text-generation-inference/server/text_generation_server/models/custom_modeling/t5_modeling.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."""

import copy
import math
import warnings
from typing import Optional, Tuple, Union

from loguru import logger

import torch
import torch.distributed
from torch import nn
from torch.nn import CrossEntropyLoss

from transformers.activations import ACT2FN
from transformers.modeling_outputs import (
    BaseModelOutput,
    BaseModelOutputWithPastAndCrossAttentions,
    Seq2SeqLMOutput,
)
from transformers.modeling_utils import PreTrainedModel
from transformers.pytorch_utils import ALL_LAYERNORM_LAYERS
from transformers.utils import (
    is_torch_fx_proxy,
)
from transformers import T5Config
from text_generation_server.utils.layers import (
    TensorParallelColumnLinear,
    TensorParallelEmbedding,
    TensorParallelRowLinear,
    TensorParallelHead,
)


class PartialTPEmbedding(nn.Module):
    def __init__(self, prefix: str, weights):
        super().__init__()
        weight = weights.get_sharded(f"{prefix}.weight", dim=1)
        self.weight = nn.Parameter(weight)

    def forward(self, input: torch.Tensor) -> torch.Tensor:
        return torch.nn.functional.embedding(input, self.weight)


@torch.jit.script
def layer_norm(hidden_states, weight, epsilon):
    # T5 uses a layer_norm which only scales and doesn't shift, which is also known as Root Mean
    # Square Layer Normalization https://arxiv.org/abs/1910.07467 thus varience is calculated
    # w/o mean and there is no bias. Additionally we want to make sure that the accumulation for
    # half-precision inputs is done in fp32

    variance = hidden_states.to(torch.float32).pow(2).mean(-1, keepdim=True)
    hidden_states = hidden_states * torch.rsqrt(variance + epsilon)

    # convert into half-precision if necessary
    if weight.dtype in [torch.float16, torch.bfloat16]:
        hidden_states = hidden_states.to(weight.dtype)

    return weight * hidden_states


class T5LayerNorm(nn.Module):
    def __init__(self, prefix, weights, eps=1e-6):
        """
        Construct a layernorm module in the T5 style. No bias and no subtraction of mean.
        """
        super().__init__()
        weight = weights.get_tensor(f"{prefix}.weight")
        self.weight = nn.Parameter(weight)
        self.variance_epsilon = torch.tensor(eps)

    def forward(self, hidden_states):
        return layer_norm(hidden_states, self.weight, self.variance_epsilon)


try:
    from apex.normalization import FusedRMSNorm

    T5LayerNorm = FusedRMSNorm  # noqa

    logger.info(
        "Discovered apex.normalization.FusedRMSNorm - will use it instead of T5LayerNorm"
    )
except ImportError:
    # using the normal T5LayerNorm
    pass
except Exception:
    logger.warning("discovered apex but it failed to load, falling back to T5LayerNorm")
    pass

ALL_LAYERNORM_LAYERS.append(T5LayerNorm)


class T5DenseActDense(nn.Module):
    def __init__(self, config: T5Config, prefix, weights):
        super().__init__()
        self.wi = TensorParallelColumnLinear.load(
            config, prefix=f"{prefix}.wi", weights=weights, bias=False
        )

        ### XXX: T5 models do not handle well both f16 and quantization.
        ### Overidding specifically this layer for that reason.
        ### https://github.com/huggingface/transformers/blob/main/src/transformers/models/t5/modeling_t5.py#L316
        ### https://github.com/huggingface/transformers/issues/20287
        _q = config.quantize
        _dtype = weights.dtype
        weights.dtype = torch.float32
        config.quantize = None
        self.wo_cast = (torch.float32, _dtype)
        self.wo = TensorParallelRowLinear.load(
            config, prefix=f"{prefix}.wo", weights=weights, bias=False
        )
        weights.dtype = _dtype
        config.quantize = _q

        self.dropout = nn.Dropout(config.dropout_rate)
        self.act = (
            ACT2FN[config.dense_act_fn]
            if "gelu" not in config.dense_act_fn
            else lambda x: torch.nn.functional.gelu(x, approximate="tanh")
        )

    def forward(self, hidden_states):
        hidden_states = self.wi(hidden_states)
        hidden_states = self.act(hidden_states)
        hidden_states = self.dropout(hidden_states)

        hidden_states = hidden_states.to(dtype=self.wo_cast[0])
        hidden_states = self.wo(hidden_states)
        # XXX: Recasting is already done within the layer norm.
        # Casting back to float16 here modifies results
        # hidden_states = hidden_states.to(dtype=self.wo_cast[1])
        return hidden_states


class T5DenseGatedActDense(nn.Module):
    def __init__(self, config: T5Config, prefix, weights):
        super().__init__()
        self.wi_0 = TensorParallelColumnLinear.load(
            config, prefix=f"{prefix}.wi_0", weights=weights, bias=False
        )
        self.wi_1 = TensorParallelColumnLinear.load(
            config, prefix=f"{prefix}.wi_1", weights=weights, bias=False
        )
        ### XXX: T5 models do not handle well both f16 and quantization.
        ### Overidding specifically this layer for that reason.
        ### https://github.com/huggingface/transformers/blob/main/src/transformers/models/t5/modeling_t5.py#L316
        ### https://github.com/huggingface/transformers/issues/20287
        _q = config.quantize
        _dtype = weights.dtype
        weights.dtype = torch.float32
        config.quantize = None
        self.wo_cast = (torch.float32, _dtype)
        self.wo = TensorParallelRowLinear.load(
            config, prefix=f"{prefix}.wo", weights=weights, bias=False
        )
        weights.dtype = _dtype
        config.quantize = _q

        self.dropout = nn.Dropout(config.dropout_rate)
        self.act = (
            ACT2FN[config.dense_act_fn]
            if "gelu" not in config.dense_act_fn
            else lambda x: torch.nn.functional.gelu(x, approximate="tanh")
        )

    def forward(self, hidden_states):
        hidden_gelu = self.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 = hidden_states.to(dtype=self.wo_cast[0])
        hidden_states = self.wo(hidden_states)
        # XXX: Recasting is already done within the layer norm.
        # Casting back to float16 here modifies results
        # hidden_states = hidden_states.to(dtype=self.wo_cast[1])
        return hidden_states


class T5LayerFF(nn.Module):
    def __init__(self, config: T5Config, prefix, weights):
        super().__init__()
        if config.is_gated_act:
            self.DenseReluDense = T5DenseGatedActDense(
                config, prefix=f"{prefix}.DenseReluDense", weights=weights
            )
        else:
            self.DenseReluDense = T5DenseActDense(
                config, prefix=f"{prefix}.DenseReluDense", weights=weights
            )

        self.layer_norm = T5LayerNorm(
            prefix=f"{prefix}.layer_norm",
            weights=weights,
            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, prefix, weights, 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.relative_attention_max_distance = config.relative_attention_max_distance
        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

        process_group = weights.process_group
        # Mesh TensorFlow initialization to avoid scaling before softmax
        assert self.n_heads % process_group.size() == 0
        self.q = TensorParallelColumnLinear.load(
            config, prefix=f"{prefix}.q", weights=weights, bias=False
        )
        self.k = TensorParallelColumnLinear.load(
            config, prefix=f"{prefix}.k", weights=weights, bias=False
        )
        self.v = TensorParallelColumnLinear.load(
            config, prefix=f"{prefix}.v", weights=weights, bias=False
        )
        self.o = TensorParallelRowLinear.load(
            config, prefix=f"{prefix}.o", weights=weights, bias=False
        )
        if self.n_heads % weights.process_group.size() != 0:
            raise ValueError(
                f"`n_heads` must be divisible by `num_shards` (got `n_heads`: {self.n_heads} "
                f"and `num_shards`: {weights.process_group.size()}"
            )
        self.n_heads = self.n_heads // process_group.size()
        self.inner_dim = self.inner_dim // process_group.size()

        if self.has_relative_attention_bias:
            self.relative_attention_bias = PartialTPEmbedding(
                prefix=f"{prefix}.relative_attention_bias", weights=weights
            )

    @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_position_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_position_if_large = torch.min(
            relative_position_if_large,
            torch.full_like(relative_position_if_large, num_buckets - 1),
        )

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

    def compute_bias(self, query_length, key_length, device=None):
        """Compute binned relative position bias"""
        if device is None:
            device = self.relative_attention_bias.weight.device
        context_position = torch.arange(query_length, dtype=torch.long, device=device)[
            :, None
        ]
        memory_position = torch.arange(key_length, dtype=torch.long, device=device)[
            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,
            max_distance=self.relative_attention_max_distance,
        )
        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,
        layer_head_mask=None,
        query_length=None,
        use_cache=False,
        output_attentions=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
            ), f"past_key_value should have 2 past states: keys and values. Got {len(past_key_value)} past states"
            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)
                elif past_key_value.shape[2] != key_value_states.shape[1]:
                    # checking that the `sequence_length` of the `past_key_value` is the same as
                    # the provided `key_value_states` to support prefix tuning
                    # cross-attn
                    # (batch_size, n_heads, seq_length, dim_per_head)
                    hidden_states = shape(proj_layer(key_value_states))
                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, device=scores.device
                )

            # 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[:, :, -hidden_states.size(1) :, :]

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

        position_bias_masked = position_bias

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

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

        attn_output = unshape(
            torch.matmul(attn_weights, value_states)
        )  # (batch_size, seq_length, dim)
        attn_output = self.o(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:
            outputs = outputs + (attn_weights,)
        return outputs


class T5LayerSelfAttention(nn.Module):
    def __init__(self, config, prefix, weights, has_relative_attention_bias=False):
        super().__init__()
        self.SelfAttention = T5Attention(
            config,
            prefix=f"{prefix}.SelfAttention",
            weights=weights,
            has_relative_attention_bias=has_relative_attention_bias,
        )
        self.layer_norm = T5LayerNorm(
            prefix=f"{prefix}.layer_norm",
            weights=weights,
            eps=config.layer_norm_epsilon,
        )
        self.dropout = nn.Dropout(config.dropout_rate)

    def forward(
        self,
        hidden_states,
        attention_mask=None,
        position_bias=None,
        layer_head_mask=None,
        past_key_value=None,
        use_cache=False,
        output_attentions=False,
    ):
        normed_hidden_states = self.layer_norm(hidden_states)
        attention_output = self.SelfAttention(
            normed_hidden_states,
            mask=attention_mask,
            position_bias=position_bias,
            layer_head_mask=layer_head_mask,
            past_key_value=past_key_value,
            use_cache=use_cache,
            output_attentions=output_attentions,
        )
        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, prefix, weights):
        super().__init__()
        self.EncDecAttention = T5Attention(
            config,
            prefix=f"{prefix}.EncDecAttention",
            weights=weights,
            has_relative_attention_bias=False,
        )
        self.layer_norm = T5LayerNorm(
            prefix=f"{prefix}.layer_norm",
            weights=weights,
            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,
        layer_head_mask=None,
        past_key_value=None,
        use_cache=False,
        query_length=None,
        output_attentions=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,
            layer_head_mask=layer_head_mask,
            past_key_value=past_key_value,
            use_cache=use_cache,
            query_length=query_length,
            output_attentions=output_attentions,
        )
        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, prefix, weights, has_relative_attention_bias: bool):
        super().__init__()
        self.is_decoder = config.is_decoder
        self.layer = nn.ModuleList()
        self.layer.append(
            T5LayerSelfAttention(
                config,
                prefix=f"{prefix}.layer.0",
                weights=weights,
                has_relative_attention_bias=has_relative_attention_bias,
            )
        )
        if self.is_decoder:
            i = 2
            self.layer.append(
                T5LayerCrossAttention(
                    config, prefix=f"{prefix}.layer.1", weights=weights
                )
            )
        else:
            i = 1

        self.layer.append(
            T5LayerFF(config, prefix=f"{prefix}.layer.{i}", weights=weights)
        )

    def forward(
        self,
        hidden_states,
        attention_mask=None,
        position_bias=None,
        encoder_hidden_states=None,
        encoder_attention_mask=None,
        encoder_decoder_position_bias=None,
        layer_head_mask=None,
        cross_attn_layer_head_mask=None,
        past_key_value=None,
        use_cache=False,
        output_attentions=False,
        return_dict=True,
    ):
        if past_key_value is not None:
            if not self.is_decoder:
                logger.warning(
                    "`past_key_values` is passed to the encoder. Please make sure this is intended."
                )
            expected_num_past_key_values = 2 if encoder_hidden_states is None else 4

            if len(past_key_value) != expected_num_past_key_values:
                raise ValueError(
                    f"There should be {expected_num_past_key_values} past states. "
                    f"{'2 (past / key) for cross attention. ' if expected_num_past_key_values == 4 else ''}"
                    f"Got {len(past_key_value)} past key / value states"
                )

            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,
            layer_head_mask=layer_head_mask,
            past_key_value=self_attn_past_key_value,
            use_cache=use_cache,
            output_attentions=output_attentions,
        )
        hidden_states, present_key_value_state = self_attention_outputs[:2]
        attention_outputs = self_attention_outputs[
            2:
        ]  # Keep self-attention outputs and relative position weights

        # clamp inf values to enable fp16 training
        if hidden_states.dtype == torch.float16:
            clamp_value = torch.where(
                torch.isinf(hidden_states).any(),
                torch.finfo(hidden_states.dtype).max - 1000,
                torch.finfo(hidden_states.dtype).max,
            )
            hidden_states = torch.clamp(
                hidden_states, min=-clamp_value, max=clamp_value
            )

        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,
                layer_head_mask=cross_attn_layer_head_mask,
                past_key_value=cross_attn_past_key_value,
                query_length=query_length,
                use_cache=use_cache,
                output_attentions=output_attentions,
            )
            hidden_states = cross_attention_outputs[0]

            # clamp inf values to enable fp16 training
            if hidden_states.dtype == torch.float16:
                clamp_value = torch.where(
                    torch.isinf(hidden_states).any(),
                    torch.finfo(hidden_states.dtype).max - 1000,
                    torch.finfo(hidden_states.dtype).max,
                )
                hidden_states = torch.clamp(
                    hidden_states, min=-clamp_value, max=clamp_value
                )

            # 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)

        # clamp inf values to enable fp16 training
        if hidden_states.dtype == torch.float16:
            clamp_value = torch.where(
                torch.isinf(hidden_states).any(),
                torch.finfo(hidden_states.dtype).max - 1000,
                torch.finfo(hidden_states.dtype).max,
            )
            hidden_states = torch.clamp(
                hidden_states, min=-clamp_value, max=clamp_value
            )

        outputs = (hidden_states,)

        if use_cache:
            outputs = outputs + (present_key_value_state,) + attention_outputs
        else:
            outputs = outputs + attention_outputs

        return outputs  # hidden-states, present_key_value_states, (self-attention position bias), (self-attention weights), (cross-attention position bias), (cross-attention weights)


class T5PreTrainedModel(PreTrainedModel):
    """
    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
    models.
    """

    config_class = T5Config

    def _shift_right(self, input_ids):
        decoder_start_token_id = self.config.decoder_start_token_id
        pad_token_id = self.config.pad_token_id

        assert decoder_start_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
        if is_torch_fx_proxy(input_ids):
            # Item assignment is not supported natively for proxies.
            shifted_input_ids = torch.full(
                input_ids.shape[:-1] + (1,), decoder_start_token_id
            )
            shifted_input_ids = torch.cat(
                [shifted_input_ids, input_ids[..., :-1]], dim=-1
            )
        else:
            shifted_input_ids = input_ids.new_zeros(input_ids.shape)
            shifted_input_ids[..., 1:] = input_ids[..., :-1].clone()
            shifted_input_ids[..., 0] = decoder_start_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)

        return shifted_input_ids


class T5Stack(T5PreTrainedModel):
    def __init__(self, config, prefix, weights, embed_tokens):
        super().__init__(config)

        self.is_decoder = config.is_decoder

        self.embed_tokens = embed_tokens
        self.block = nn.ModuleList(
            [
                T5Block(
                    config,
                    prefix=f"{prefix}.block.{layer_id}",
                    weights=weights,
                    has_relative_attention_bias=(layer_id == 0),
                )
                for layer_id in range(config.num_layers)
            ]
        )
        self.final_layer_norm = T5LayerNorm(
            prefix=f"{prefix}.final_layer_norm",
            weights=weights,
            eps=config.layer_norm_epsilon,
        )
        self.dropout = nn.Dropout(config.dropout_rate)

    def forward(
        self,
        input_ids=None,
        attention_mask=None,
        encoder_hidden_states=None,
        encoder_attention_mask=None,
        inputs_embeds=None,
        head_mask=None,
        cross_attn_head_mask=None,
        past_key_values=None,
        use_cache=None,
        output_attentions=None,
        output_hidden_states=None,
        return_dict=None,
    ):
        # Model parallel
        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}input_ids 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}input_ids 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

        # 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
            ), f"`use_cache` can only be set to `True` if {self} is used as a decoder"

        if attention_mask is None:
            attention_mask = torch.ones(
                batch_size, mask_seq_length, device=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,
            )

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

        # 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.
        extended_attention_mask = self.get_extended_attention_mask(
            attention_mask, input_shape
        )

        # If a 2D or 3D attention mask is provided for the cross-attention
        # we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]
        if self.is_decoder and encoder_hidden_states is not None:
            (
                encoder_batch_size,
                encoder_sequence_length,
                _,
            ) = encoder_hidden_states.size()
            encoder_hidden_shape = (encoder_batch_size, encoder_sequence_length)
            if encoder_attention_mask is None:
                encoder_attention_mask = torch.ones(
                    encoder_hidden_shape, device=inputs_embeds.device
                )
            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)
        cross_attn_head_mask = self.get_head_mask(
            cross_attn_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)
        ):
            layer_head_mask = head_mask[i]
            cross_attn_layer_head_mask = cross_attn_head_mask[i]
            # Model parallel
            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,
                layer_head_mask=layer_head_mask,
                cross_attn_layer_head_mask=cross_attn_layer_head_mask,
                past_key_value=past_key_value,
                use_cache=use_cache,
                output_attentions=output_attentions,
            )

            # layer_outputs is a tuple with:
            # hidden-states, key-value-states, (self-attention position bias), (self-attention weights), (cross-attention position bias), (cross-attention weights)
            if use_cache is False:
                layer_outputs = layer_outputs[:1] + (None,) + layer_outputs[1:]

            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 position bias), (self-attention weights),
            # (cross-attention position bias), (cross-attention weights)
            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,
        )


class T5ForConditionalGeneration(T5PreTrainedModel):
    def __init__(self, config: T5Config, weights):
        super().__init__(config)
        self.model_dim = config.d_model

        self.shared = TensorParallelEmbedding(prefix="shared", weights=weights)

        encoder_config = copy.deepcopy(config)
        encoder_config.is_decoder = False
        encoder_config.use_cache = False
        encoder_config.is_encoder_decoder = False
        self.encoder = T5Stack(
            config=encoder_config,
            prefix="encoder",
            weights=weights,
            embed_tokens=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(
            config=decoder_config,
            prefix="decoder",
            weights=weights,
            embed_tokens=self.shared,
        )

        self.lm_head = TensorParallelHead.load(
            config, prefix="lm_head", weights=weights
        )

    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.FloatTensor] = None,
        decoder_input_ids: Optional[torch.LongTensor] = None,
        decoder_attention_mask: Optional[torch.BoolTensor] = None,
        head_mask: Optional[torch.FloatTensor] = None,
        decoder_head_mask: Optional[torch.FloatTensor] = None,
        cross_attn_head_mask: Optional[torch.Tensor] = None,
        encoder_outputs: Optional[Tuple[Tuple[torch.Tensor]]] = None,
        past_key_values: Optional[Tuple[Tuple[torch.Tensor]]] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        decoder_inputs_embeds: Optional[torch.FloatTensor] = None,
        labels: Optional[torch.LongTensor] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[Tuple[torch.FloatTensor], Seq2SeqLMOutput]:
        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
        )

        # FutureWarning: head_mask was separated into two input args - head_mask, decoder_head_mask
        if head_mask is not None and decoder_head_mask is None:
            if self.config.num_layers == self.config.num_decoder_layers:
                warnings.warn(__HEAD_MASK_WARNING_MSG, FutureWarning)
                decoder_head_mask = head_mask

        # 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_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)

        # 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=decoder_head_mask,
            cross_attn_head_mask=cross_attn_head_mask,
            use_cache=use_cache,
            output_attentions=output_attentions,
            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)
            # move labels to correct device to enable PP
            labels = labels.to(lm_logits.device)
            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_key_values=None,
        attention_mask=None,
        head_mask=None,
        decoder_head_mask=None,
        decoder_attention_mask=None,
        cross_attn_head_mask=None,
        use_cache=None,
        encoder_outputs=None,
        **kwargs,
    ):
        # cut decoder_input_ids if past is used
        if past_key_values is not None:
            input_ids = input_ids[:, -1:]

        return {
            "decoder_input_ids": input_ids,
            "past_key_values": past_key_values,
            "encoder_outputs": encoder_outputs,
            "attention_mask": attention_mask,
            "head_mask": head_mask,
            "decoder_head_mask": decoder_head_mask,
            "decoder_attention_mask": decoder_attention_mask,
            "cross_attn_head_mask": cross_attn_head_mask,
            "use_cache": use_cache,
        }

    def prepare_decoder_input_ids_from_labels(self, labels: torch.Tensor):
        return self._shift_right(labels)

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

        reordered_decoder_past = ()
        for layer_past_states in past_key_values:
            # 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.to(layer_past_state.device)
                    ),
                )

            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