        self.output_attentions = config.output_attentions
        self.relative_attention_num_buckets = config.relative_attention_num_buckets
        self.d_model = config.d_model
        self.d_kv = config.d_kv
        self.n_heads = config.num_heads
        self.dropout = config.dropout_rate
        self.inner_dim = self.n_heads * self.d_kv

        # Mesh TensorFlow initialization to avoid scaling before softmax
        self.q = nn.Linear(self.d_model, self.inner_dim, bias=False)
        self.k = nn.Linear(self.d_model, self.inner_dim, bias=False)
        self.v = nn.Linear(self.d_model, self.inner_dim, bias=False)
        self.o = nn.Linear(self.inner_dim, self.d_model, bias=False)

        if self.has_relative_attention_bias:
            self.relative_attention_bias = nn.Embedding(self.relative_attention_num_buckets, self.n_heads)
        self.pruned_heads = set()

    def prune_heads(self, heads):
        if len(heads) == 0:
            return
        mask = torch.ones(self.n_heads, self.d_kv)
        heads = set(heads) - self.pruned_heads
        for head in heads:
            head -= sum(1 if h < head else 0 for h in self.pruned_heads)
            mask[head] = 0
        mask = mask.view(-1).contiguous().eq(1)
        index = torch.arange(len(mask))[mask].long()
        # Prune linear layers
        self.q = prune_linear_layer(self.q, index)
        self.k = prune_linear_layer(self.k, index)
        self.v = prune_linear_layer(self.v, index)
        self.o = prune_linear_layer(self.o, index, dim=1)
        # Update hyper params
        self.n_heads = self.n_heads - len(heads)
        self.inner_dim = self.d_kv * self.n_heads
        self.pruned_heads = self.pruned_heads.union(heads)

    @staticmethod
    def _relative_position_bucket(relative_position,
                                  bidirectional=True,
                                  num_buckets=32,
                                  max_distance=128):
        """
        Adapted from Mesh Tensorflow:
        https://github.com/tensorflow/mesh/blob/0cb87fe07da627bf0b7e60475d59f95ed6b5be3d/mesh_tensorflow/transformer/transformer_layers.py#L593

        Translate relative position to a bucket number for relative attention.
        The relative position is defined as memory_position - query_position, i.e.
        the distance in tokens from the attending position to the attended-to
        position.  If bidirectional=False, then positive relative positions are
        invalid.
        We use smaller buckets for small absolute relative_position and larger buckets
        for larger absolute relative_positions.  All relative positions >=max_distance
        map to the same bucket.  All relative positions <=-max_distance map to the
        same bucket.  This should allow for more graceful generalization to longer
        sequences than the model has been trained on.
        Args:
            relative_position: an int32 Tensor
            bidirectional: a boolean - whether the attention is bidirectional
            num_buckets: an integer
            max_distance: an integer
        Returns:
            a Tensor with the same shape as relative_position, containing int32
            values in the range [0, num_buckets)
        """
        ret = 0
        n = -relative_position
        if bidirectional:
            num_buckets //= 2
            ret += (n < 0).to(torch.long) * num_buckets  # mtf.to_int32(mtf.less(n, 0)) * num_buckets
            n = torch.abs(n)
        else:
            n = torch.max(n, torch.zeros_like(n))
        # now n is in the range [0, inf)

        # half of the buckets are for exact increments in positions
        max_exact = num_buckets // 2
        is_small = (n < max_exact)

        # The other half of the buckets are for logarithmically bigger bins in positions up to max_distance
        val_if_large = max_exact + (
            torch.log(n.float() / max_exact)
            / math.log(max_distance / max_exact) * (num_buckets - max_exact)).to(torch.long)
        val_if_large = torch.min(val_if_large, torch.full_like(val_if_large, num_buckets - 1))

        ret += torch.where(is_small, n, val_if_large)
        return ret

    def compute_bias(self, qlen, klen):
        """ Compute binned relative position bias """
        context_position = torch.arange(qlen, dtype=torch.long)[:, None]
        memory_position = torch.arange(klen, dtype=torch.long)[None, :]
        relative_position = memory_position - context_position  # shape (qlen, klen)
        rp_bucket = self._relative_position_bucket(relative_position,  # shape (qlen, klen)
                                                   bidirectional=not self.is_decoder,
                                                   num_buckets=self.relative_attention_num_buckets)
        values = self.relative_attention_bias(rp_bucket)  # shape (qlen, klen, num_heads)
        values = values.permute([2, 0, 1]).unsqueeze(0) # shape (1, num_heads, qlen, klen)
        return values

    def forward(self, input, mask=None, kv=None, position_bias=None, cache=None, head_mask=None):
        """
        Self-attention (if kv is None) or attention over source sentence (provided by kv).
        """
        # Input is (bs, qlen, dim)
        # Mask is (bs, klen) (non-causal) or (bs, klen, klen)
        bs, qlen, dim = input.size()
        if kv is None:
            klen = qlen if cache is None else cache['slen'] + qlen
        else:
            klen = kv.size(1)

        def shape(x):
            """  projection """
            return x.view(bs, -1, self.n_heads, self.d_kv).transpose(1, 2)

        def unshape(x):
            """  compute context """
            return x.transpose(1, 2).contiguous().view(bs, -1, self.inner_dim)

        q = shape(self.q(input))                                          # (bs, n_heads, qlen, dim_per_head)
        if kv is None:
            k = shape(self.k(input))                                      # (bs, n_heads, qlen, dim_per_head)
            v = shape(self.v(input))                                      # (bs, n_heads, qlen, dim_per_head)
        elif cache is None or self.layer_id not in cache:
            k = v = kv
            k = shape(self.k(k))                                          # (bs, n_heads, qlen, dim_per_head)
            v = shape(self.v(v))                                          # (bs, n_heads, qlen, dim_per_head)

        if cache is not None:
            if self.layer_id in cache:
                if kv is None:
                    k_, v_ = cache[self.layer_id]
                    k = torch.cat([k_, k], dim=2)                             # (bs, n_heads, klen, dim_per_head)
                    v = torch.cat([v_, v], dim=2)                             # (bs, n_heads, klen, dim_per_head)
                else:
                    k, v = cache[self.layer_id]
            cache[self.layer_id] = (k, v)

        # q = q / math.sqrt(dim_per_head)                                     # No scaling in T5
        scores = torch.einsum('bnqd,bnkd->bnqk', q, k)                        # (bs, n_heads, qlen, klen)

        if position_bias is None:
            if not self.has_relative_attention_bias:
                raise ValueError("No position_bias provided and no weights to compute position_bias")
            position_bias = self.compute_bias(qlen, klen)
            if mask is not None:
                position_bias = position_bias + mask                          # (bs, n_heads, qlen, klen)

        scores += position_bias
        weights = F.softmax(scores.float(), dim=-1).type_as(scores)           # (bs, n_heads, qlen, klen)
        weights = F.dropout(weights, p=self.dropout, training=self.training)  # (bs, n_heads, qlen, klen)

        # Mask heads if we want to
        if head_mask is not None:
            weights = weights * head_mask

        context = torch.matmul(weights, v)                                    # (bs, n_heads, qlen, dim_per_head)
        context = unshape(context)                                            # (bs, qlen, dim)

        context = self.o(context)

        outputs = (context,)
        if self.output_attentions:
            outputs = outputs + (weights,)
        if self.has_relative_attention_bias:
            outputs = outputs + (position_bias,)
        return outputs


class T5LayerSelfAttention(nn.Module):
    def __init__(self, config, has_relative_attention_bias=False):
        super(T5LayerSelfAttention, self).__init__()
        self.SelfAttention = T5Attention(config, has_relative_attention_bias=has_relative_attention_bias)
        self.layer_norm = T5LayerNorm(config.d_model, eps=config.layer_norm_epsilon)
        self.dropout = nn.Dropout(config.dropout_rate)

    def forward(self, hidden_states, attention_mask=None, position_bias=None, head_mask=None):
        norm_x = self.layer_norm(hidden_states)
        attention_output = self.SelfAttention(norm_x,
                                              mask=attention_mask,
                                              position_bias=position_bias,
                                              head_mask=head_mask)
        y = attention_output[0]
        layer_output = hidden_states + self.dropout(y)
        outputs = (layer_output,) + attention_output[1:]  # add attentions if we output them
        return outputs


class T5LayerCrossAttention(nn.Module):
    def __init__(self, config, has_relative_attention_bias=False):
        super(T5LayerCrossAttention, self).__init__()
        self.EncDecAttention = T5Attention(config, has_relative_attention_bias=has_relative_attention_bias)
        self.layer_norm = T5LayerNorm(config.d_model, eps=config.layer_norm_epsilon)
        self.dropout = nn.Dropout(config.dropout_rate)

    def forward(self, hidden_states, kv, attention_mask=None, position_bias=None, head_mask=None):
        norm_x = self.layer_norm(hidden_states)
        attention_output = self.EncDecAttention(norm_x,
                                                mask=attention_mask,
                                                kv=kv,
                                                position_bias=position_bias,
                                                head_mask=head_mask)
        y = attention_output[0]
        layer_output = hidden_states + self.dropout(y)
        outputs = (layer_output,) + attention_output[1:]  # add attentions if we output them
        return outputs


class T5Block(nn.Module):
    def __init__(self, config, has_relative_attention_bias=False):
        super(T5Block, self).__init__()
        self.is_decoder = config.is_decoder
        self.layer = nn.ModuleList()
        self.layer.append(T5LayerSelfAttention(config, has_relative_attention_bias=has_relative_attention_bias))
        if self.is_decoder:
            self.layer.append(T5LayerCrossAttention(config, has_relative_attention_bias=has_relative_attention_bias))
            self.layer.append(T5LayerFF(config))
        else:
            self.layer.append(T5LayerFF(config))

    def forward(self, hidden_states, attention_mask=None, position_bias=None,
                encoder_hidden_states=None, encoder_attention_mask=None, encoder_decoder_position_bias=None,
                head_mask=None):
        self_attention_outputs = self.layer[0](hidden_states,
                                                attention_mask=attention_mask,
                                                position_bias=position_bias,
                                                head_mask=head_mask)
        hidden_states = self_attention_outputs[0]
        outputs = self_attention_outputs[1:]  # Keep self-attention outputs and relative position weights

        if not self.is_decoder:
            hidden_states = self.layer[1](hidden_states)
        else:
            cross_attention_outputs = self.layer[1](hidden_states,
                                                    kv=encoder_hidden_states,
                                                    attention_mask=encoder_attention_mask,
                                                    position_bias=encoder_decoder_position_bias,
                                                    head_mask=head_mask)
            hidden_states = cross_attention_outputs[0]
            outputs = outputs + cross_attention_outputs[1:]  # Keep cross-attention outputs and relative position weights
            hidden_states = self.layer[2](hidden_states)

        outputs = (hidden_states,) + outputs  # add attentions if we output them
        return outputs  # hidden-states, (self-attention weights), (self-attention position bias), (cross-attention weights), (cross-attention position bias)


class T5PreTrainedModel(PreTrainedModel):
    """ An abstract class to handle weights initialization and
        a simple interface for dowloading and loading pretrained models.
    """
    config_class = T5Config
    pretrained_model_archive_map = T5_PRETRAINED_MODEL_ARCHIVE_MAP
    load_tf_weights = load_tf_weights_in_t5
    base_model_prefix = "transformer"

    @property
    def dummy_inputs(self):
        input_ids = torch.tensor(DUMMY_INPUTS)
        input_mask = torch.tensor(DUMMY_MASK)
        dummy_inputs = {'decoder_input_ids': input_ids,
                        'encoder_input_ids': input_ids,
                        'decoder_attention_mask': input_mask}
        return dummy_inputs

    def _init_weights(self, module):
        """ Initialize the weights """
        factor = self.config.initializer_factor  # Used for testing weights initialization
        if isinstance(module, T5LayerNorm):
            module.weight.data.fill_(factor*1.0)
        elif isinstance(module, (T5Model, T5WithLMHeadModel)):
            # Mesh TensorFlow embeddings initialization
            # See https://github.com/tensorflow/mesh/blob/fa19d69eafc9a482aff0b59ddd96b025c0cb207d/mesh_tensorflow/layers.py#L1624
            module.shared.weight.data.normal_(mean=0.0, std=factor*1.0)
        elif isinstance(module, T5DenseReluDense):
            # Mesh TensorFlow FF initialization
            # See https://github.com/tensorflow/mesh/blob/master/mesh_tensorflow/transformer/transformer_layers.py#L56
            # and https://github.com/tensorflow/mesh/blob/fa19d69eafc9a482aff0b59ddd96b025c0cb207d/mesh_tensorflow/layers.py#L89
            module.wi.weight.data.normal_(mean=0.0, std=factor*((self.config.d_model) ** -0.5))
            if hasattr(module.wi, 'bias') and module.wi.bias is not None:
                module.wi.bias.data.zero_()
            module.wo.weight.data.normal_(mean=0.0, std=factor*((self.config.d_ff) ** -0.5))
            if hasattr(module.wo, 'bias') and module.wo.bias is not None:
                module.wo.bias.data.zero_()
        elif isinstance(module, T5Attention):
            # Mesh TensorFlow attention initialization to avoid scaling before softmax
            # See https://github.com/tensorflow/mesh/blob/fa19d69eafc9a482aff0b59ddd96b025c0cb207d/mesh_tensorflow/transformer/attention.py#L136
            d_model = self.config.d_model
            d_kv = self.config.d_kv
            n_heads = self.config.num_heads
            module.q.weight.data.normal_(mean=0.0, std=factor*((d_model * d_kv) ** -0.5))
            module.k.weight.data.normal_(mean=0.0, std=factor*(d_model ** -0.5))
            module.v.weight.data.normal_(mean=0.0, std=factor*(d_model ** -0.5))
            module.o.weight.data.normal_(mean=0.0, std=factor*((n_heads * d_kv) ** -0.5))
            if module.has_relative_attention_bias:
                module.relative_attention_bias.weight.data.normal_(mean=0.0, std=factor*((d_model) ** -0.5))


class T5Stack(T5PreTrainedModel):
    def __init__(self, config):
        super(T5Stack, self).__init__(config)
        self.output_attentions = config.output_attentions
        self.output_hidden_states = config.output_hidden_states
        self.is_decoder = config.is_decoder

        self.block = nn.ModuleList([T5Block(config, has_relative_attention_bias=bool(i == 0))
                                    for i in range(config.num_layers)])
        self.final_layer_norm = T5LayerNorm(config.d_model, eps=config.layer_norm_epsilon)
        self.dropout = nn.Dropout(config.dropout_rate)

        self.init_weights()

    def forward(self,
                hidden_states,
                attention_mask=None,
                encoder_hidden_states=None,
                encoder_attention_mask=None,
                head_mask=None):

        batch_size, seq_length = hidden_states.shape[0], hidden_states.shape[1]
        if attention_mask is None:
            attention_mask = torch.ones(batch_size, seq_length).to(hidden_states.device)
        if self.is_decoder and encoder_attention_mask is None:
            encoder_seq_length = encoder_hidden_states.shape[1]
            encoder_attention_mask = torch.ones(batch_size, encoder_seq_length).to(hidden_states.device)

        # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
        # ourselves in which case we just need to make it broadcastable to all heads.
        if attention_mask.dim() == 3:
            extended_attention_mask = attention_mask[:, None, :, :]
        elif attention_mask.dim() == 2:
        # Provided a padding mask of dimensions [batch_size, seq_length]
        # - if the model is a decoder, apply a causal mask in addition to the padding mask
        # - if the model is an encoder, make the mask broadcastable to [batch_size, num_heads, seq_length, seq_length]
            if self.config.is_decoder:
                seq_ids = torch.arange(seq_length, device=hidden_states.device)
                causal_mask = seq_ids[None, None, :].repeat(batch_size, seq_length, 1) <= seq_ids[None, :, None]
                causal_mask = causal_mask.to(attention_mask)
                extended_attention_mask = causal_mask[:, None, :, :] * attention_mask[:, None, None, :]
            else:
                extended_attention_mask = attention_mask[:, None, None, :]

        # Since attention_mask is 1.0 for positions we want to attend and 0.0 for
        # masked positions, this operation will create a tensor which is 0.0 for
        # positions we want to attend and -1e9 for masked positions.
        # Since we are adding it to the raw scores before the softmax, this is
        # effectively the same as removing these entirely.

        # T5 has a mask that can compare sequence ids, we can simulate this here with this transposition
        # Cf. https://github.com/tensorflow/mesh/blob/8d2465e9bc93129b913b5ccc6a59aa97abd96ec6/mesh_tensorflow/transformer/transformer_layers.py#L270
        # extended_attention_mask = (extended_attention_mask == extended_attention_mask.transpose(-1, -2))

        extended_attention_mask = extended_attention_mask.to(dtype=next(self.parameters()).dtype)  # fp16 compatibility
        extended_attention_mask = (1.0 - extended_attention_mask) * -1e9

        if self.is_decoder:
            # If a 2D ou 3D attention mask is provided for the cross-attention
            # we need to make broadcastabe to [batch_size, num_heads, seq_length, seq_length]
            if encoder_attention_mask.dim() == 3:
                encoder_extended_attention_mask = encoder_attention_mask[:, None, :, :]
            if encoder_attention_mask.dim() == 2:
                encoder_extended_attention_mask = encoder_attention_mask[:, None, None, :]

            # T5 has a mask that can compare sequence ids, we can simulate this here with this transposition
            # Cf. https://github.com/tensorflow/mesh/blob/8d2465e9bc93129b913b5ccc6a59aa97abd96ec6/mesh_tensorflow/transformer/transformer_layers.py#L270
            # encoder_extended_attention_mask = (encoder_extended_attention_mask == encoder_extended_attention_mask.transpose(-1, -2))

            encoder_extended_attention_mask = encoder_extended_attention_mask.to(dtype=next(self.parameters()).dtype)  # fp16 compatibility
            encoder_extended_attention_mask = (1.0 - encoder_extended_attention_mask) * -1e9
        else:
            encoder_extended_attention_mask = None

        # Prepare head mask if needed
        # 1.0 in head_mask indicate we keep the head
        # attention_probs has shape bsz x n_heads x N x N
        # input head_mask has shape [num_heads] or [num_hidden_layers x num_heads]
        # and head_mask is converted to shape [num_hidden_layers x batch x num_heads x seq_length x seq_length]
        if head_mask is not None:
            if head_mask.dim() == 1:
                head_mask = head_mask.unsqueeze(0).unsqueeze(0).unsqueeze(-1).unsqueeze(-1)
                head_mask = head_mask.expand(self.config.num_layers, -1, -1, -1, -1)
            elif head_mask.dim() == 2:
                head_mask = head_mask.unsqueeze(1).unsqueeze(-1).unsqueeze(-1)  # We can specify head_mask for each layer
            head_mask = head_mask.to(dtype=next(self.parameters()).dtype) # switch to fload if need + fp16 compatibility
        else:
            head_mask = [None] * self.config.num_layers

        all_hidden_states = ()
        all_attentions = ()
        position_bias = None
        encoder_decoder_position_bias = None

        hidden_states = self.dropout(hidden_states)
        for i, layer_module in enumerate(self.block):
            if self.output_hidden_states:
                all_hidden_states = all_hidden_states + (hidden_states,)

            layer_outputs = layer_module(hidden_states,
                                         attention_mask=extended_attention_mask,
                                         position_bias=position_bias,
                                         encoder_hidden_states=encoder_hidden_states,
                                         encoder_attention_mask=encoder_extended_attention_mask,
                                         encoder_decoder_position_bias=encoder_decoder_position_bias,
                                         head_mask=head_mask[i])
            # layer_outputs is a tuple with:
            # hidden-states, (self-attention weights), (self-attention position bias), (cross-attention weights), (cross-attention position bias)
            hidden_states = layer_outputs[0]
            if i == 0:
                # We share the position biases between the layers - the first layer store them
                # layer_outputs = hidden-states, (self-attention weights), (self-attention position bias), (cross-attention weights), (cross-attention position bias)
                position_bias = layer_outputs[2 if self.output_attentions else 1]
                if self.is_decoder:
                    encoder_decoder_position_bias = layer_outputs[4 if self.output_attentions else 2]

            if self.output_attentions:
                all_attentions = all_attentions + (layer_outputs[1],)  # We keep only self-attention weights for now

        hidden_states = self.final_layer_norm(hidden_states)
        layer_output = self.dropout(hidden_states)

        # Add last layer
        if self.output_hidden_states:
            all_hidden_states = all_hidden_states + (hidden_states,)

        outputs = (hidden_states,)
        if self.output_hidden_states:
            outputs = outputs + (all_hidden_states,)
        if self.output_attentions:
            outputs = outputs + (all_attentions,)
        return outputs  # last-layer hidden state, (all hidden states), (all attentions)


T5_START_DOCSTRING = r"""    The T5 model was proposed in
    `Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer`_
    by Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, Peter J. Liu.
    It's an encoder decoder transformer pre-trained in a text-to-text denoising generative setting.

    This model is a PyTorch `torch.nn.Module`_ sub-class. Use it as a regular PyTorch Module and
    refer to the PyTorch documentation for all matter related to general usage and behavior.

    .. _`Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer`:
        https://arxiv.org/abs/1910.10683

    .. _`torch.nn.Module`:
        https://pytorch.org/docs/stable/nn.html#module

    Parameters:
        config (:class:`~transformers.T5Config`): Model configuration class with all the parameters of the model. 
            Initializing with a config file does not load the weights associated with the model, only the configuration.
            Check out the :meth:`~transformers.PreTrainedModel.from_pretrained` method to load the model weights.
"""

T5_INPUTS_DOCSTRING = r"""
    Inputs:
        **input_ids**: ``torch.LongTensor`` of shape ``(batch_size, sequence_length)``:
            Indices of input sequence tokens in the vocabulary.
            To match pre-training, T5 input sequence should be formatted with [CLS] and [SEP] tokens as follows:

            (a) For sequence pairs:

                ``tokens:         [CLS] is this jack ##son ##ville ? [SEP] no it is not . [SEP]``

            (b) For single sequences:

                ``tokens:         [CLS] the dog is hairy . [SEP]``

            T5 is a model with relative position embeddings so you should be able to pad the inputs on
            the right or the left.

            Indices can be obtained using :class:`transformers.T5Tokenizer`.
            See :func:`transformers.PreTrainedTokenizer.encode` and
            :func:`transformers.PreTrainedTokenizer.convert_tokens_to_ids` for details.
        **attention_mask**: (`optional`) ``torch.FloatTensor`` of shape ``(batch_size, sequence_length)``:
            Mask to avoid performing attention on padding token indices.
            Mask values selected in ``[0, 1]``:
            ``1`` for tokens that are NOT MASKED, ``0`` for MASKED tokens.
        **head_mask**: (`optional`) ``torch.FloatTensor`` of shape ``(num_heads,)`` or ``(num_layers, num_heads)``:
            Mask to nullify selected heads of the self-attention modules.
            Mask values selected in ``[0, 1]``:
            ``1`` indicates the head is **not masked**, ``0`` indicates the head is **masked**.
"""

@add_start_docstrings("The bare T5 Model transformer outputting raw hidden-states"
                      "without any specific head on top.",
                      T5_START_DOCSTRING, T5_INPUTS_DOCSTRING)
class T5Model(T5PreTrainedModel):
    r"""
    Outputs: `Tuple` comprising various elements depending on the configuration (config) and inputs:
        **last_hidden_state**: ``torch.FloatTensor`` of shape ``(batch_size, sequence_length, hidden_size)``
            Sequence of hidden-states at the output of the last layer of the model.
        **hidden_states**: (`optional`, returned when ``config.output_hidden_states=True``)
            list of ``torch.FloatTensor`` (one for the output of each layer + the output of the embeddings)
            of shape ``(batch_size, sequence_length, hidden_size)``:
            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
        **attentions**: (`optional`, returned when ``config.output_attentions=True``)
            list of ``torch.FloatTensor`` (one for each layer) of shape ``(batch_size, num_heads, sequence_length, sequence_length)``:
            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.

    Examples::

        tokenizer = T5Tokenizer.from_pretrained('t5-small')
        model = T5Model.from_pretrained('t5-small')
        input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0)  # Batch size 1
        outputs = model(input_ids)
        last_hidden_states = outputs[0]  # The last hidden-state is the first element of the output tuple

    """
    def __init__(self, config):
        super(T5Model, self).__init__(config)
        self.shared = nn.Embedding(config.vocab_size, config.d_model)

        encoder_config = copy.deepcopy(config)
        self.encoder = T5Stack(encoder_config)

        decoder_config = copy.deepcopy(config)
        decoder_config.is_decoder = True
        self.decoder = T5Stack(decoder_config)

        self.init_weights()

    def get_input_embeddings(self):
        return self.shared

    def set_input_embeddings(self, new_embeddings):
        self.shared = new_embeddings

    def _prune_heads(self, heads_to_prune):
        """ Prunes heads of the model.
            heads_to_prune: dict of {layer_num: list of heads to prune in this layer}
            See base class PreTrainedModel
        """
        for layer, heads in heads_to_prune.items():
            self.encoder.layer[layer].attention.prune_heads(heads)

    def forward(self, **kwargs):
        # keyword arguments come in 3 flavors: encoder-specific (prefixed by
        # `encoder_`), decoder-specific (prefixed by `decoder_`) and those
        # that apply to the model as whole.
        # We let the specific kwargs override the common ones in case of conflict.
        kwargs_common = dict((k, v) for k, v in kwargs.items()
                             if not k.startswith("encoder_") and not k.startswith("decoder_"))
        kwargs_encoder = kwargs_common.copy()
        kwargs_decoder = kwargs_common.copy()
        kwargs_encoder.update(dict((k[len("encoder_"):], v) for k, v in kwargs.items() if k.startswith("encoder_")))
        kwargs_decoder.update(dict((k[len("decoder_"):], v) for k, v in kwargs.items() if k.startswith("decoder_")))

        # Encode if needed (training, first prediction pass)
        encoder_hidden_states = kwargs_encoder.pop("hidden_states", None)
        encoder_attention_mask = kwargs_encoder.get("attention_mask", None)
        if encoder_hidden_states is None:
            # Convert encoder inputs in embeddings if needed
            hidden_states = kwargs_encoder.pop("inputs_embeds", None)
            if hidden_states is None:
                encoder_inputs_ids = kwargs_encoder.pop("input_ids")
                hidden_states = self.shared(encoder_inputs_ids)  # Convert inputs in embeddings

            if encoder_attention_mask is not None:
                # Apply masking
                encoder_attention_mask = (encoder_attention_mask != 0).to(hidden_states)
                hidden_states = hidden_states * encoder_attention_mask.unsqueeze(-1)

            encoder_outputs = self.encoder(hidden_states, **kwargs_encoder)
            encoder_hidden_states = encoder_outputs[0]
        else:
            encoder_outputs = ()

        # Decode
        # Convert decoder inputs in embeddings if needed
        hidden_states = kwargs_decoder.pop("inputs_embeds", None)
        if hidden_states is None:
            decoder_inputs_ids = kwargs_decoder.pop("input_ids")
            hidden_states = self.shared(decoder_inputs_ids)

        kwargs_decoder["encoder_hidden_states"] = encoder_hidden_states
        kwargs_decoder["encoder_attention_mask"] = encoder_attention_mask
        decoder_outputs = self.decoder(hidden_states, **kwargs_decoder)

        return decoder_outputs + encoder_outputs


@add_start_docstrings("""T5 Model with a `language modeling` head on top. """,
    T5_START_DOCSTRING, T5_INPUTS_DOCSTRING)
class T5WithLMHeadModel(T5PreTrainedModel):
    r"""
        **lm_labels**: (`optional`) ``torch.LongTensor`` of shape ``(batch_size, sequence_length)``:
            Labels for computing the masked language modeling loss.
            Indices should be in ``[-1, 0, ..., config.vocab_size]`` (see ``input_ids`` docstring)
            Tokens with indices set to ``-1`` are ignored (masked), the loss is only computed for the tokens with labels
            in ``[0, ..., config.vocab_size]``

    Outputs: `Tuple` comprising various elements depending on the configuration (config) and inputs:
        **loss**: (`optional`, returned when ``lm_labels`` is provided) ``torch.FloatTensor`` of shape ``(1,)``:
            Masked language modeling loss.
        **prediction_scores**: ``torch.FloatTensor`` of shape ``(batch_size, sequence_length, config.vocab_size)``
            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
        **hidden_states**: (`optional`, returned when ``config.output_hidden_states=True``)
            list of ``torch.FloatTensor`` (one for the output of each layer + the output of the embeddings)
            of shape ``(batch_size, sequence_length, hidden_size)``:
            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
        **attentions**: (`optional`, returned when ``config.output_attentions=True``)
            list of ``torch.FloatTensor`` (one for each layer) of shape ``(batch_size, num_heads, sequence_length, sequence_length)``:
            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.

    Examples::

        tokenizer = T5Tokenizer.from_pretrained('t5-small')
        model = T5WithLMHeadModel.from_pretrained('t5-small')
        input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0)  # Batch size 1
        outputs = model(input_ids, lm_labels=input_ids)
        loss, prediction_scores = outputs[:2]

    """
    def __init__(self, config):
        super(T5WithLMHeadModel, self).__init__(config)
        self.model_dim = config.d_model

        self.shared = nn.Embedding(config.vocab_size, config.d_model)

        encoder_config = copy.deepcopy(config)
        self.encoder = T5Stack(encoder_config)

        decoder_config = copy.deepcopy(config)
        decoder_config.is_decoder = True
        self.decoder = T5Stack(decoder_config)

        self.lm_head = nn.Linear(config.d_model, config.vocab_size, bias=False)

        self.init_weights()

    def get_input_embeddings(self):
        return self.shared

    def set_input_embeddings(self, new_embeddings):
        self.shared = new_embeddings

    def get_output_embeddings(self):
        return self.lm_head

    def forward(self, **kwargs):
        # keyword arguments come in 3 flavors: encoder-specific (prefixed by
        # `encoder_`), decoder-specific (prefixed by `decoder_`) and those
        # that apply to the model as whole.
        # We let the specific kwargs override the common ones in case of conflict.

        lm_labels = kwargs.pop('decoder_lm_labels', None)

        kwargs_common = dict((k, v) for k, v in kwargs.items()
                             if not k.startswith("encoder_") and not k.startswith("decoder_"))
        kwargs_encoder = kwargs_common.copy()
        kwargs_decoder = kwargs_common.copy()
        kwargs_encoder.update(dict((k[len("encoder_"):], v) for k, v in kwargs.items() if k.startswith("encoder_")))
        kwargs_decoder.update(dict((k[len("decoder_"):], v) for k, v in kwargs.items() if k.startswith("decoder_")))

        # Encode if needed (training, first prediction pass)
        encoder_hidden_states = kwargs_encoder.pop("hidden_states", None)
        if encoder_hidden_states is None:
            # Convert encoder inputs in embeddings if needed
            hidden_states = kwargs_encoder.pop("inputs_embeds", None)
            if hidden_states is None:
                encoder_inputs_ids = kwargs_encoder.pop("input_ids")
                hidden_states = self.shared(encoder_inputs_ids)  # Convert inputs in embeddings

            encoder_outputs = self.encoder(hidden_states, **kwargs_encoder)
            encoder_hidden_states = encoder_outputs[0]
        else:
            encoder_outputs = ()

        # Decode
        # Convert decoder inputs in embeddings if needed
        hidden_states = kwargs_decoder.pop("inputs_embeds", None)
        if hidden_states is None:
            decoder_inputs_ids = kwargs_decoder.pop("input_ids")
            hidden_states = self.shared(decoder_inputs_ids)

        kwargs_decoder["encoder_hidden_states"] = encoder_hidden_states
        kwargs_decoder["encoder_attention_mask"] = kwargs_encoder.get("attention_mask", None)
        decoder_outputs = self.decoder(hidden_states, **kwargs_decoder)

        sequence_output = decoder_outputs[0]
        # Rescale output before projecting on vocab
        # See https://github.com/tensorflow/mesh/blob/fa19d69eafc9a482aff0b59ddd96b025c0cb207d/mesh_tensorflow/transformer/transformer.py#L586
        sequence_output = sequence_output * (self.model_dim ** -0.5)
        lm_logits = self.lm_head(sequence_output)

        decoder_outputs = (lm_logits,) + decoder_outputs[1:]  # Add hidden states and attention if they are here
        if lm_labels is not None:
            shift_logits = lm_logits[..., :-1, :].contiguous()
            shift_labels = lm_labels[..., 1:].contiguous()
            loss_fct = CrossEntropyLoss(ignore_index=-1)
            loss = loss_fct(shift_logits.view(-1, shift_logits.size(-1)),
                            shift_labels.view(-1))
            decoder_outputs = (loss,) + decoder_outputs  # TODO(thom): Add z_loss https://github.com/tensorflow/mesh/blob/fa19d69eafc9a482aff0b59ddd96b025c0cb207d/mesh_tensorflow/layers.py#L666

        return decoder_outputs + encoder_outputs