cl_finetune_trainer.py

import logging
import os
import sys
import time
from dataclasses import dataclass, field
from typing import Optional

import transformers
from cl_seq2seq_trainer import Seq2SeqTrainerCL
from seq2seq_training_args import Seq2SeqTrainingArguments
from transformers import AutoConfig, AutoModelForSeq2SeqLM, AutoTokenizer, HfArgumentParser, MBartTokenizer, set_seed
from transformers.trainer_utils import EvaluationStrategy, is_main_process
# from transformers.training_args import ParallelMode
from utils import (
    Seq2SeqDataCollator,
    Seq2SeqDataset,
    assert_all_frozen,
    build_compute_metrics_fn,
    check_output_dir,
    freeze_embeds,
    freeze_params,
    lmap,
    save_json,
    use_task_specific_params,
    write_txt_file,
)

logger = logging.getLogger(__name__)


@dataclass
class ModelArguments:
    """
    Arguments pertaining to which model/config/tokenizer we are going to fine-tune from.
    """

    model_name_or_path: str = field(
        metadata={"help": "Path to pretrained model or model identifier from huggingface.co/models"}
    )
    config_name: Optional[str] = field(
        default=None, metadata={"help": "Pretrained config name or path if not the same as model_name"}
    )
    tokenizer_name: Optional[str] = field(
        default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
    )
    cache_dir: Optional[str] = field(
        default=None,
        metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
    )
    alpha: Optional[float] = field(default=0.5, metadata={"help": "weight for final loss function"})
    sentence_n: Optional[int] = field(default=2, metadata={"help": "sentences for augmentation"})
    temperature: Optional[float] = field(default=0.5, metadata={"help": "temperature for cl loss"})

    hidden_state_representation: Optional[str] = field(default='cls', metadata={"help": "representation"})

    freeze_embeds: bool = field(default=False, metadata={"help": "Whether  to freeze the embeddings."})
    freeze_encoder: bool = field(default=False, metadata={"help": "Whether tp freeze the encoder."})
    freeze_decoder: bool = field(default=False, metadata={"help": "Whether tp freeze the decoder."})
    freeze_encoder_layer: int = field(default=-1, metadata={"help": "Freeze the first n layers in the encoder"})
    freeze_decoder_layer: int = field(default=-1, metadata={"help": "Freeze the first n layers in the decoder"})
    eval_metric: str = field(default='loss', metadata={"help": "eval metrics for validation set"})

    continue_trainer: bool = field(default=False, metadata={"help": "flag variable to continue training"})
    continue_trainer_path: str = field(default=None, metadata={"help": "checkpoint path to continue training"})


@dataclass
class DataTrainingArguments:
    """
    Arguments pertaining to what data we are going to input our model for training and eval.
    """

    data_dir: str = field(
        metadata={"help": "The input data dir. Should contain the .tsv files (or other data files) for the task."}
    )
    task: Optional[str] = field(
        default="summarization",
        metadata={"help": "Task name, summarization (or summarization_{dataset} for pegasus) or translation"},
    )
    max_source_length: Optional[int] = field(
        default=1024,
        metadata={
            "help": "The maximum total input sequence length after tokenization. Sequences longer "
                    "than this will be truncated, sequences shorter will be padded."
        },
    )
    max_target_length: Optional[int] = field(
        default=128,
        metadata={
            "help": "The maximum total sequence length for target text after tokenization. Sequences longer "
                    "than this will be truncated, sequences shorter will be padded."
        },
    )
    val_max_target_length: Optional[int] = field(
        default=142,
        metadata={
            "help": "The maximum total sequence length for validation target text after tokenization. Sequences longer "
                    "than this will be truncated, sequences shorter will be padded."
        },
    )
    test_max_target_length: Optional[int] = field(
        default=142,
        metadata={
            "help": "The maximum total sequence length for test target text after tokenization. Sequences longer "
                    "than this will be truncated, sequences shorter will be padded."
        },
    )
    n_train: Optional[int] = field(default=-1, metadata={"help": "# training examples. -1 means use all."})
    n_val: Optional[int] = field(default=-1, metadata={"help": "# validation examples. -1 means use all."})
    n_test: Optional[int] = field(default=-1, metadata={"help": "# test examples. -1 means use all."})
    src_lang: Optional[str] = field(default=None, metadata={"help": "Source language id for translation."})
    tgt_lang: Optional[str] = field(default=None, metadata={"help": "Target language id for translation."})
    eval_beams: Optional[int] = field(default=None, metadata={"help": "# num_beams to use for evaluation."})
    ignore_pad_token_for_loss: bool = field(
        default=True,
        metadata={"help": "If only pad tokens should be ignored. This assumes that `config.pad_token_id` is defined."},
    )


def speed_metrics(split, start_time, num_samples):
    """
    Measure and return speed performance metrics.

    This function requires a time snapshot `start_time` before the operation to be measured starts and this
    function should be run immediately after the operation to be measured has completed.

    Args:
    - split: one of train, val, test
    - start_time: operation start time
    - num_samples: number of samples processed

    """
    runtime = time.time() - start_time
    result = {}

    samples_per_second = 1 / (runtime / num_samples)
    result[f"{split}_samples_per_second"] = round(samples_per_second, 3)
    result[f"{split}_runtime"] = round(runtime, 4)

    result[f"{split}_n_ojbs"] = num_samples
    return result


def handle_metrics(split, metrics, output_dir):
    """
    Log and save metrics

    Args:
    - split: one of train, val, test
    - metrics: metrics dict
    - output_dir: where to save the metrics
    """

    logger.info(f"***** {split} metrics *****")
    for key, value in metrics.items():
        logger.info(f"  {key} = {value}")
    save_json(metrics, os.path.join(output_dir, f"{split}_results.json"))


def main():
    # load the parameters
    parser = HfArgumentParser((ModelArguments, DataTrainingArguments, Seq2SeqTrainingArguments))

    if len(sys.argv) == 2 and sys.argv[1].endswith(".json"):
        # If we pass only one argument to the script and it's the path to a json file,
        # let's parse it to get our arguments.
        model_args, data_args, training_args = parser.parse_json_file(json_file=os.path.abspath(sys.argv[1]))
    else:
        model_args, data_args, training_args = parser.parse_args_into_dataclasses()

    check_output_dir(training_args)

    # Set seed
    set_seed(training_args.seed)

    # Load pretrained model and tokenizer
    #
    # Distributed training:
    # The .from_pretrained methods guarantee that only one local process can concurrently
    # download model & vocab.

    config = AutoConfig.from_pretrained(
        model_args.config_name if model_args.config_name else model_args.model_name_or_path,
        cache_dir=model_args.cache_dir,
    )

    extra_model_params = ("encoder_layerdrop", "decoder_layerdrop", "dropout", "attention_dropout")
    for p in extra_model_params:
        if getattr(training_args, p, None):
            assert hasattr(config, p), f"({config.__class__.__name__}) doesn't have a `{p}` attribute"
            setattr(config, p, getattr(training_args, p))

    tokenizer = AutoTokenizer.from_pretrained(
        model_args.tokenizer_name if model_args.tokenizer_name else model_args.model_name_or_path,
        cache_dir=model_args.cache_dir,
    )

    model = AutoModelForSeq2SeqLM.from_pretrained(
        model_args.model_name_or_path,
        from_tf=".ckpt" in model_args.model_name_or_path,
        config=config,
        cache_dir=model_args.cache_dir,
    )

    # use task specific params
    use_task_specific_params(model, data_args.task)

    # set num_beams for evaluation
    if data_args.eval_beams is None:
        data_args.eval_beams = model.config.num_beams

    # set decoder_start_token_id for MBart
    if model.config.decoder_start_token_id is None and isinstance(tokenizer, MBartTokenizer):
        assert (
                data_args.tgt_lang is not None and data_args.src_lang is not None
        ), "mBart requires --tgt_lang and --src_lang"
        model.config.decoder_start_token_id = tokenizer.lang_code_to_id[data_args.tgt_lang]

    if model_args.freeze_embeds:
        freeze_embeds(model)
        logger.info(f"Freeze the embeddings")

    if model_args.freeze_encoder:
        freeze_params(model.get_encoder())
        assert_all_frozen(model.get_encoder())
        logger.info(f"Freeze the encoder")

    if model_args.freeze_decoder:
        freeze_params(model.get_decoder())
        assert_all_frozen(model.get_decoder())
        logger.info(f"Freeze the decoder")

    # freeze the first N layers in the encoder
    if model_args.freeze_encoder_layer > 0:
        freeze_params(model.get_encoder().layers[:model_args.freeze_encoder_layer])
        assert_all_frozen(model.get_encoder().layers[:model_args.freeze_encoder_layer])
        freeze_params(model.get_encoder().layernorm_embedding)
        assert_all_frozen(model.get_encoder().layernorm_embedding)
        logger.info(f"Freeze the first {model_args.freeze_encoder_layer} layers in the encoder")

    if model_args.freeze_decoder_layer > 0:
        freeze_params(model.get_decoder().layers[:model_args.freeze_encoder_layer])
        assert_all_frozen(model.get_decoder().layers[:model_args.freeze_encoder_layer])
        logger.info(f"Freeze the first {model_args.freeze_decoder_layer} layers in the decoder")

    dataset_class = Seq2SeqDataset

    # Get datasets
    train_dataset = (
        dataset_class(
            tokenizer,
            type_path="train",
            data_dir=data_args.data_dir,
            n_obs=data_args.n_train,
            max_target_length=data_args.max_target_length,
            max_source_length=data_args.max_source_length,
            prefix=model.config.prefix or "",
        )
        if training_args.do_train
        else None
    )
    eval_dataset = (
        dataset_class(
            tokenizer,
            type_path="val",
            data_dir=data_args.data_dir,
            n_obs=data_args.n_val,
            max_target_length=data_args.val_max_target_length,
            max_source_length=data_args.max_source_length,
            prefix=model.config.prefix or "",
        )
        if training_args.do_eval or training_args.evaluation_strategy != EvaluationStrategy.NO
        else None
    )
    test_dataset = (
        dataset_class(
            tokenizer,
            type_path="test",
            data_dir=data_args.data_dir,
            n_obs=data_args.n_test,
            max_target_length=data_args.test_max_target_length,
            max_source_length=data_args.max_source_length,
            prefix=model.config.prefix or "",
        )
        if training_args.do_predict
        else None
    )

    # Initialize our Trainer
    compute_metrics_fn = (
        build_compute_metrics_fn(data_args.task, tokenizer) if training_args.predict_with_generate else None
    )
    trainer = Seq2SeqTrainerCL(
        model=model,
        tokenizer=tokenizer,
        config=config,
        args=training_args,
        train_dataset=train_dataset,
        eval_dataset=eval_dataset,
        data_collator=Seq2SeqDataCollator(tokenizer, data_args, training_args.tpu_num_cores),
        compute_metrics=compute_metrics_fn,
        data_args=data_args,
        alpha=model_args.alpha,
        temperature=model_args.temperature,
        hidden_state_representation=model_args.hidden_state_representation,
        eval_metric=model_args.eval_metric,
    )

    all_metrics = {}
    # Training
    if training_args.do_train:
        logger.info("*** Train ***")

        start_time = time.time()

        if not model_args.continue_trainer:
            trainer.train(
                model_path=model_args.model_name_or_path if os.path.isdir(model_args.model_name_or_path) else None
            )
        else:
            trainer.train(
                model_path=model_args.continue_trainer_path if os.path.isdir(model_args.continue_trainer_path) else None
            )

        metrics = speed_metrics("train", start_time, data_args.n_train)

        trainer.save_model()  # this also saves the tokenizer

        if trainer.is_world_process_zero():
            handle_metrics("train", metrics, training_args.output_dir)
            all_metrics.update(metrics)

            # Need to save the state, since Trainer.save_model saves only the tokenizer with the model
            trainer.state.save_to_json(os.path.join(training_args.output_dir, "trainer_state.json"))

            # For convenience, we also re-save the tokenizer to the same directory,
            # so that you can share your model easily on huggingface.co/models =)
            tokenizer.save_pretrained(training_args.output_dir)

    # Evaluation
    if training_args.do_eval:
        logger.info("*** Evaluate ***")

        start_time = time.time()
        metrics = trainer.evaluate(metric_key_prefix="val")
        metrics.update(speed_metrics("val", start_time, data_args.n_val))
        metrics["val_loss"] = round(metrics["val_loss"], 4)

        if trainer.is_world_process_zero():
            handle_metrics("val", metrics, training_args.output_dir)
            all_metrics.update(metrics)

    if training_args.do_predict:
        logger.info("*** Predict ***")

        start_time = time.time()
        test_output = trainer.predict(test_dataset=test_dataset, metric_key_prefix="test")
        metrics = test_output.metrics
        metrics.update(speed_metrics("test", start_time, data_args.n_test))

        if trainer.is_world_process_zero():
            metrics["test_loss"] = round(metrics["test_loss"], 4)
            handle_metrics("test", metrics, training_args.output_dir)
            all_metrics.update(metrics)

            if training_args.predict_with_generate:
                test_preds = tokenizer.batch_decode(
                    test_output.predictions, skip_special_tokens=True, clean_up_tokenization_spaces=True
                )
                test_preds = lmap(str.strip, test_preds)
                write_txt_file(test_preds, os.path.join(training_args.output_dir, "test_generations.txt"))

    if trainer.is_world_process_zero():
        save_json(all_metrics, os.path.join(training_args.output_dir, "all_results.json"))

    return all_metrics


if __name__ == "__main__":
    main()