train_lora.py

import logging
import math
import os
import random
import sys
from itertools import chain

import datasets
import torch
import transformers
from config import DataArguments, ModelArguments, TrainingArguments
from datasets import load_dataset
from dialogues import get_dialogue_template, mask_user_labels, prepare_dialogue
from transformers import (AutoModelForCausalLM, AutoTokenizer, Trainer,
                          default_data_collator, set_seed)
from transformers.testing_utils import CaptureLogger
from transformers.trainer_utils import get_last_checkpoint
from utils import StarChatArgumentParser, hf_login
from transformers import DataCollatorWithPadding
from peft import LoraConfig, get_peft_model, TaskType

logger = logging.getLogger(__name__)


def main():
    logging.basicConfig(
        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
        datefmt="%Y-%m-%d %H:%M:%S",
        handlers=[logging.StreamHandler(sys.stdout)],
        level=logging.DEBUG
    )

    parser = StarChatArgumentParser((ModelArguments, DataArguments, TrainingArguments))
    if len(sys.argv) == 2 and sys.argv[1].endswith(".yaml"):
        # If we pass only one argument to the script and it's the path to a YAML file,
        # let's parse it to get our arguments.
        model_args, data_args, training_args = parser.parse_yaml_file(os.path.abspath(sys.argv[1]))
    # parse command line args and yaml file
    elif len(sys.argv) > 2 and sys.argv[1].endswith(".yaml"):
        model_args, data_args, training_args = parser.parse_yaml_and_args(os.path.abspath(sys.argv[1]), sys.argv[2:])
    # parse command line args only
    else:
        model_args, data_args, training_args = parser.parse_args_into_dataclasses()
        
    # Set custom training arguments

    # Set seed for reproducibility
    set_seed(training_args.seed)

    ###############
    # Setup logging
    ###############
    logging.basicConfig(
        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
        datefmt="%Y-%m-%d %H:%M:%S",
        handlers=[logging.StreamHandler(sys.stdout)],
    )
    log_level = training_args.get_process_log_level()
    logger.setLevel(log_level)
    datasets.utils.logging.set_verbosity(log_level)
    transformers.utils.logging.set_verbosity(log_level)
    transformers.utils.logging.enable_default_handler()
    transformers.utils.logging.enable_explicit_format()

    # Log on each process a small summary
    logger.warning(
        f"Process rank: {training_args.local_rank}, device: {training_args.device}, n_gpu: {training_args.n_gpu}"
        + f" distributed training: {bool(training_args.local_rank != -1)}, 16-bits training: {training_args.fp16}"
    )
    logger.info(f"Model parameters {model_args}")
    logger.info(f"Data parameters {data_args}")
    logger.info(f"Training/evaluation parameters {training_args}")

    # Login to HuggingFace Hub if needed
    hf_login()

    ###########################
    # Detecting last checkpoint
    ###########################
    last_checkpoint = None
    if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
        last_checkpoint = get_last_checkpoint(training_args.output_dir)
        if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
            raise ValueError(
                f"Output directory ({training_args.output_dir}) already exists and is not empty. "
                "Use --overwrite_output_dir to overcome."
            )
        elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
            logger.info(
                f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
                "the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
            )

    ###############
    # Load datasets
    ###############
    raw_datasets = load_dataset('json', data_files={'train': './gcc_test_dataset.json', 'test': './gcc_test_dataset.json'})
    logger.info(
        f"Training on the following datasets and their proportions: {[split + ' : ' + str(dset.num_rows) for split, dset in raw_datasets.items()]}"
    )
    with training_args.main_process_first(desc="Log a few random samples from the raw training set"):
        for index in random.sample(range(len(raw_datasets["train"])), 3):
            logger.info(f"Sample {index} of the raw training set:\n\n{raw_datasets['train'][index]['messages']}")

    #########################
    # Apply dialogue template
    #########################
    dialogue_template = get_dialogue_template(data_args.dialogue_template)
    logger.info(f"System prompt for dialogue template: {dialogue_template.system}")
    raw_datasets = raw_datasets.map(prepare_dialogue, fn_kwargs={"dialogue_template": dialogue_template})

    #####################################
    # Load tokenizer and process datasets
    #####################################
    tokenizer = AutoTokenizer.from_pretrained(
        model_args.model_name_or_path,
        revision=model_args.model_revision,
    )
    tokenizer.add_special_tokens({'pad_token': '[PAD]'})

    # Note that we must call `add_tokens` before adding any special tokens
    dialogue_tokens = dialogue_template.get_special_tokens()
    num_added_tokens = tokenizer.add_special_tokens({"additional_special_tokens": dialogue_tokens})
    logger.info(f"Added {num_added_tokens} new tokens: {dialogue_tokens}")

    if training_args.do_train:
        column_names = list(raw_datasets["train"].features)
    else:
        column_names = list(raw_datasets["test"].features)
    text_column_name = "text" if "text" in column_names else column_names[0]

    with training_args.main_process_first(desc="Log a few random samples from the training set"):
        for index in random.sample(range(len(raw_datasets["train"])), 3):
            logger.info(f"Sample {index} of the raw training set:\n\n{raw_datasets['train'][index]['text']}")

    # since this will be pickled to avoid _LazyModule error in Hasher force logger loading before tokenize_function
    tok_logger = transformers.utils.logging.get_logger("transformers.tokenization_utils_base")

    def tokenize_function(examples):
        with CaptureLogger(tok_logger) as cl:
            output = tokenizer(
                examples[text_column_name], 
                return_token_type_ids=False,
                truncation=True,  # 添加此行以避免过长的序列
                max_length=256,
                padding="max_length"
            )
        # clm input could be much much longer than block_size
        if "Token indices sequence length is longer than the" in cl.out:
            tok_logger.warning(
                "^^^^^^^^^^^^^^^^ Please ignore the warning above - this long input will be chunked into smaller bits"
                " before being passed to the model."
            )
        return output

    with training_args.main_process_first(desc="dataset map tokenization"):
        tokenized_datasets = raw_datasets.map(
            tokenize_function,
            batched=True,
            num_proc=data_args.preprocessing_num_workers,
            remove_columns=column_names,
            load_from_cache_file=not data_args.overwrite_cache,
            desc="Running tokenizer on dataset",
        )

    ##############################
    # Concatenate and chunk corpus
    ##############################
    if data_args.block_size is None:
        block_size = tokenizer.model_max_length
        if block_size > 1024:
            logger.warning(
                "The chosen tokenizer supports a `model_max_length` that is longer than the default `block_size` value"
                " of 1024. If you would like to use a longer `block_size` up to `tokenizer.model_max_length` you can"
                " override this default with `--block_size xxx`."
            )
            block_size = 1024
    else:
        if data_args.block_size > tokenizer.model_max_length:
            logger.warning(
                f"The block_size passed ({data_args.block_size}) is larger than the maximum length for the model"
                f"({tokenizer.model_max_length}). Using block_size={tokenizer.model_max_length}."
            )
        block_size = min(data_args.block_size, tokenizer.model_max_length)

    # Main data processing function that will concatenate all texts from our dataset and generate chunks of block_size.
    def group_texts(examples):
        # 添加调试打印语句，检查 examples 的结构
        print(f"Process {os.getpid()}: Examples structure: {examples}")

        concatenated_examples = {k: list(chain(*examples[k])) for k in examples.keys()}
        total_length = len(concatenated_examples[list(examples.keys())[0]])
        if total_length >= block_size:
            total_length = (total_length // block_size) * block_size

        result = {
            k: [t[i: i + block_size] for i in range(0, total_length, block_size)]
            for k, t in concatenated_examples.items()
        }
        labels = result["input_ids"].copy()
        mask_user_labels(tokenizer, dialogue_template, labels)
        result["labels"] = labels

        # 填充并确保结果为 PyTorch 张量
        result = tokenizer.pad(result, padding=True, return_tensors="pt")
        return result

    # Note that with `batched=True`, this map processes 1,000 texts together, so group_texts throws away a remainder
    # for each of those groups of 1,000 texts. You can adjust that batch_size here but a higher value might be slower
    # to preprocess.
    with training_args.main_process_first(desc="grouping texts together"):
        lm_datasets = tokenized_datasets.map(
            group_texts,
            batched=True,
            num_proc=data_args.preprocessing_num_workers,
            load_from_cache_file=not data_args.overwrite_cache,
            desc=f"Grouping texts in chunks of {block_size}",
        )

    if training_args.do_train:
        if "train" not in tokenized_datasets:
            raise ValueError("--do_train requires a train dataset")
        train_dataset = lm_datasets["train"]
        if data_args.max_train_samples is not None:
            max_train_samples = min(len(train_dataset), data_args.max_train_samples)
            train_dataset = train_dataset.select(range(max_train_samples))

    if training_args.do_eval:
        if "test" not in tokenized_datasets:
            raise ValueError("--do_eval requires a validation dataset")
        eval_dataset = lm_datasets["test"]
        if data_args.max_eval_samples is not None:
            max_eval_samples = min(len(eval_dataset), data_args.max_eval_samples)
            eval_dataset = eval_dataset.select(range(max_eval_samples))

    #######################
    # Load pretrained model
    #######################
    logger.info("*** Load pretrained model ***")
    torch_dtype = (
        model_args.torch_dtype if model_args.torch_dtype in ["auto", None] else getattr(torch, model_args.torch_dtype)
    )
    model = AutoModelForCausalLM.from_pretrained(
        model_args.model_name_or_path,
        revision=model_args.model_revision,
        torch_dtype=torch_dtype,
        use_cache=False if training_args.gradient_checkpointing else True,
    )
    model.enable_input_require_grads()
    model.resize_token_embeddings(len(tokenizer))

    ####################
    # Apply LoRA Adapter
    ####################
    lora_config = LoraConfig(
        r=8,  # Rank of the low-rank matrices
        lora_alpha=32,
        target_modules=["q_proj","v_proj"],
        lora_dropout=0.1,
        bias="none",
        task_type=TaskType.CAUSAL_LM
    )
    model = get_peft_model(model, lora_config)

    ########################
    # Initialize the Trainer
    ########################
    trainer = Trainer(
        model=model,
        args=training_args,
        train_dataset=train_dataset if training_args.do_train else None,
        eval_dataset=eval_dataset if training_args.do_eval else None,
        tokenizer=tokenizer,
        data_collator=DataCollatorWithPadding(tokenizer)
    )

    ###############
    # Training loop
    ###############
    if training_args.do_train:
        logger.info("*** Train ***")
        checkpoint = None
        if training_args.resume_from_checkpoint is not None:
            checkpoint = training_args.resume_from_checkpoint
        elif last_checkpoint is not None:
            checkpoint = last_checkpoint
        train_result = trainer.train(resume_from_checkpoint=checkpoint)

        metrics = train_result.metrics

        max_train_samples = (
            data_args.max_train_samples if data_args.max_train_samples is not None else len(train_dataset)
        )
        metrics["train_samples"] = min(max_train_samples, len(train_dataset))

        trainer.log_metrics("train", metrics)
        trainer.save_metrics("train", metrics)
        trainer.save_state()

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

        metrics = trainer.evaluate()

        max_eval_samples = data_args.max_eval_samples if data_args.max_eval_samples is not None else len(eval_dataset)
        metrics["eval_samples"] = min(max_eval_samples, len(eval_dataset))
        try:
            perplexity = math.exp(metrics["eval_loss"])
        except OverflowError:
            perplexity = float("inf")
        metrics["perplexity"] = perplexity

        trainer.log_metrics("eval", metrics)
        trainer.save_metrics("eval", metrics)

    #################################
    # Create model card & push to Hub
    #################################
    kwargs = {"finetuned_from": model_args.model_name_or_path, "tasks": "text-generation"}
    if data_args.dataset_name is not None:
        kwargs["dataset_tags"] = data_args.dataset_name
        if hasattr(data_args, 'dataset_config_name') and data_args.dataset_config_name is not None:
            kwargs["dataset_args"] = data_args.dataset_config_name
            kwargs["dataset"] = f"{data_args.dataset_name} {data_args.dataset_config_name}"
        else:
            kwargs["dataset"] = data_args.dataset_name
            kwargs["dataset_args"] = "default"

    # Store dialogue template so we can load it at deployment time
    dialogue_template.save_pretrained(training_args.output_dir)

    if training_args.push_to_hub:
        trainer.push_to_hub(**kwargs)
    else:
        trainer.save_model(training_args.output_dir)
        trainer.create_model_card(**kwargs)

    with training_args.main_process_first(desc="Generate a sample from the model"):
        inputs = tokenizer(
            "<|system|>\n<|end|>\n<|user|>\nHow many helicopters can a human eat in one sitting?<|end|>\n<|assistant|>",
            return_tensors="pt",
            return_token_type_ids=False,
        ).to(training_args.device)
        outputs = model.generate(
            **inputs,
            max_new_tokens=256,
            pad_token_id=tokenizer.eos_token_id,
            eos_token_id=tokenizer.convert_tokens_to_ids(dialogue_template.end_token),
        )
        logger.info(f"=== SAMPLE OUTPUT ==\n\n{tokenizer.decode(outputs[0], skip_special_tokens=False)}")


if __name__ == "__main__":
    main()