QuestionAnswering.py

from transformers import LukePreTrainedModel, LukeModel, AutoTokenizer, TrainingArguments, default_data_collator, Trainer, AutoModelForQuestionAnswering
from transformers.modeling_outputs import ModelOutput
from typing import Optional, Tuple, Union

import numpy as np
from tqdm import tqdm
import evaluate
import torch
from dataclasses import dataclass
from datasets import load_dataset, concatenate_datasets
from torch import nn
from torch.nn import CrossEntropyLoss
import collections

PEFT = False
tf32 = True
fp16= True
train = False
test = True
trained_model = "LUKE_squadshift"
train_checkpoint = None

base_tokenizer = "roberta-base"
base_model = "studio-ousia/luke-base"

# base_tokenizer = "xlnet-base-cased"
# base_model = "xlnet-base-cased"

# base_tokenizer = "bert-base-cased"
# base_model = "SpanBERT/spanbert-base-cased"

torch.backends.cuda.matmul.allow_tf32 = tf32
torch.backends.cudnn.allow_tf32 = tf32
device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")

if tf32:
    trained_model += "_tf32"

# https://github.com/huggingface/transformers/blob/v4.34.1/src/transformers/models/luke/modeling_luke.py#L319-L353
# Taken from HF repository, easier to include additional features -- Currently identical to LukeForQuestionAnswering by HF

@dataclass
class LukeQuestionAnsweringModelOutput(ModelOutput):
    """
    Outputs of question answering models.


    Args:
        loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
            Total span extraction loss is the sum of a Cross-Entropy for the start and end positions.
        start_logits (`torch.FloatTensor` of shape `(batch_size, sequence_length)`):
            Span-start scores (before SoftMax).
        end_logits (`torch.FloatTensor` of shape `(batch_size, sequence_length)`):
            Span-end scores (before SoftMax).
        hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `torch.FloatTensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.


            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
        entity_hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `torch.FloatTensor` (one for the output of the embeddings + one for the output of each layer) of
            shape `(batch_size, entity_length, hidden_size)`. Entity hidden-states of the model at the output of each
            layer plus the initial entity embedding outputs.
        attentions (`tuple(torch.FloatTensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `torch.FloatTensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.


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


    loss: Optional[torch.FloatTensor] = None
    start_logits: torch.FloatTensor = None
    end_logits: torch.FloatTensor = None
    hidden_states: Optional[Tuple[torch.FloatTensor]] = None
    entity_hidden_states: Optional[Tuple[torch.FloatTensor]] = None
    attentions: Optional[Tuple[torch.FloatTensor]] = None

class AugmentedLukeForQuestionAnswering(LukePreTrainedModel):
    def __init__(self, config):
        super().__init__(config)

        # This is 2.
        self.num_labels = config.num_labels

        self.luke = LukeModel(config, add_pooling_layer=False)

        '''
        Any improvement to the model are expected here. Additional features, anything...
        '''
        self.qa_outputs = nn.Linear(config.hidden_size, config.num_labels)


        # Initialize weights and apply final processing
        self.post_init()

    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.FloatTensor] = None,
        token_type_ids: Optional[torch.LongTensor] = None,
        position_ids: Optional[torch.FloatTensor] = None,
        entity_ids: Optional[torch.LongTensor] = None,
        entity_attention_mask: Optional[torch.FloatTensor] = None,
        entity_token_type_ids: Optional[torch.LongTensor] = None,
        entity_position_ids: Optional[torch.LongTensor] = None,
        head_mask: Optional[torch.FloatTensor] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        start_positions: Optional[torch.LongTensor] = None,
        end_positions: Optional[torch.LongTensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[Tuple, LukeQuestionAnsweringModelOutput]:
        
        r"""
        start_positions (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
            Labels for position (index) of the start of the labelled span for computing the token classification loss.
            Positions are clamped to the length of the sequence (`sequence_length`). Position outside of the sequence
            are not taken into account for computing the loss.
        end_positions (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
            Labels for position (index) of the end of the labelled span for computing the token classification loss.
            Positions are clamped to the length of the sequence (`sequence_length`). Position outside of the sequence
            are not taken into account for computing the loss.
        """
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict


        outputs = self.luke(
            input_ids=input_ids,
            attention_mask=attention_mask,
            token_type_ids=token_type_ids,
            position_ids=position_ids,
            entity_ids=entity_ids,
            entity_attention_mask=entity_attention_mask,
            entity_token_type_ids=entity_token_type_ids,
            entity_position_ids=entity_position_ids,
            head_mask=head_mask,
            inputs_embeds=inputs_embeds,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=True,
        )


        sequence_output = outputs.last_hidden_state


        logits = self.qa_outputs(sequence_output)
        start_logits, end_logits = logits.split(1, dim=-1)
        start_logits = start_logits.squeeze(-1)
        end_logits = end_logits.squeeze(-1)


        total_loss = None
        if start_positions is not None and end_positions is not None:
            # If we are on multi-GPU, split add a dimension
            if len(start_positions.size()) > 1:
                start_positions = start_positions.squeeze(-1)
            if len(end_positions.size()) > 1:
                end_positions = end_positions.squeeze(-1)
            # sometimes the start/end positions are outside our model inputs, we ignore these terms
            ignored_index = start_logits.size(1)
            start_positions.clamp_(0, ignored_index)
            end_positions.clamp_(0, ignored_index)

            loss_fct = CrossEntropyLoss(ignore_index=ignored_index)
            start_loss = loss_fct(start_logits, start_positions)
            end_loss = loss_fct(end_logits, end_positions)
            total_loss = (start_loss + end_loss) / 2


        if not return_dict:
            return tuple(
                v
                for v in [
                    total_loss,
                    start_logits,
                    end_logits,
                    outputs.hidden_states,
                    outputs.entity_hidden_states,
                    outputs.attentions,
                ]
                if v is not None
            )


        return LukeQuestionAnsweringModelOutput(
            loss=total_loss,
            start_logits=start_logits,
            end_logits=end_logits,
            hidden_states=outputs.hidden_states,
            entity_hidden_states=outputs.entity_hidden_states,
            attentions=outputs.attentions,
        )

# Get data to train model - squadshift is designed as a validation/testing set, so there are multiple answers, take the shortest
def get_squadshifts_training():
    wiki = load_dataset("squadshifts", "new_wiki")["test"]
    nyt = load_dataset("squadshifts", "nyt")["test"]
    reddit = load_dataset("squadshifts", "reddit")["test"]
    raw_dataset = concatenate_datasets([wiki, nyt, reddit])
    updated = raw_dataset.map(validation_to_train)
    return updated

def validation_to_train(example):
    answers = example["answers"]
    answer_text = answers["text"]
    index_min = min(range(len(answer_text)), key=lambda x : len(answer_text.__getitem__(x)))
    answers["text"] = answers["text"][index_min:index_min+1]
    answers["answer_start"] = answers["answer_start"][index_min:index_min+1]
    return example

if __name__ == "__main__":
    # Setting up tokenizer and helper functions
    # Work-around for FastTokenizer - RoBERTa and LUKE share the same subword vocab, and we are not using entities functions of LUKE-tokenizer anyways
    tokenizer = AutoTokenizer.from_pretrained(base_tokenizer)

    # Necessary initialization
    max_length = 500
    stride = 128
    batch_size = 8
    n_best = 20
    max_answer_length = 30
    metric = evaluate.load("squad")
    raw_datasets = load_dataset("squad")

    raw_train = raw_datasets["train"]
    raw_validation = raw_datasets["validation"]

    def compute_metrics(start_logits, end_logits, features, examples):
        example_to_features = collections.defaultdict(list)
        for idx, feature in enumerate(features):
            example_to_features[feature["example_id"]].append(idx)

        predicted_answers = []
        for example in tqdm(examples):
            example_id = example["id"]
            context = example["context"]
            answers = []

            # Loop through all features associated with that example
            for feature_index in example_to_features[example_id]:
                start_logit = start_logits[feature_index]
                end_logit = end_logits[feature_index]
                offsets = features[feature_index]["offset_mapping"]

                start_indexes = np.argsort(start_logit)[-1 : -n_best - 1 : -1].tolist()
                end_indexes = np.argsort(end_logit)[-1 : -n_best - 1 : -1].tolist()
                for start_index in start_indexes:
                    for end_index in end_indexes:
                        # Skip answers that are not fully in the context
                        if offsets[start_index] is None or offsets[end_index] is None:
                            continue
                        # Skip answers with a length that is either < 0 or > max_answer_length
                        if (
                            end_index < start_index
                            or end_index - start_index + 1 > max_answer_length
                        ):
                            continue

                        answer = {
                            "text": context[offsets[start_index][0] : offsets[end_index][1]],
                            "logit_score": start_logit[start_index] + end_logit[end_index],
                        }
                        answers.append(answer)

            # Select the answer with the best score
            if len(answers) > 0:
                best_answer = max(answers, key=lambda x: x["logit_score"])
                predicted_answers.append(
                    {"id": example_id, "prediction_text": best_answer["text"]}
                )
            else:
                predicted_answers.append({"id": example_id, "prediction_text": ""})

        theoretical_answers = [{"id": ex["id"], "answers": ex["answers"]} for ex in examples]
        return metric.compute(predictions=predicted_answers, references=theoretical_answers)

    def preprocess_training_examples(examples):

        questions = [q.strip() for q in examples["question"]]
        inputs = tokenizer(
            questions,
            examples["context"],
            max_length=max_length,
            truncation="only_second",
            stride=stride,
            return_overflowing_tokens=True,
            return_offsets_mapping=True,
            padding="max_length",
        )

        offset_mapping = inputs.pop("offset_mapping")
        sample_map = inputs.pop("overflow_to_sample_mapping")
        answers = examples["answers"]
        start_positions = []
        end_positions = []

        for i, offset in enumerate(offset_mapping):
            sample_idx = sample_map[i]
            answer = answers[sample_idx]
            start_char = answer["answer_start"][0]
            end_char = answer["answer_start"][0] + len(answer["text"][0])
            sequence_ids = inputs.sequence_ids(i)

            # Find the start and end of the context
            idx = 0
            while sequence_ids[idx] != 1:
                idx += 1
            context_start = idx
            while sequence_ids[idx] == 1:
                idx += 1
            context_end = idx - 1

            # If the answer is not fully inside the context, label is (0, 0)
            if offset[context_start][0] > start_char or offset[context_end][1] < end_char:
                start_positions.append(0)
                end_positions.append(0)
            else:
                # Otherwise it's the start and end token positions
                idx = context_start
                while idx <= context_end and offset[idx][0] <= start_char:
                    idx += 1
                start_positions.append(idx - 1)

                idx = context_end
                while idx >= context_start and offset[idx][1] >= end_char:
                    idx -= 1
                end_positions.append(idx + 1)

        inputs["start_positions"] = start_positions
        inputs["end_positions"] = end_positions
        return inputs

    def preprocess_validation_examples(examples):
        questions = [q.strip() for q in examples["question"]]
        inputs = tokenizer(
            questions,
            examples["context"],
            max_length=max_length,
            truncation="only_second",
            stride=stride,
            return_overflowing_tokens=True,
            return_offsets_mapping=True,
            padding="max_length",
        )


        sample_map = inputs.pop("overflow_to_sample_mapping")
        example_ids = []

        for i in range(len(inputs["input_ids"])):
            sample_idx = sample_map[i]
            example_ids.append(examples["id"][sample_idx])

            sequence_ids = inputs.sequence_ids(i)
            offset = inputs["offset_mapping"][i]
            inputs["offset_mapping"][i] = [
                o if sequence_ids[k] == 1 else None for k, o in enumerate(offset)
            ]

        inputs["example_id"] = example_ids
        return inputs

    if train:

        model = AutoModelForQuestionAnswering.from_pretrained(base_model).to(device)

        # For squadshift
        raw_train = get_squadshifts_training()

        train_dataset = raw_train.map(
            preprocess_training_examples,
            batched=True,
            remove_columns=raw_train.column_names,
        )

        validation_dataset = raw_validation.map(
            preprocess_validation_examples,
            batched=True,
            remove_columns=raw_validation.column_names,
        )

        

        # --------------- PEFT -------------------- # One epoch without PEFT took about 2h on my computer with CUDA - performance of PEFT kinda ass though
        if PEFT:
            from peft import get_peft_config, get_peft_model, LoraConfig, TaskType

            # ---- For all linear layers ----
            import re
            pattern = r'\((\w+)\): Linear'
            linear_layers = re.findall(pattern, str(model.modules))
            target_modules = list(set(linear_layers))

            # If using peft, can consider increaisng r for better performance
            peft_config = LoraConfig(
                task_type=TaskType.QUESTION_ANS, inference_mode=False, r=8, lora_alpha=32, lora_dropout=0.1, target_modules=target_modules, bias='all'
            ) 

            model = get_peft_model(model, peft_config)
            model.print_trainable_parameters()

            trained_model += "_PEFT"

        # ------------------------------------------ #

        args = TrainingArguments(
            trained_model,
            evaluation_strategy = "no",
            save_strategy="epoch",
            learning_rate=2e-5,
            per_device_train_batch_size=batch_size,
            per_device_eval_batch_size=batch_size,
            num_train_epochs=3,
            weight_decay=0.01,
            push_to_hub=True,
            fp16=fp16
        )

        trainer = Trainer(
            model,
            args,
            train_dataset=train_dataset,
            eval_dataset=validation_dataset,
            data_collator=default_data_collator,
            tokenizer=tokenizer
        )

        trainer.train(train_checkpoint)

    if test:
        model = AutoModelForQuestionAnswering.from_pretrained(trained_model).to(device)

        interval = len(raw_datasets["validation"]) // 100
        exact_match = 0
        f1 = 0

        with torch.no_grad():
            for i in range(1, 101):
                start = interval * (i - 1)
                end = interval * i
                small_eval_set = raw_datasets["validation"].select(range(start ,end))
                eval_set = small_eval_set.map(
                    preprocess_validation_examples,
                    batched=True,
                    remove_columns=raw_datasets["validation"].column_names
                )
                eval_set_for_model = eval_set.remove_columns(["example_id", "offset_mapping"])
                eval_set_for_model.set_format("torch")
                batch = {k: eval_set_for_model[k].to(device) for k in eval_set_for_model.column_names}
                outputs = model(**batch)
                start_logits = outputs.start_logits.cpu().numpy()
                end_logits = outputs.end_logits.cpu().numpy()
                res = compute_metrics(start_logits, end_logits, eval_set, small_eval_set)
                exact_match += res['exact_match']
                f1 += res["f1"]
            
        print("F1 score: {}".format(f1 / 100))
        print("Exact match: {}".format(exact_match / 100))

# XLNET
# F1 score: 91.54154256653278
# Exact match: 84.86666666666666

# SpanBERT
# F1 score: 92.160285362531
# Exact match: 85.73333333333333

# LUKE SQUADSHIFT (SQUAD then SQUADSHIFT)
# F1 score: 91.27683543983473
# Exact match: 84.96190476190473