run_kbqg.py

#!/usr/bin/env python
# coding=utf-8
# Copyright The HuggingFace Team and The HuggingFace Inc. team. All rights reserved.
#
# 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.
"""
Fine-tuning a 🤗 Transformers model on summarization.
"""
# You can also adapt this script on your own summarization task. Pointers for this are left as comments.

import argparse
import logging
import math
import os
import random

import nltk
import numpy as np
import torch

from torch.utils.data import DataLoader
from tqdm.auto import tqdm

import transformers
from accelerate import Accelerator

from transformers import (
    AdamW,
    BartConfig,
    BartForConditionalGeneration,
    BartTokenizer,
    SchedulerType,
    get_scheduler,
    set_seed,
)

from dataset import KBQGDataset

from nltk.translate.bleu_score import corpus_bleu


logger = logging.getLogger(__name__)


def parse_args():
    parser = argparse.ArgumentParser(description="Finetune a transformers model on a summarization task")
    parser.add_argument(
        "--data_dir",
        type=str,
        default=None,
        help="Directory of the dataset.",
    )
    parser.add_argument(
        "--ignore_pad_token_for_loss",
        type=bool,
        default=True,
        help="Whether to ignore the tokens corresponding to " "padded labels in the loss computation or not.",
    )
    parser.add_argument(
        "--max_source_length",
        type=int,
        default=256,
        help="The maximum total input sequence length after "
        "tokenization.Sequences longer than this will be truncated, sequences shorter will be padded.",
    )
    parser.add_argument(
        "--preprocessing_num_workers",
        type=int,
        default=None,
        help="The number of processes to use for the preprocessing.",
    )
    parser.add_argument(
        "--max_target_length",
        type=int,
        default=128,
        help="The maximum total sequence length for target text after "
        "tokenization. Sequences longer than this will be truncated, sequences shorter will be padded."
        "during ``evaluate`` and ``predict``.",
    )
    parser.add_argument(
        "--num_beams",
        type=int,
        default=None,
        help="Number of beams to use for evaluation. This argument will be "
        "passed to ``model.generate``, which is used during ``evaluate`` and ``predict``.",
    )
    parser.add_argument(
        "--pad_to_max_length",
        action="store_true",
        help="If passed, pad all samples to `max_length`. Otherwise, dynamic padding is used.",
    )
    parser.add_argument(
        "--model_name_or_path",
        type=str,
        help="Path to pretrained model or model identifier from huggingface.co/models.",
        required=True,
    )
    parser.add_argument(
        "--per_device_train_batch_size",
        type=int,
        default=8,
        help="Batch size (per device) for the training dataloader.",
    )
    parser.add_argument(
        "--per_device_eval_batch_size",
        type=int,
        default=8,
        help="Batch size (per device) for the evaluation dataloader.",
    )
    parser.add_argument(
        "--learning_rate",
        type=float,
        default=5e-5,
        help="Initial learning rate (after the potential warmup period) to use.",
    )
    parser.add_argument("--weight_decay", type=float, default=0.0, help="Weight decay to use.")
    parser.add_argument("--num_train_epochs", type=int, default=3, help="Total number of training epochs to perform.")
    parser.add_argument(
        "--max_train_steps",
        type=int,
        default=None,
        help="Total number of training steps to perform. If provided, overrides num_train_epochs.",
    )
    parser.add_argument(
        "--gradient_accumulation_steps",
        type=int,
        default=1,
        help="Number of updates steps to accumulate before performing a backward/update pass.",
    )
    parser.add_argument(
        "--lr_scheduler_type",
        type=SchedulerType,
        default="linear",
        help="The scheduler type to use.",
        choices=["linear", "cosine", "cosine_with_restarts", "polynomial", "constant", "constant_with_warmup"],
    )
    parser.add_argument(
        "--num_warmup_steps", type=int, default=0, help="Number of steps for the warmup in the lr scheduler."
    )
    parser.add_argument("--output_dir", type=str, default=None, help="Where to store the final model.")
    parser.add_argument("--seed", type=int, default=None, help="A seed for reproducible training.")

    args = parser.parse_args()

    return args


def main():
    args = parse_args()

    # Initialize the accelerator. We will let the accelerator handle device placement for us in this example.
    accelerator = Accelerator()
    # Make one log on every process with the configuration for debugging.
    logging.basicConfig(
        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
        datefmt="%m/%d/%Y %H:%M:%S",
        level=logging.INFO,
    )
    logger.info(accelerator.state)

    # Setup logging, we only want one process per machine to log things on the screen.
    # accelerator.is_local_main_process is only True for one process per machine.
    logger.setLevel(logging.INFO if accelerator.is_local_main_process else logging.ERROR)

    # If passed along, set the training seed now.
    if args.seed is not None:
        set_seed(args.seed)
    
    # Load pretrained model and tokenizer
    model = BartForConditionalGeneration.from_pretrained(args.model_name_or_path)
    tokenizer = BartTokenizer.from_pretrained(args.model_name_or_path)

    # Load dataset
    train_dataset = KBQGDataset(tokenizer, args.data_dir, 'train', args.max_source_length, args.max_target_length)
    eval_dataset = KBQGDataset(tokenizer, args.data_dir, 'val', args.max_source_length, args.max_target_length)


    model.resize_token_embeddings(len(tokenizer))
    if model.config.decoder_start_token_id is None:
        raise ValueError("Make sure that `config.decoder_start_token_id` is correctly defined")

    # Temporarily set max_target_length for training.
    max_target_length = args.max_target_length
    padding = "max_length" if args.pad_to_max_length else False

    # Log a few random samples from the training set:
    for index in random.sample(range(len(train_dataset)), 1):
        logger.info(f"Sample {index} of the training set: {train_dataset[index]}.")

    label_pad_token_id = -100 if args.ignore_pad_token_for_loss else tokenizer.pad_token_id

    def postprocess_text(preds, labels):
        preds = [pred.strip() for pred in preds]
        labels = [label.strip() for label in labels]

        # rougeLSum expects newline after each sentence
        preds = ["\n".join(nltk.sent_tokenize(pred)) for pred in preds]
        labels = ["\n".join(nltk.sent_tokenize(label)) for label in labels]

        return preds, labels

    train_dataloader = DataLoader(
        train_dataset, shuffle=True, batch_size=args.per_device_train_batch_size
    )
    eval_dataloader = DataLoader(eval_dataset, batch_size=args.per_device_eval_batch_size)

    # Optimizer
    # Split weights in two groups, one with weight decay and the other not.
    no_decay = ["bias", "LayerNorm.weight"]
    optimizer_grouped_parameters = [
        {
            "params": [p for n, p in model.named_parameters() if not any(nd in n for nd in no_decay)],
            "weight_decay": args.weight_decay,
        },
        {
            "params": [p for n, p in model.named_parameters() if any(nd in n for nd in no_decay)],
            "weight_decay": 0.0,
        },
    ]
    optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate)

    # Prepare everything with our `accelerator`.
    model, optimizer, train_dataloader, eval_dataloader = accelerator.prepare(
        model, optimizer, train_dataloader, eval_dataloader
    )

    # Note -> the training dataloader needs to be prepared before we grab his length below (cause its length will be
    # shorter in multiprocess)

    # Scheduler and math around the number of training steps.
    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
    if args.max_train_steps is None:
        args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
    else:
        args.num_train_epochs = math.ceil(args.max_train_steps / num_update_steps_per_epoch)

    lr_scheduler = get_scheduler(
        name=args.lr_scheduler_type,
        optimizer=optimizer,
        num_warmup_steps=args.num_warmup_steps,
        num_training_steps=args.max_train_steps,
    )

    # Train!
    total_batch_size = args.per_device_train_batch_size * accelerator.num_processes * args.gradient_accumulation_steps

    logger.info("***** Running training *****")
    logger.info(f"  Num examples = {len(train_dataset)}")
    logger.info(f"  Num Epochs = {args.num_train_epochs}")
    logger.info(f"  Instantaneous batch size per device = {args.per_device_train_batch_size}")
    logger.info(f"  Total train batch size (w. parallel, distributed & accumulation) = {total_batch_size}")
    logger.info(f"  Gradient Accumulation steps = {args.gradient_accumulation_steps}")
    logger.info(f"  Total optimization steps = {args.max_train_steps}")
    # Only show the progress bar once on each machine.
    progress_bar = tqdm(range(args.max_train_steps), disable=not accelerator.is_local_main_process)
    completed_steps = 0

    for epoch in range(args.num_train_epochs):
        model.train()
        for step, batch in enumerate(train_dataloader):
            outputs = model(
                input_ids=batch["source_ids"],
                attention_mask=batch["source_mask"],
                labels=batch["target_ids"],
            )
            loss = outputs.loss
            loss = loss / args.gradient_accumulation_steps
            accelerator.backward(loss)
            if step % args.gradient_accumulation_steps == 0 or step == len(train_dataloader) - 1:
                optimizer.step()
                lr_scheduler.step()
                optimizer.zero_grad()
                progress_bar.update(1)
                completed_steps += 1

            if completed_steps >= args.max_train_steps:
                break

        model.eval()

        gen_kwargs = {
            "max_length": args.max_target_length if args is not None else config.max_length,
            "num_beams": args.num_beams,
        }
        hypotheses = []
        list_of_references = []
        for step, batch in enumerate(eval_dataloader):
            with torch.no_grad():
                generated_tokens = accelerator.unwrap_model(model).generate(
                    batch["source_ids"],
                    attention_mask=batch["source_mask"],
                    **gen_kwargs,
                )

                generated_tokens = accelerator.pad_across_processes(
                    generated_tokens, dim=1, pad_index=tokenizer.pad_token_id
                )
                labels = batch["target_ids"]
                if not args.pad_to_max_length:
                    # If we did not pad to max length, we need to pad the labels too
                    labels = accelerator.pad_across_processes(batch["target_ids"], dim=1, pad_index=tokenizer.pad_token_id)

                generated_tokens = accelerator.gather(generated_tokens).cpu().numpy()
                labels = accelerator.gather(labels).cpu().numpy()

                if args.ignore_pad_token_for_loss:
                    # Replace -100 in the labels as we can't decode them.
                    labels = np.where(labels != -100, labels, tokenizer.pad_token_id)
                if isinstance(generated_tokens, tuple):
                    generated_tokens = generated_tokens[0]
                decoded_preds = tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)
                decoded_labels = tokenizer.batch_decode(labels, skip_special_tokens=True)

                decoded_preds, decoded_labels = postprocess_text(decoded_preds, decoded_labels)

                hypotheses.extend(decoded_preds)
                list_of_references.extend(decoded_labels)
        
        result = corpus_bleu(
            list_of_references=list_of_references,
            hypotheses=hypotheses,
            weights=[
                (1, 0, 0, 0),
                (0.5, 0.5, 0, 0),
                (0.33, 0.33, 0.33, 0),
                (0.25, 0.25,  0.25, 0.25)
            ]
        )

        logger.info(f"evaluation at epoch {epoch}, result: {result}")

        if epoch < args.num_train_epochs - 1:
            accelerator.wait_for_everyone()
            unwrapped_model = accelerator.unwrap_model(model)
            unwrapped_model.save_pretrained(args.output_dir, save_function=accelerator.save)
            if accelerator.is_main_process:
                tokenizer.save_pretrained(args.output_dir)

    if args.output_dir is not None:
        accelerator.wait_for_everyone()
        unwrapped_model = accelerator.unwrap_model(model)
        unwrapped_model.save_pretrained(args.output_dir, save_function=accelerator.save)
        if accelerator.is_main_process:
            tokenizer.save_pretrained(args.output_dir)


if __name__ == "__main__":
    main()