modeling_p5.py

from dataclasses import dataclass

from transformers.models.t5.modeling_t5 import (
    T5Stack, T5Block, T5LayerNorm, T5LayerSelfAttention, T5LayerFF, T5LayerCrossAttention,
    T5PreTrainedModel, T5ForConditionalGeneration
)

import torch
import torch.nn as nn
from torch.nn import CrossEntropyLoss

from typing import Any, Callable, Dict, Iterable, List, Optional, Tuple
import copy

from transformers.modeling_outputs import ModelOutput, BaseModelOutput, BaseModelOutputWithPast, BaseModelOutputWithPastAndCrossAttentions, Seq2SeqLMOutput, Seq2SeqModelOutput
from transformers.modeling_utils import PreTrainedModel, find_pruneable_heads_and_indices, prune_linear_layer
from transformers.utils import logging
from transformers import BeamScorer, BeamSearchScorer

logger = logging.get_logger(__name__)

# The encoder for input token sequence
class JointEncoder(T5Stack):
    def __init__(self, config, embed_tokens=None):
        super(T5Stack, self).__init__(config)
        self.config = config

        self.embed_tokens = embed_tokens
        self.is_decoder = self.config.is_decoder
        assert self.config.is_decoder is False

        self.block = nn.ModuleList(
            [T5Block(config, has_relative_attention_bias=(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)
        
        ## Set maximum 512 whole words in a source text
        self.whole_word_embeddings = nn.Embedding(
            512, config.d_model   ## config.d_model is 768 for base
        )
        self.init_weights()
        self.model_parallel = False
        self.device_map = None

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

    def forward(
        self,
        input_ids=None,
        whole_word_ids=None,
        attention_mask=None,
        inputs_embeds=None,
        head_mask=None,
        past_key_values=None,
        use_cache=None,
        output_attentions=None,
        output_hidden_states=None,
        return_dict=None,
    ):

        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) ### embedding step - add HERE ###
            if whole_word_ids is not None:
                whole_word_embeds = self.whole_word_embeddings(whole_word_ids)
                assert whole_word_embeds.shape[-1] == inputs_embeds.shape[-1]
                inputs_embeds = inputs_embeds + whole_word_embeds

        B, L = inputs_embeds.size()[:-1]

        if attention_mask is None:
            attention_mask = input_ids.ne(self.config.pad_token_id).to(dtype=inputs_embeds.dtype, device=inputs_embeds.device)

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

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

        # 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

        hidden_states = self.dropout(inputs_embeds)

        if self.config.num_layers > 0:

            assert self.block[0].layer[0].SelfAttention.has_relative_attention_bias

            seq_length = L
            q_len = seq_length
            k_len = seq_length

            # [1, n_heads, Q_len, K_len]
            text_position_bias = self.block[0].layer[0].SelfAttention.compute_bias(
                L, L)
            num_heads = text_position_bias.size(1)
            position_bias = text_position_bias.new_zeros(
                1, num_heads, seq_length, seq_length)
            position_bias[:, :, :L, :L] = text_position_bias

            position_bias = position_bias + extended_attention_mask

            for i, (layer_module, past_key_value) in enumerate(zip(self.block, past_key_values)):
                layer_head_mask = head_mask[i]
                layer_outputs = layer_module(
                    hidden_states,
                    attention_mask=extended_attention_mask,
                    position_bias=position_bias,
                    encoder_hidden_states=None,
                    encoder_attention_mask=None,
                    encoder_decoder_position_bias=None,
                    # head_mask=head_mask[i],
                    layer_head_mask=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 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]

                # append next layer key value states
                if use_cache:
                    present_key_value_states = present_key_value_states + \
                        (present_key_value_state,)

        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 P5(T5ForConditionalGeneration):
    _keys_to_ignore_on_load_missing = [
        r"encoder\.embed_tokens\.weight",
        r"decoder\.embed_tokens\.weight",
        r"lm_head\.weight",
    ]
    _keys_to_ignore_on_load_unexpected = [
        r"decoder\.block\.0\.layer\.1\.EncDecAttention\.relative_attention_bias\.weight",
    ]

    def __init__(self, config):
        super(T5ForConditionalGeneration, self).__init__(config)

        self.config = config

        self.model_dim = config.d_model

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

        encoder_config = copy.deepcopy(config)
        encoder_config.is_decoder = False
        encoder_config.use_cache = False
        encoder_config.is_encoder_decoder = False

        self.encoder = JointEncoder(encoder_config, self.shared)

        decoder_config = copy.deepcopy(config)
        decoder_config.is_decoder = True
        decoder_config.is_encoder_decoder = False

        self.decoder = T5Stack(decoder_config, self.shared)

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

        self.init_weights()

        self.model_parallel = False
        self.device_map = None

    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 extend_vocab(self, vocab_size):

        new_shared = nn.Embedding(vocab_size, self.config.d_model)
        old_weight = self.shared.weight.data.detach().clone()
        old_vocab_size = old_weight.size(0)
        new_shared.weight.data[:old_vocab_size, :] = old_weight
        self.shared = new_shared

        new_lm_head = nn.Linear(self.config.d_model, vocab_size, bias=False)
        old_weight = self.lm_head.weight.data.detach().clone()
        old_vocab_size = old_weight.size(0)
        new_lm_head.weight.data[:old_vocab_size, :] = old_weight
        self.lm_head = new_lm_head

        self.encoder.embed_tokens = self.shared
        self.decoder.embed_tokens = self.shared

        self.lm_head.weight = self.shared.weight

        self.config.vocab_size = vocab_size
        self.encoder.config.vocab_size = vocab_size
        self.decoder.config.vocab_size = vocab_size

    def forward(
        self,
        input_ids=None,
        whole_word_ids=None,
        attention_mask=None,
        encoder_outputs=None,
        decoder_input_ids=None,
        decoder_attention_mask=None,
        past_key_values=None,
        use_cache=None,
        labels=None,
        inputs_embeds=None,
        decoder_inputs_embeds=None,
        head_mask=None,
        output_attentions=None,
        output_hidden_states=None,
        return_dict=None,
        reduce_loss=False,

        return_hidden_state=False,

        **kwargs,
    ):

        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

        if encoder_outputs is None:
            encoder_outputs = self.encoder(
                input_ids=input_ids,
                whole_word_ids=whole_word_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)

        # 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:]

        if attention_mask is None:
            attention_mask = input_ids.ne(self.config.pad_token_id).to(dtype=hidden_states.dtype, device=hidden_states.device)
        encoder_attention_mask = attention_mask

        # 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=encoder_attention_mask,

            head_mask=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]

        assert self.config.tie_word_embeddings is True

        if self.config.tie_word_embeddings:
            sequence_output = sequence_output * (self.model_dim ** -0.5)

        if return_hidden_state:
            return sequence_output

        lm_logits = self.lm_head(sequence_output)

        loss = None
        if labels is not None:
            if reduce_loss:
                loss_fct = CrossEntropyLoss(ignore_index=-100)
            else:
                loss_fct = CrossEntropyLoss(ignore_index=-100, reduction='none')
            loss = loss_fct(
                lm_logits.view(-1, lm_logits.size(-1)),
                labels.view(-1))

        return P5Seq2SeqLMOutput(
            loss=loss,
            logits=lm_logits,
            past_key_values=decoder_outputs.past_key_values,
            decoder_last_hidden_state=decoder_outputs.last_hidden_state,
            decoder_hidden_states=decoder_outputs.hidden_states,
        )

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

        if past is not None:
            input_ids = input_ids[:, -1:]

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

        return output

    @staticmethod
    def _expand_inputs_for_generation(
        input_ids: torch.LongTensor,
        expand_size: int = 1,
        is_encoder_decoder: bool = False,
        attention_mask: torch.LongTensor = None,
        encoder_outputs: ModelOutput = None,
        **model_kwargs
    ) -> Tuple[torch.LongTensor, Dict[str, Any]]:
        expanded_return_idx = (
            torch.arange(input_ids.shape[0]).view(-1, 1).repeat(1,
                                                                expand_size).view(-1).to(input_ids.device)
        )
        input_ids = input_ids.index_select(0, expanded_return_idx)

        if "token_type_ids" in model_kwargs:
            token_type_ids = model_kwargs["token_type_ids"]
            model_kwargs["token_type_ids"] = token_type_ids.index_select(
                0, expanded_return_idx)

        if attention_mask is not None:
            model_kwargs["attention_mask"] = attention_mask.index_select(
                0, expanded_return_idx)

        if is_encoder_decoder:
            assert encoder_outputs is not None
            encoder_outputs["last_hidden_state"] = encoder_outputs.last_hidden_state.index_select(
                0, expanded_return_idx
            )
            model_kwargs["encoder_outputs"] = encoder_outputs

        return input_ids, model_kwargs


@dataclass
class P5Seq2SeqLMOutput(ModelOutput):
    """
    Base class for sequence-to-sequence language models outputs.

    Args:
        loss (:obj:`torch.FloatTensor` of shape :obj:`(1,)`, `optional`, returned when :obj:`labels` is provided):
            Languaged modeling loss.
        logits (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, config.vocab_size)`):
            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
        past_key_values (:obj:`List[torch.FloatTensor]`, `optional`, returned when ``use_cache=True`` is passed or when ``config.use_cache=True``):
            List of :obj:`torch.FloatTensor` of length :obj:`config.n_layers`,  with each tensor of shape
            :obj:`(2, batch_size, num_heads, sequence_length, embed_size_per_head)`).

            Contains pre-computed hidden-states (key and values in the attention blocks) of the decoder that can be
            used (see ``past_key_values`` input) to speed up sequential decoding.
        decoder_hidden_states (:obj:`tuple(torch.FloatTensor)`, `optional`, returned when ``output_hidden_states=True`` is passed or when ``config.output_hidden_states=True``):
            Tuple of :obj:`torch.FloatTensor` (one for the output of the embeddings + one for the output of each layer)
            of shape :obj:`(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the decoder at the output of each layer plus the initial embedding outputs.
        decoder_attentions (:obj:`tuple(torch.FloatTensor)`, `optional`, returned when ``output_attentions=True`` is passed or when ``config.output_attentions=True``):
            Tuple of :obj:`torch.FloatTensor` (one for each layer) of shape
            :obj:`(batch_size, num_heads, sequence_length, sequence_length)`.

            Attentions weights of the decoder, after the attention softmax, used to compute the weighted average in the
            self-attention heads.
        encoder_last_hidden_state (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):
            Sequence of hidden-states at the output of the last layer of the encoder of the model.
        encoder_hidden_states (:obj:`tuple(torch.FloatTensor)`, `optional`, returned when ``output_hidden_states=True`` is passed or when ``config.output_hidden_states=True``):
            Tuple of :obj:`torch.FloatTensor` (one for the output of the embeddings + one for the output of each layer)
            of shape :obj:`(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the encoder at the output of each layer plus the initial embedding outputs.
        encoder_attentions (:obj:`tuple(torch.FloatTensor)`, `optional`, returned when ``output_attentions=True`` is passed or when ``config.output_attentions=True``):
            Tuple of :obj:`torch.FloatTensor` (one for each layer) of shape
            :obj:`(batch_size, num_heads, sequence_length, sequence_length)`.

            Attentions weights of the encoder, after the attention softmax, used to compute the weighted average in the
            self-attention heads.
    """

    loss: Optional[torch.FloatTensor] = None
    logits: torch.FloatTensor = None
    past_key_values: Optional[List[torch.FloatTensor]] = None
    decoder_last_hidden_state: Optional[Tuple[torch.FloatTensor]] = None
    decoder_hidden_states: Optional[Tuple[torch.FloatTensor]] = None
    decoder_attentions: Optional[Tuple[torch.FloatTensor]] = None
    encoder_last_hidden_state: Optional[torch.FloatTensor] = None
    encoder_hidden_states: Optional[Tuple[torch.FloatTensor]] = None
    encoder_attentions: Optional[Tuple[torch.FloatTensor]] = None