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. """
""" Copy-pasted from Transformers 4.0.1 """


import copy
import math
import os

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

from transformers.activations import ACT2FN
from transformers.file_utils import (
    DUMMY_INPUTS,
    DUMMY_MASK,
    add_start_docstrings,
    add_start_docstrings_to_model_forward,
    replace_return_docstrings,
)
from transformers.modeling_outputs import (
    BaseModelOutput,
    BaseModelOutputWithPastAndCrossAttentions,
    Seq2SeqLMOutput,
    Seq2SeqModelOutput,
)
from transformers.modeling_utils import PreTrainedModel, find_pruneable_heads_and_indices, prune_linear_layer
from transformers.utils import logging
from transformers.models.t5.configuration_t5 import T5Config

from lm_experiments_tools.utils import get_cls_by_name


logger = logging.get_logger(__name__)

_CONFIG_FOR_DOC = "T5Config"
_TOKENIZER_FOR_DOC = "T5Tokenizer"

####################################################
# This dict contains ids and associated url
# for the pretrained weights provided with the models
####################################################
T5_PRETRAINED_MODEL_ARCHIVE_LIST = [
    "t5-small",
    "t5-base",
    "t5-large",
    "t5-3b",
    "t5-11b",
    # See all T5 models at https://huggingface.co/models?filter=t5
]


####################################################
# 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", "AdamWeightDecayOptimizer", "AdamWeightDecayOptimizer_1", "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):
                scope_names = re.split(r"_(\d+)", m_name)
            else:
                scope_names = [m_name]
            if scope_names[0] in ["kernel", "scale", "embedding"]:
                pointer = getattr(pointer, "weight")
            elif scope_names[0] == "self_attention":
                pointer = getattr(pointer, "layer")
                pointer = pointer[0]
            elif scope_names[0] == "enc_dec_attention":
                pointer = getattr(pointer, "layer")
                pointer = pointer[1]
            elif scope_names[0] == "dense_relu_dense":
                pointer = getattr(pointer, "layer")
                pointer = pointer[2]
            elif scope_names[0] == "rms_norm":
                if hasattr(pointer, "layer_norm"):
                    pointer = getattr(pointer, "layer_norm")
                elif hasattr(pointer, "final_layer_norm"):
                    pointer = getattr(pointer, "final_layer_norm")
            elif scope_names[0] == "scale":
                pointer = getattr(pointer, "weight")
            elif scope_names[0] == "output_bias" or scope_names[0] == "beta":
                pointer = getattr(pointer, "bias")
            elif scope_names[0] == "squad":
                pointer = getattr(pointer, "classifier")
            elif scope_names[0] == "decoder" and name[1] == "logits":
                continue
            elif scope_names[0] == "logits":
                pointer = getattr(pointer, "lm_head")
            elif scope_names[0] == "wi" and len(scope_names) > 1 and scope_names[1].isdigit():
                pointer = getattr(pointer, f"wi_{scope_names[1]}")
                continue
            else:
                try:
                    pointer = getattr(pointer, scope_names[0])
                except AttributeError:
                    logger.info("Skipping {}".format("/".join(name)))
                    continue
            if len(scope_names) >= 2:
                num = int(scope_names[1])
                pointer = pointer[num]
        if scope_names[0] not in ["kernel", "scale", "embedding"]:
            pointer = getattr(pointer, "weight")
        if scope_names[0] != "embedding":
            logger.info("Transposing numpy weight of shape {} for {}".format(array.shape, name))
            array = np.transpose(array)
        try:
            assert (
                pointer.shape == array.shape
            ), f"Pointer shape {pointer.shape} and array shape {array.shape} mismatched"
        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())))
    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 subtraction of mean.
        """
        super().__init__()
        self.weight = nn.Parameter(torch.ones(hidden_size))
        self.variance_epsilon = eps

    def forward(self, hidden_states):
        # layer norm should always be calculated in float32
        variance = hidden_states.to(torch.float32).pow(2).mean(-1, keepdim=True)
        hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)

        # convert into float16 if necessary
        if self.weight.dtype == torch.float16:
            hidden_states = hidden_states.to(torch.float16)
        return self.weight * hidden_states


class T5DenseReluDense(nn.Module):
    def __init__(self, config):
        super().__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):
        hidden_states = self.wi(hidden_states)
        hidden_states = F.relu(hidden_states)
        hidden_states = self.dropout(hidden_states)
        hidden_states = self.wo(hidden_states)
        return hidden_states


class T5DenseGatedGeluDense(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.wi_0 = nn.Linear(config.d_model, config.d_ff, bias=False)
        self.wi_1 = 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)
        self.gelu_act = ACT2FN["gelu_new"]

    def forward(self, hidden_states):
        hidden_gelu = self.gelu_act(self.wi_0(hidden_states))
        hidden_linear = self.wi_1(hidden_states)
        hidden_states = hidden_gelu * hidden_linear
        hidden_states = self.dropout(hidden_states)
        hidden_states = self.wo(hidden_states)
        return hidden_states


class T5LayerFF(nn.Module):
    def __init__(self, config):
        super().__init__()
        if config.feed_forward_proj == "relu":
            self.DenseReluDense = T5DenseReluDense(config)
        elif config.feed_forward_proj == "gated-gelu":
            self.DenseReluDense = T5DenseGatedGeluDense(config)
        else:
            raise ValueError(
                f"{self.config.feed_forward_proj} is not supported. Choose between `relu` and `gated-gelu`"
            )

        self.layer_norm = T5LayerNorm(config.d_model, eps=config.layer_norm_epsilon)
        self.dropout = nn.Dropout(config.dropout_rate)

    def forward(self, hidden_states):
        forwarded_states = self.layer_norm(hidden_states)
        forwarded_states = self.DenseReluDense(forwarded_states)
        hidden_states = hidden_states + self.dropout(forwarded_states)
        return hidden_states


class T5Attention(nn.Module):
    def __init__(self, config: T5Config, has_relative_attention_bias=False):
        super().__init__()
        self.is_decoder = config.is_decoder
        self.has_relative_attention_bias = has_relative_attention_bias

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

        # 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
        heads, index = find_pruneable_heads_and_indices(
            heads, self.n_heads, self.key_value_proj_dim, self.pruned_heads
        )
        # 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.key_value_proj_dim * 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)
        """
        relative_buckets = 0
        if bidirectional:
            num_buckets //= 2
            relative_buckets += (relative_position > 0).to(torch.long) * num_buckets
            relative_position = torch.abs(relative_position)
        else:
            relative_position = -torch.min(relative_position, torch.zeros_like(relative_position))
        # now relative_position is in the range [0, inf)

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

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

        relative_buckets += torch.where(is_small, relative_position, relative_postion_if_large)
        return relative_buckets

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

    def forward(
        self,
        hidden_states,
        mask=None,
        key_value_states=None,
        position_bias=None,
        past_key_value=None,
        head_mask=None,
        query_length=None,
        use_cache=False,
        output_attentions=False,
        output_normalized_attentions=False,
        output_head_hiddens=False,
    ):
        """
        Self-attention (if key_value_states is None) or attention over source sentence (provided by key_value_states).
        """
        # Input is (batch_size, seq_length, dim)
        # Mask is (batch_size, key_length) (non-causal) or (batch_size, key_length, key_length)
        # past_key_value[0] is (batch_size, n_heads, q_len - 1, dim_per_head)
        batch_size, seq_length = hidden_states.shape[:2]

        real_seq_length = seq_length

        if past_key_value is not None:
            assert (
                len(past_key_value) == 2
            ), "past_key_value should have 2 past states: keys and values. Got {} past states".format(
                len(past_key_value)
            )
            real_seq_length += past_key_value[0].shape[2] if query_length is None else query_length

        key_length = real_seq_length if key_value_states is None else key_value_states.shape[1]

        def shape(states):
            """  projection """
            return states.view(batch_size, -1, self.n_heads, self.key_value_proj_dim).transpose(1, 2)

        def unshape(states):
            """  reshape """
            return states.transpose(1, 2).contiguous().view(batch_size, -1, self.inner_dim)

        def project(hidden_states, proj_layer, key_value_states, past_key_value):
            """ projects hidden states correctly to key/query states """
            if key_value_states is None:
                # self-attn
                # (batch_size, n_heads, seq_length, dim_per_head)
                hidden_states = shape(proj_layer(hidden_states))
            elif past_key_value is None:
                # cross-attn
                # (batch_size, n_heads, seq_length, dim_per_head)
                hidden_states = shape(proj_layer(key_value_states))

            if past_key_value is not None:
                if key_value_states is None:
                    # self-attn
                    # (batch_size, n_heads, key_length, dim_per_head)
                    hidden_states = torch.cat([past_key_value, hidden_states], dim=2)
                else:
                    # cross-attn
                    hidden_states = past_key_value
            return hidden_states

        # get query states
        query_states = shape(self.q(hidden_states))  # (batch_size, n_heads, seq_length, dim_per_head)

        # get key/value states
        key_states = project(
            hidden_states, self.k, key_value_states, past_key_value[0] if past_key_value is not None else None
        )
        value_states = project(
            hidden_states, self.v, key_value_states, past_key_value[1] if past_key_value is not None else None
        )

        # compute scores
        scores = torch.matmul(
            query_states, key_states.transpose(3, 2)
        )  # equivalent of torch.einsum("bnqd,bnkd->bnqk", query_states, key_states), compatible with onnx op>9

        if position_bias is None:
            if not self.has_relative_attention_bias:
                position_bias = torch.zeros(
                    (1, self.n_heads, real_seq_length, key_length), device=scores.device, dtype=scores.dtype
                )
            else:
                position_bias = self.compute_bias(real_seq_length, key_length)

            # if key and values are already calculated
            # we want only the last query position bias
            if past_key_value is not None:
                position_bias = position_bias[:, :, -seq_length:, :]

            if mask is not None:
                position_bias = position_bias + mask  # (batch_size, n_heads, seq_length, key_length)

        scores += position_bias
        attn_weights = F.softmax(scores.float(), dim=-1).type_as(
            scores
        )  # (batch_size, n_heads, seq_length, key_length)
        attn_weights = F.dropout(
            attn_weights, p=self.dropout, training=self.training
        )  # (batch_size, n_heads, seq_length, key_length)

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

        attn_heads = unshape(torch.matmul(attn_weights, value_states))  # (batch_size, seq_length, dim)
        attn_output = self.o(attn_heads)

        present_key_value_state = (key_states, value_states) if (self.is_decoder and use_cache) else None
        outputs = (attn_output,) + (present_key_value_state,) + (position_bias,)

        if output_attentions:
            if output_normalized_attentions:
                attn_weights = attn_weights * torch.linalg.norm(value_states, dim=-1).unsqueeze(2)
            if output_head_hiddens:
                outputs = outputs + ((attn_heads, attn_weights),)
            else:
                outputs = outputs + (attn_weights,)
        return outputs


class T5LayerSelfAttention(nn.Module):
    def __init__(self, config, has_relative_attention_bias=False):
        super().__init__()
        if config.is_decoder and hasattr(config, 'decoder_self_attention_layer'):
            Attention = get_cls_by_name(config.decoder_self_attention_layer)
        elif not config.is_decoder and hasattr(config, 'encoder_self_attention_layer'):
            Attention = get_cls_by_name(config.encoder_self_attention_layer)
        else:
            Attention = T5Attention
        self.SelfAttention = Attention(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,
        past_key_value=None,
        use_cache=False,
        output_attentions=False,
        output_normalized_attentions=False,
        output_head_hiddens=False,
    ):
        normed_hidden_states = self.layer_norm(hidden_states)
        attention_output = self.SelfAttention(
            normed_hidden_states,
            mask=attention_mask,
            position_bias=position_bias,
            head_mask=head_mask,
            past_key_value=past_key_value,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_normalized_attentions=output_normalized_attentions,
            output_head_hiddens=output_head_hiddens
        )
        hidden_states = hidden_states + self.dropout(attention_output[0])
        outputs = (hidden_states,) + attention_output[1:]  # add attentions if we output them
        return outputs


class T5LayerCrossAttention(nn.Module):
    def __init__(self, config):
        super().__init__()
        if config.is_decoder and hasattr(config, 'decoder_cross_attention_layer'):
            Attention = get_cls_by_name(config.decoder_cross_attention_layer)
        else:
            Attention = T5Attention
        self.EncDecAttention = Attention(config, has_relative_attention_bias=False)
        self.layer_norm = T5LayerNorm(config.d_model, eps=config.layer_norm_epsilon)
        self.dropout = nn.Dropout(config.dropout_rate)

    def forward(
        self,
        hidden_states,
        key_value_states,
        attention_mask=None,
        position_bias=None,
        head_mask=None,
        past_key_value=None,
        use_cache=False,
        query_length=None,
        output_attentions=False,
        output_normalized_attentions=False,
        output_head_hiddens=False,
    ):
        normed_hidden_states = self.layer_norm(hidden_states)
        attention_output = self.EncDecAttention(
            normed_hidden_states,
            mask=attention_mask,
            key_value_states=key_value_states,
            position_bias=position_bias,
            head_mask=head_mask,
            past_key_value=past_key_value,
            use_cache=use_cache,
            query_length=query_length,
            output_attentions=output_attentions,
            output_normalized_attentions=output_normalized_attentions,
            output_head_hiddens=output_head_hiddens
        )
        layer_output = hidden_states + self.dropout(attention_output[0])
        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().__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))

        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,
        past_key_value=None,
        use_cache=False,
        output_attentions=False,
        output_normalized_attentions=False,
        output_head_hiddens=False,
        return_dict=True,
    ):

        if past_key_value is not None:
            assert self.is_decoder, "Only decoder can use `past_key_values`"
            expected_num_past_key_values = 2 if encoder_hidden_states is None else 4

            error_message = "There should be {} past states. 2 (past / key) for self attention.{} Got {} past key / value states".format(
                expected_num_past_key_values,
                "2 (past / key) for cross attention" if expected_num_past_key_values == 4 else "",
                len(past_key_value),
            )
            assert len(past_key_value) == expected_num_past_key_values, error_message

            self_attn_past_key_value = past_key_value[:2]
            cross_attn_past_key_value = past_key_value[2:]
        else:
            self_attn_past_key_value, cross_attn_past_key_value = None, None

        self_attention_outputs = self.layer[0](
            hidden_states,
            attention_mask=attention_mask,
            position_bias=position_bias,
            head_mask=head_mask,
            past_key_value=self_attn_past_key_value,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_normalized_attentions=output_normalized_attentions,
            output_head_hiddens=output_head_hiddens
        )
        hidden_states, present_key_value_state = self_attention_outputs[:2]
        attention_outputs = self_attention_outputs[2:]  # Keep self-attention outputs and relative position weights

        do_cross_attention = self.is_decoder and encoder_hidden_states is not None
        if do_cross_attention:
            # the actual query length is unknown for cross attention
            # if using past key value states. Need to inject it here
            if present_key_value_state is not None:
                query_length = present_key_value_state[0].shape[2]
            else:
                query_length = None

            cross_attention_outputs = self.layer[1](
                hidden_states,
                key_value_states=encoder_hidden_states,
                attention_mask=encoder_attention_mask,
                position_bias=encoder_decoder_position_bias,
                head_mask=head_mask,
                past_key_value=cross_attn_past_key_value,
                query_length=query_length,
                use_cache=use_cache,
                output_attentions=output_attentions,
                output_normalized_attentions=output_normalized_attentions,
                output_head_hiddens=output_head_hiddens,
            )
            hidden_states = cross_attention_outputs[0]
            # Combine self attn and cross attn key value states
            if present_key_value_state is not None:
                present_key_value_state = present_key_value_state + cross_attention_outputs[1]

            # Keep cross-attention outputs and relative position weights
            attention_outputs = attention_outputs + cross_attention_outputs[2:]

        # Apply Feed Forward layer
        hidden_states = self.layer[-1](hidden_states)
        outputs = (hidden_states,)

        outputs = outputs + (present_key_value_state,) + attention_outputs
        return outputs  # hidden-states, present_key_value_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 downloading and loading pretrained
    models.
    """

    config_class = T5Config
    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,
            "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, T5ForConditionalGeneration)):
            # 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, T5DenseGatedGeluDense):
            module.wi_0.weight.data.normal_(mean=0.0, std=factor * ((self.config.d_model) ** -0.5))
            if hasattr(module.wi_0, "bias") and module.wi_0.bias is not None:
                module.wi_0.bias.data.zero_()
            module.wi_1.weight.data.normal_(mean=0.0, std=factor * ((self.config.d_model) ** -0.5))
            if hasattr(module.wi_1, "bias") and module.wi_1.bias is not None:
                module.wi_1.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, T5CDQAttention):
            d_model = self.config.d_model
            key_value_proj_dim = self.config.d_kv
            n_heads = self.config.num_heads - self.config.num_cdq_heads
            n_cdq_heads = self.config.num_cdq_heads
            if n_heads > 0:
                module.q.weight.data.normal_(mean=0.0, std=factor * ((d_model * key_value_proj_dim) ** -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 + n_cdq_heads) * key_value_proj_dim) ** -0.5))
            module.T.weight.data.normal_(mean=0.0, std=factor * ((n_cdq_heads * d_model) ** -0.5))
            module.W_t.weight.data.normal_(mean=0.0, std=factor * ((d_model * d_model) ** -0.5))
            module.cd_q.weight.data.normal_(mean=0.0, std=factor * ((d_model * key_value_proj_dim) ** -0.5))
            module.cd_w.weight.data.normal_(mean=0.0, std=factor * ((d_model * 2 * d_model)) ** -0.5)
            if module.has_relative_attention_bias:
                module.relative_attention_bias.weight.data.normal_(mean=0.0, std=factor * ((d_model) ** -0.5))
        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
            key_value_proj_dim = self.config.d_kv
            n_heads = self.config.num_heads
            module.q.weight.data.normal_(mean=0.0, std=factor * ((d_model * key_value_proj_dim) ** -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 * key_value_proj_dim) ** -0.5))
            if module.has_relative_attention_bias:
                module.relative_attention_bias.weight.data.normal_(mean=0.0, std=factor * ((d_model) ** -0.5))

    def _shift_right(self, input_ids, insert_token_id=None):
        if insert_token_id is None:
            insert_token_id = self.config.decoder_start_token_id
        pad_token_id = self.config.pad_token_id

        assert (
            insert_token_id is not None
        ), "self.model.config.decoder_start_token_id has to be defined. In T5 it is usually set to the pad_token_id. See T5 docs for more information"

        # shift inputs to the right
        shifted_input_ids = input_ids.new_zeros(input_ids.shape)
        shifted_input_ids[..., 1:] = input_ids[..., :-1].clone()
        shifted_input_ids[..., 0] = insert_token_id

        assert pad_token_id is not None, "self.model.config.pad_token_id has to be defined."
        # replace possible -100 values in labels by `pad_token_id`
        shifted_input_ids.masked_fill_(shifted_input_ids == -100, pad_token_id)

        assert torch.all(shifted_input_ids >= 0).item(), "Verify that `shifted_input_ids` has only positive values"

        return shifted_input_ids


class T5Stack(T5PreTrainedModel):
    def __init__(self, config, embed_tokens=None, memory_tokens=None):
        super().__init__(config)

        self.embed_tokens = embed_tokens
        self.memory_tokens = memory_tokens
        self.num_memory_tokens = config.num_memory_tokens if hasattr(config, 'num_memory_tokens') else 0
        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 get_input_embeddings(self):
        return self.embed_tokens

    def get_output_embeddings(self):
        return self.embed_tokens

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

    def forward(
        self,
        input_ids=None,
        attention_mask=None,
        encoder_hidden_states=None,
        encoder_attention_mask=None,
        inputs_embeds=None,
        head_mask=None,
        past_key_values=None,
        use_cache=None,
        output_attentions=None,
        output_normalized_attentions=False,
        output_head_hiddens=False,
        output_hidden_states=None,
        return_dict=None,
    ):

        use_cache = use_cache if use_cache is not None else self.config.use_cache
        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
        output_hidden_states = (
            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
        )
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        if input_ids is not None and inputs_embeds is not None:
            err_msg_prefix = "decoder_" if self.is_decoder else ""
            raise ValueError(
                f"You cannot specify both {err_msg_prefix}inputs and {err_msg_prefix}inputs_embeds at the same time"
            )
        elif input_ids is not None:
            input_shape = input_ids.size()
            input_ids = input_ids.view(-1, input_shape[-1])
        elif inputs_embeds is not None:
            input_shape = inputs_embeds.size()[:-1]
        else:
            err_msg_prefix = "decoder_" if self.is_decoder else ""
            raise ValueError(f"You have to specify either {err_msg_prefix}inputs or {err_msg_prefix}inputs_embeds")

        if inputs_embeds is None:
            assert self.embed_tokens is not None, "You have to initialize the model with valid token embeddings"
            inputs_embeds = self.embed_tokens(input_ids)

        batch_size, seq_length = input_shape

        if self.memory_tokens is not None:
            # used in encoder
            seq_length += self.num_memory_tokens
            inputs_embeds = torch.cat([self.memory_tokens.repeat(batch_size, 1, 1), inputs_embeds], dim=1)
            if attention_mask is not None:
                attention_mask = F.pad(attention_mask, (self.num_memory_tokens, 0), value=1)

        # required mask seq length can be calculated via length of past
        mask_seq_length = past_key_values[0][0].shape[2] + seq_length if past_key_values is not None else seq_length

        if use_cache is True:
            assert self.is_decoder, ":obj:`use_cache` can only be set to `True` if {} is used as a decoder".format(
                self
            )

        if attention_mask is None:
            attention_mask = torch.ones(batch_size, mask_seq_length).to(inputs_embeds.device)
        if self.is_decoder and encoder_attention_mask is None and encoder_hidden_states is not None:
            encoder_seq_length = encoder_hidden_states.shape[1]
            encoder_attention_mask = torch.ones(
                batch_size, encoder_seq_length, device=inputs_embeds.device, dtype=torch.long
            )
        elif self.is_decoder and self.num_memory_tokens > 0 and encoder_attention_mask is not None:
            encoder_attention_mask = F.pad(encoder_attention_mask, (self.num_memory_tokens, 0), value=1)

        # initialize past_key_values with `None` if past does not exist
        if past_key_values is None:
            past_key_values = [None] * len(self.block)

        # ourselves in which case we just need to make it broadcastable to all heads.
        extended_attention_mask = self.get_extended_attention_mask(attention_mask, input_shape, inputs_embeds.device)

        if self.is_decoder and encoder_attention_mask is not None:
            encoder_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask)
        else:
            encoder_extended_attention_mask = None

        # Prepare head mask if needed
        head_mask = self.get_head_mask(head_mask, self.config.num_layers)
        present_key_value_states = () if use_cache else None
        all_hidden_states = () if output_hidden_states else None
        all_attentions = () if output_attentions else None
        all_cross_attentions = () if (output_attentions and self.is_decoder) else None
        position_bias = None
        encoder_decoder_position_bias = None

        hidden_states = self.dropout(inputs_embeds)

        for i, (layer_module, past_key_value) in enumerate(zip(self.block, past_key_values)):
            if 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],
                past_key_value=past_key_value,
                use_cache=use_cache,
                output_attentions=output_attentions,
                output_normalized_attentions=output_normalized_attentions,
                output_head_hiddens=output_head_hiddens,
            )
            # layer_outputs is a tuple with:
            # hidden-states, key-value-states, (self-attention weights), (self-attention position bias), (cross-attention weights), (cross-attention position bias)
            hidden_states, present_key_value_state = layer_outputs[:2]

            # We share the position biases between the layers - the first layer store them
            # layer_outputs = hidden-states, key-value-states (self-attention weights),
            # (self-attention position bias), (cross-attention weights), (cross-attention position bias)
            position_bias = layer_outputs[2]
            if self.is_decoder and encoder_hidden_states is not None:
                encoder_decoder_position_bias = layer_outputs[4 if output_attentions else 3]
            # append next layer key value states
            if use_cache:
                present_key_value_states = present_key_value_states + (present_key_value_state,)

            if output_attentions:
                all_attentions = all_attentions + (layer_outputs[3],)
                if self.is_decoder:
                    all_cross_attentions = all_cross_attentions + (layer_outputs[5],)

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

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

        if not return_dict:
            return tuple(
                v
                for v in [
                    hidden_states,
                    present_key_value_states,
                    all_hidden_states,
                    all_attentions,
                    all_cross_attentions,
                ]
                if v is not None
            )
        return BaseModelOutputWithPastAndCrossAttentions(
            last_hidden_state=hidden_states,
            past_key_values=present_key_value_states,
            hidden_states=all_hidden_states,
            attentions=all_attentions,
            cross_attentions=all_cross_attentions,
        )


T5_START_DOCSTRING = r"""

    The T5 model was proposed in `Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer
    <https://arxiv.org/abs/1910.10683>`__ 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 inherits from :class:`~transformers.PreTrainedModel`. Check the superclass documentation for the generic
    methods the library implements for all its model (such as downloading or saving, resizing the input embeddings,
    pruning heads etc.)

    This model is also a PyTorch `torch.nn.Module <https://pytorch.org/docs/stable/nn.html#torch.nn.Module>`__
    subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to
    general usage and behavior.

    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"""
    Args:
        input_ids (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`):
            Indices of input sequence tokens in the vocabulary. T5 is a model with relative position embeddings so you
            should be able to pad the inputs on both the right and the left.

            Indices can be obtained using :class:`~transformers.T5Tokenizer`. See
            :meth:`transformers.PreTrainedTokenizer.encode` and :meth:`transformers.PreTrainedTokenizer.__call__` for
            detail.

            To know more on how to prepare :obj:`input_ids` for pretraining take a look a `T5 Training
            <./t5.html#training>`__.
        attention_mask (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
            Mask to avoid performing attention on padding token indices. Mask values selected in ``[0, 1]``:

            - 1 for tokens that are **not masked**,
            - 0 for tokens that are **masked**.

            `What are attention masks? <../glossary.html#attention-mask>`__
        decoder_input_ids (:obj:`torch.LongTensor` of shape :obj:`(batch_size, target_sequence_length)`, `optional`):
            Provide for sequence to sequence training. T5 uses the :obj:`pad_token_id` as the starting token for
            :obj:`decoder_input_ids` generation. If :obj:`past_key_values` is used, optionally only the last
            :obj:`decoder_input_ids` have to be input (see :obj:`past_key_values`).

            To know more on how to prepare :obj:`decoder_input_ids` for pretraining take a look at `T5 Training
            <./t5.html#training>`__. If :obj:`decoder_input_ids` and :obj:`decoder_inputs_embeds` are both unset,
            :obj:`decoder_input_ids` takes the value of :obj:`input_ids`.
        decoder_attention_mask (:obj:`torch.BoolTensor` of shape :obj:`(batch_size, tgt_seq_len)`, `optional`):
            Default behavior: generate a tensor that ignores pad tokens in :obj:`decoder_input_ids`. Causal mask will
            also be used by default.
        encoder_outputs (:obj:`tuple(tuple(torch.FloatTensor)`, `optional`):
            Tuple consists of (:obj:`last_hidden_state`, :obj:`optional`: `hidden_states`, :obj:`optional`:
            `attentions`) :obj:`last_hidden_state` of shape :obj:`(batch_size, sequence_length, hidden_size)` is a
            sequence of hidden states at the output of the last layer of the encoder. Used in the cross-attention of
            the decoder.
        past_key_values (:obj:`tuple(tuple(torch.FloatTensor))` of length :obj:`config.n_layers` with each tuple having 4 tensors of shape :obj:`(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`):
            Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up decoding.

            If :obj:`past_key_values` are used, the user can optionally input only the last :obj:`decoder_input_ids`
            (those that don't have their past key value states given to this model) of shape :obj:`(batch_size, 1)`
            instead of all :obj:`decoder_input_ids` of shape :obj:`(batch_size, sequence_length)`.
        head_mask (:obj:`torch.FloatTensor` of shape :obj:`(num_heads,)` or :obj:`(num_layers, num_heads)`, `optional`):
            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**.

        inputs_embeds (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):
            Optionally, instead of passing :obj:`input_ids` you can choose to directly pass an embedded representation.
            This is useful if you want more control over how to convert :obj:`input_ids` indices into associated
            vectors than the model's internal embedding lookup matrix.
        decoder_inputs_embeds (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, target_sequence_length, hidden_size)`, `optional`):
            Optionally, instead of passing :obj:`decoder_input_ids` you can choose to directly pass an embedded
            representation. If :obj:`past_key_values` is used, optionally only the last :obj:`decoder_inputs_embeds`
            have to be input (see :obj:`past_key_values`). This is useful if you want more control over how to convert
            :obj:`decoder_input_ids` indices into associated vectors than the model's internal embedding lookup matrix.

            If :obj:`decoder_input_ids` and :obj:`decoder_inputs_embeds` are both unset, :obj:`decoder_inputs_embeds`
            takes the value of :obj:`inputs_embeds`.

        use_cache (:obj:`bool`, `optional`):
            If set to :obj:`True`, :obj:`past_key_values` key value states are returned and can be used to speed up
            decoding (see :obj:`past_key_values`).

        output_attentions (:obj:`bool`, `optional`):
            Whether or not to return the attentions tensors of all attention layers. See ``attentions`` under returned
            tensors for more detail.
        output_hidden_states (:obj:`bool`, `optional`):
            Whether or not to return the hidden states of all layers. See ``hidden_states`` under returned tensors for
            more detail.
        return_dict (:obj:`bool`, `optional`):
            Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple.
"""


@add_start_docstrings(
    "The bare T5 Model transformer outputting raw hidden-states" "without any specific head on top.",
    T5_START_DOCSTRING,
)
class T5Model(T5PreTrainedModel):
    _keys_to_ignore_on_load_missing = [
        r"encoder\.embed_tokens\.weight",
        r"decoder\.embed_tokens\.weight",
        r"decoder\.block\.0\.layer\.1\.EncDecAttention\.relative_attention_bias\.weight",
    ]

    def __init__(self, config: T5Config):
        super().__init__(config)
        self.shared = nn.Embedding(config.vocab_size, config.d_model)

        encoder_config = copy.deepcopy(config)
        encoder_config.use_cache = False
        encoder_config.is_encoder_decoder = False
        self.encoder = T5Stack(encoder_config, self.shared)

        decoder_config = copy.deepcopy(config)
        decoder_config.is_decoder = True
        decoder_config.is_encoder_decoder = False
        decoder_config.num_layers = config.num_decoder_layers
        self.decoder = T5Stack(decoder_config, self.shared)

        self.init_weights()

    def get_input_embeddings(self):
        return self.shared

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

    def get_encoder(self):
        return self.encoder

    def get_decoder(self):
        return self.decoder

    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)

    @add_start_docstrings_to_model_forward(T5_INPUTS_DOCSTRING)
    @replace_return_docstrings(output_type=Seq2SeqModelOutput, config_class=_CONFIG_FOR_DOC)
    def forward(
        self,
        input_ids=None,
        attention_mask=None,
        decoder_input_ids=None,
        decoder_attention_mask=None,
        encoder_outputs=None,
        past_key_values=None,
        head_mask=None,
        inputs_embeds=None,
        decoder_inputs_embeds=None,
        use_cache=None,
        output_attentions=None,
        output_hidden_states=None,
        return_dict=None,
    ):
        r"""
        Returns:

        Example::

            >>> from transformers import T5Tokenizer, T5Model

            >>> tokenizer = T5Tokenizer.from_pretrained('t5-small')
            >>> model = T5Model.from_pretrained('t5-small')

            >>> input_ids = tokenizer("Studies have been shown that owning a dog is good for you", return_tensors="pt").input_ids  # Batch size 1
            >>> decoder_input_ids = tokenizer("Studies show that", return_tensors="pt").input_ids  # Batch size 1
            >>> outputs = model(input_ids=input_ids, decoder_input_ids=decoder_input_ids)

            >>> last_hidden_states = outputs.last_hidden_state
        """
        use_cache = use_cache if use_cache is not None else self.config.use_cache
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        # Encode if needed (training, first prediction pass)
        if encoder_outputs is None:
            encoder_outputs = self.encoder(
                input_ids=input_ids,
                attention_mask=attention_mask,
                inputs_embeds=inputs_embeds,
                head_mask=head_mask,
                output_attentions=output_attentions,
                output_hidden_states=output_hidden_states,
                return_dict=return_dict,
            )
        elif return_dict and not isinstance(encoder_outputs, BaseModelOutput):
            encoder_outputs = BaseModelOutput(
                last_hidden_state=encoder_outputs[0],
                hidden_states=encoder_outputs[1] if len(encoder_outputs) > 1 else None,
                attentions=encoder_outputs[2] if len(encoder_outputs) > 2 else None,
            )

        hidden_states = encoder_outputs[0]

        # Decode
        decoder_outputs = self.decoder(
            input_ids=decoder_input_ids,
            attention_mask=decoder_attention_mask,
            inputs_embeds=decoder_inputs_embeds,
            past_key_values=past_key_values,
            encoder_hidden_states=hidden_states,
            encoder_attention_mask=attention_mask,
            head_mask=head_mask,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )

        if not return_dict:
            return decoder_outputs + encoder_outputs

        return Seq2SeqModelOutput(
            last_hidden_state=decoder_outputs.last_hidden_state,
            past_key_values=decoder_outputs.past_key_values,
            decoder_hidden_states=decoder_outputs.hidden_states,
            decoder_attentions=decoder_outputs.attentions,
            cross_attentions=decoder_outputs.cross_attentions,
            encoder_last_hidden_state=encoder_outputs.last_hidden_state,
            encoder_hidden_states=encoder_outputs.hidden_states,
            encoder_attentions=encoder_outputs.attentions,
        )


@add_start_docstrings("""T5 Model with a `language modeling` head on top. """, T5_START_DOCSTRING)
class T5ForConditionalGeneration(T5PreTrainedModel):
    _keys_to_ignore_on_load_missing = [
        r"encoder\.embed_tokens\.weight",
        r"decoder\.embed_tokens\.weight",
        r"lm_head\.weight",
        r"decoder\.block\.0\.layer\.1\.EncDecAttention\.relative_attention_bias\.weight",
    ]

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

        self.shared = nn.Embedding(config.vocab_size, config.d_model)
        self.num_memory_tokens = config.num_memory_tokens if hasattr(config, 'num_memory_tokens') else 0
        if self.num_memory_tokens > 0:
            self.memory_tokens = nn.Parameter(
                torch.randn(config.num_memory_tokens, config.d_model) * config.initializer_factor)
        else:
            self.memory_tokens = None

        encoder_config = copy.deepcopy(config)
        encoder_config.use_cache = False
        encoder_config.is_encoder_decoder = False
        self.encoder = T5Stack(encoder_config, self.shared, self.memory_tokens)

        decoder_config = copy.deepcopy(config)
        decoder_config.is_decoder = True
        decoder_config.is_encoder_decoder = False
        decoder_config.num_layers = config.num_decoder_layers
        self.decoder = T5Stack(decoder_config, self.shared)

        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
        self.encoder.set_input_embeddings(new_embeddings)
        self.decoder.set_input_embeddings(new_embeddings)

    def get_output_embeddings(self):
        return self.lm_head

    def get_encoder(self):
        return self.encoder

    def get_decoder(self):
        return self.decoder

    @add_start_docstrings_to_model_forward(T5_INPUTS_DOCSTRING)
    @replace_return_docstrings(output_type=Seq2SeqLMOutput, config_class=_CONFIG_FOR_DOC)
    def forward(
        self,
        input_ids=None,
        attention_mask=None,
        decoder_input_ids=None,
        decoder_attention_mask=None,
        encoder_outputs=None,
        past_key_values=None,
        head_mask=None,
        inputs_embeds=None,
        decoder_inputs_embeds=None,
        labels=None,
        use_cache=None,
        output_attentions=None,
        output_normalized_attentions=False,
        output_head_hiddens=False,
        output_hidden_states=None,
        return_dict=None,
    ):
        r"""
        labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size,)`, `optional`):
            Labels for computing the sequence classification/regression loss. Indices should be in :obj:`[-100, 0, ...,
            config.vocab_size - 1]`. All labels set to ``-100`` are ignored (masked), the loss is only computed for
            labels in ``[0, ..., config.vocab_size]``

        Returns:

        Examples::

            >>> from transformers import T5Tokenizer, T5ForConditionalGeneration

            >>> tokenizer = T5Tokenizer.from_pretrained('t5-small')
            >>> model = T5ForConditionalGeneration.from_pretrained('t5-small')

            >>> input_ids = tokenizer('The <extra_id_0> walks in <extra_id_1> park', return_tensors='pt').input_ids
            >>> labels = tokenizer('<extra_id_0> cute dog <extra_id_1> the <extra_id_2> </s>', return_tensors='pt').input_ids
            >>> outputs = model(input_ids=input_ids, labels=labels)
            >>> loss = outputs.loss
            >>> logits = outputs.logits

            >>> input_ids = tokenizer("summarize: studies have shown that owning a dog is good for you ", return_tensors="pt").input_ids  # Batch size 1
            >>> outputs = model.generate(input_ids)
        """
        use_cache = use_cache if use_cache is not None else self.config.use_cache
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        # Encode if needed (training, first prediction pass)
        if encoder_outputs is None:
            # Convert encoder inputs in embeddings if needed
            encoder_outputs = self.encoder(
                input_ids=input_ids,
                attention_mask=attention_mask,
                inputs_embeds=inputs_embeds,
                head_mask=head_mask,
                output_attentions=output_attentions,
                output_normalized_attentions=output_normalized_attentions,
                output_head_hiddens=output_head_hiddens,
                output_hidden_states=output_hidden_states,
                return_dict=return_dict,
            )
        elif return_dict and not isinstance(encoder_outputs, BaseModelOutput):
            encoder_outputs = BaseModelOutput(
                last_hidden_state=encoder_outputs[0],
                hidden_states=encoder_outputs[1] if len(encoder_outputs) > 1 else None,
                attentions=encoder_outputs[2] if len(encoder_outputs) > 2 else None,
            )

        hidden_states = encoder_outputs[0]

        if labels is not None and decoder_input_ids is None and decoder_inputs_embeds is None:
            # get decoder inputs from shifting lm labels to the right
            decoder_input_ids = self._shift_right(labels)
            # fixed: decoder_attention_mask should also be shifted if labels are shifted
            if decoder_attention_mask is not None:
                decoder_attention_mask = self._shift_right(decoder_attention_mask, insert_token_id=1)

        # If decoding with past key value states, only the last tokens
        # should be given as an input
        if past_key_values is not None:
            assert labels is None, "Decoder should not use cached key value states when training."
            if decoder_input_ids is not None:
                decoder_input_ids = decoder_input_ids[:, -1:]
            if decoder_inputs_embeds is not None:
                decoder_inputs_embeds = decoder_inputs_embeds[:, -1:]

        # Decode
        decoder_outputs = self.decoder(
            input_ids=decoder_input_ids,
            attention_mask=decoder_attention_mask,
            inputs_embeds=decoder_inputs_embeds,
            past_key_values=past_key_values,
            encoder_hidden_states=hidden_states,
            encoder_attention_mask=attention_mask,
            head_mask=head_mask,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_normalized_attentions=output_normalized_attentions,
            output_head_hiddens=output_head_hiddens,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )

        sequence_output = decoder_outputs[0]

        if self.config.tie_word_embeddings:
            # 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)

        loss = None
        if labels is not None:
            loss_fct = CrossEntropyLoss(ignore_index=-100)
            loss = loss_fct(lm_logits.view(-1, lm_logits.size(-1)), labels.view(-1))
            # TODO(thom): Add z_loss https://github.com/tensorflow/mesh/blob/fa19d69eafc9a482aff0b59ddd96b025c0cb207d/mesh_tensorflow/layers.py#L666

        if not return_dict:
            output = (lm_logits,) + decoder_outputs[1:] + encoder_outputs
            return ((loss,) + output) if loss is not None else output

        return Seq2SeqLMOutput(
            loss=loss,
            logits=lm_logits,
            past_key_values=decoder_outputs.past_key_values,
            decoder_hidden_states=decoder_outputs.hidden_states,
            decoder_attentions=decoder_outputs.attentions,
            cross_attentions=decoder_outputs.cross_attentions,
            encoder_last_hidden_state=encoder_outputs.last_hidden_state,
            encoder_hidden_states=encoder_outputs.hidden_states,
            encoder_attentions=encoder_outputs.attentions,
        )

    def prepare_inputs_for_generation(
        self, input_ids, past=None, attention_mask=None, use_cache=None, encoder_outputs=None, **kwargs
    ):

        # cut decoder_input_ids if past is used
        if past is not None:
            input_ids = input_ids[:, -1:]

        return {
            "decoder_input_ids": input_ids,
            "past_key_values": past,
            "encoder_outputs": encoder_outputs,
            "attention_mask": attention_mask,
            "use_cache": use_cache,
        }

    def _reorder_cache(self, past, beam_idx):
        # if decoder past is not included in output
        # speedy decoding is disabled and no need to reorder
        if past is None:
            logger.warning("You might want to consider setting `use_cache=True` to speed up decoding")
            return past

        reordered_decoder_past = ()
        for layer_past_states in past:
            # get the correct batch idx from layer past batch dim
            # batch dim of `past` is at 2nd position
            reordered_layer_past_states = ()
            for layer_past_state in layer_past_states:
                # need to set correct `past` for each of the four key / value states
                reordered_layer_past_states = reordered_layer_past_states + (
                    layer_past_state.index_select(0, beam_idx),
                )

            assert reordered_layer_past_states[0].shape == layer_past_states[0].shape
            assert len(reordered_layer_past_states) == len(layer_past_states)

            reordered_decoder_past = reordered_decoder_past + (reordered_layer_past_states,)
        return reordered_decoder_past


class T5CDQAttention(nn.Module):
    def __init__(self, config: T5Config, has_relative_attention_bias=False):
        super().__init__()
        self.is_decoder = config.is_decoder
        if self.is_decoder:
            raise RuntimeError('T5CDQAttention is not implemented for decoder cause casual masking!')
        self.has_relative_attention_bias = has_relative_attention_bias

        self.relative_attention_num_buckets = config.relative_attention_num_buckets
        self.d_model = config.d_model
        self.key_value_proj_dim = config.d_kv
        # half of the attention heads are regular attention heads
        # another half of the attention heads uses context-dependent Q
        self.n_heads = config.num_heads - config.num_cdq_heads
        self.cdq_n_heads = config.num_cdq_heads
        # we share W_key, W_values, W_outputs matrices for regular and cdq attention
        self.n_shared_heads = self.n_heads if self.n_heads == self.cdq_n_heads else self.cdq_n_heads
        self.dropout = config.dropout_rate
        self.inner_dim = self.n_shared_heads * self.key_value_proj_dim

        # Mesh TensorFlow initialization to avoid scaling before softmax
        if self.n_heads > 0:
            self.q = nn.Linear(self.d_model, self.inner_dim, bias=False)
        # cd_* are shared for all cd heads
        self.cd_q = nn.Linear(self.d_model, self.key_value_proj_dim, bias=False)
        # todo: make less parameters in cd_w
        self.cd_w = nn.Linear(self.d_model * 2, self.d_model, bias=False)

        # n_heads vectors T = [t1; t2, ...] that are used to compute attention over H to produce P = [p1; p2; ...]
        self.T = nn.Linear(self.d_model, self.cdq_n_heads, bias=False)
        self.W_t = nn.Linear(self.d_model, self.d_model, bias=False)

        # k, v are shared between MHA and CDQ-MHA -> 1 head from MHA and 1 head from CDQ-MHA will have the same K and V
        # todo: maybe not to share?
        # todo: make k,v also context-dependent as q?
        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 is applied to concatenation of regular MHA and CDQ-MHA
        self.o = nn.Linear((self.n_heads + self.cdq_n_heads) * self.key_value_proj_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_shared_heads)
        self.pruned_heads = set()

    def prune_heads(self, heads):
        if len(heads) == 0:
            return
        heads, index = find_pruneable_heads_and_indices(
            heads, self.n_heads, self.key_value_proj_dim, self.pruned_heads
        )
        # 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.key_value_proj_dim * 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)
        """
        relative_buckets = 0
        if bidirectional:
            num_buckets //= 2
            relative_buckets += (relative_position > 0).to(torch.long) * num_buckets
            relative_position = torch.abs(relative_position)
        else:
            relative_position = -torch.min(relative_position, torch.zeros_like(relative_position))
        # now relative_position is in the range [0, inf)

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

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

        relative_buckets += torch.where(is_small, relative_position, relative_postion_if_large)
        return relative_buckets

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

    def forward(
        self,
        hidden_states,
        mask=None,
        key_value_states=None,
        position_bias=None,
        past_key_value=None,
        head_mask=None,
        query_length=None,
        use_cache=False,
        output_attentions=False,
        output_normalized_attentions=False,
        output_head_hiddens=False,
    ):
        """
        Self-attention (if key_value_states is None) or attention over source sentence (provided by key_value_states).
        """
        # Input is (batch_size, seq_length, dim)
        # Mask is (batch_size, key_length) (non-causal) or (batch_size, key_length, key_length)
        # past_key_value[0] is (batch_size, n_heads, q_len - 1, dim_per_head)
        batch_size, seq_length = hidden_states.shape[:2]

        real_seq_length = seq_length

        if past_key_value is not None:
            assert (
                len(past_key_value) == 2
            ), "past_key_value should have 2 past states: keys and values. Got {} past states".format(
                len(past_key_value)
            )
            real_seq_length += past_key_value[0].shape[2] if query_length is None else query_length

        key_length = real_seq_length if key_value_states is None else key_value_states.shape[1]

        def shape(states):
            """  projection """
            return states.view(batch_size, -1, self.n_shared_heads, self.key_value_proj_dim).transpose(1, 2)

        def unshape(states):
            """  reshape """
            return states.transpose(1, 2).contiguous().view(batch_size, -1, self.inner_dim)

        def project(hidden_states, proj_layer, key_value_states, past_key_value):
            """ projects hidden states correctly to key/query states """
            if key_value_states is None:
                # self-attn
                # (batch_size, n_heads, seq_length, dim_per_head)
                hidden_states = shape(proj_layer(hidden_states))
            elif past_key_value is None:
                # cross-attn
                # (batch_size, n_heads, seq_length, dim_per_head)
                hidden_states = shape(proj_layer(key_value_states))

            if past_key_value is not None:
                if key_value_states is None:
                    # self-attn
                    # (batch_size, n_heads, key_length, dim_per_head)
                    hidden_states = torch.cat([past_key_value, hidden_states], dim=2)
                else:
                    # cross-attn
                    hidden_states = past_key_value
            return hidden_states

        # T = [t1,t2,t3,..,t_{n_heads}], T x relu(W_t x H) = T_att_scores # (bs, n_heads, seq_len)
        # T x relu(W_t x H) aka multiplicative attention with non-linearity
        # todo: try without relu, or with relu(T)
        T_att_scores = self.T(F.relu(self.W_t(hidden_states))).transpose(1, 2)
        if mask is not None:
            if self.is_decoder:
                # wow, we are in decoder now
                print('Hey, Decoder!')
                # if it is not SA ignore mask
                # SA mask has shape (bs, 1, dec_len, dec_len)
                # CA mask has shape (bs, 1, 1, enc_len)
                if mask.shape[-1] == T_att_scores.shape[-1]:
                    ...
                    print('wow, how to use u, mask?')
            else:
                # mask: (bs, 1, 1, seq_len)
                T_att_scores += mask.view(batch_size, 1, seq_length)

        T_attn_weights = F.softmax(T_att_scores.float(), dim=-1).type_as(T_att_scores)
        T_attn_weights = F.dropout(T_attn_weights, p=self.dropout, training=self.training)
        P = torch.matmul(T_attn_weights, hidden_states)  # (bs, n_heads, model_dim)
        # PH = (bs, n_heads, seq_len, d_model x 2)
        PH = torch.cat([
                P.view(batch_size, self.cdq_n_heads, 1, self.d_model).repeat(1, 1, seq_length, 1),
                hidden_states.view(batch_size, 1, -1, self.d_model).repeat(1, self.cdq_n_heads, 1, 1)
            ], dim=-1)
        # input to cd_q is from [-1;1] while input to q is not restricted to these values
        # cd_query_states might have diffent scale compared to query_states -> try cd_q with bias to restore scale
        cd_query_states = self.cd_q(self.cd_w(PH))  # (bs, n_heads, seq_len, dim_per_head)

        # get query states
        if self.n_heads > 0:
            query_states = shape(self.q(hidden_states))  # (batch_size, n_heads, seq_length, dim_per_head)
        # get key/value states
        key_states = project(
            hidden_states, self.k, key_value_states, past_key_value[0] if past_key_value is not None else None
        )
        value_states = project(
            hidden_states, self.v, key_value_states, past_key_value[1] if past_key_value is not None else None
        )

        # compute scores
        if self.n_heads > 0:
            scores = torch.matmul(
                query_states, key_states.transpose(3, 2)
            )  # equivalent of torch.einsum("bnqd,bnkd->bnqk", query_states, key_states), compatible with onnx op>9
        cd_scores = torch.matmul(cd_query_states, key_states.transpose(3, 2))

        if position_bias is None:
            if not self.has_relative_attention_bias:
                position_bias = torch.zeros(
                    (1, self.n_heads, real_seq_length, key_length), device=scores.device, dtype=scores.dtype
                )
            else:
                position_bias = self.compute_bias(real_seq_length, key_length)

            # if key and values are already calculated
            # we want only the last query position bias
            if past_key_value is not None:
                position_bias = position_bias[:, :, -seq_length:, :]

            if mask is not None:
                position_bias = position_bias + mask  # (batch_size, n_heads, seq_length, key_length)

        if self.n_heads > 0:
            scores += position_bias
            attn_weights = F.softmax(scores.float(), dim=-1).type_as(scores)
            attn_weights = F.dropout(attn_weights, p=self.dropout, training=self.training)
        else:
            attn_weights = None

        cd_scores += position_bias  # todo: not to use position bias? idk, makes sense to use it..
        # attn_weights: (batch_size, n_heads, seq_length, key_length)
        cd_attn_weights = F.softmax(cd_scores.float(), dim=-1).type_as(cd_scores)
        cd_attn_weights = F.dropout(cd_attn_weights, p=self.dropout, training=self.training)

        # Mask heads if we want to
        if head_mask is not None and self.n_heads > 0:
            # head_mask only for regular MHA
            attn_weights = attn_weights * head_mask

        cd_attn_output = unshape(torch.matmul(cd_attn_weights, value_states))  # (batch_size, seq_length, dim)
        if self.n_heads > 0:
            attn_output = unshape(torch.matmul(attn_weights, value_states))  # (batch_size, seq_length, dim)
            attn_output = self.o(torch.cat([attn_output, cd_attn_output], dim=-1))
        else:
            attn_output = self.o(cd_attn_output)

        present_key_value_state = (key_states, value_states) if (self.is_decoder and use_cache) else None
        outputs = (attn_output,) + (present_key_value_state,) + (position_bias,)

        if output_attentions:
            if output_normalized_attentions:
                v_norms = torch.linalg.norm(value_states, dim=-1).unsqueeze(2)
                attn_weights = attn_weights * v_norms
                cd_attn_weights = cd_attn_weights * v_norms
                # todo: T_attnt_weights normalize by values too?
            outputs = outputs + ((attn_weights, cd_attn_weights, T_attn_weights), )
        return outputs