Lora_SeqCLS.py

import wandb
import sys
import os
from copy import deepcopy

import traceback
import logging

from transformers import AutoTokenizer, AutoConfig, set_seed
from transformers import TrainingArguments, HfArgumentParser, TrainerCallback
from peft import get_peft_model, LoraConfig, TaskType

import torch

from dataset import TrainCollator, TrainDataset, DPTrainDataset, EvalDataset, TrainNCCDataset, EvalNCCDataset, TrainNCCCollator
from arguments import DataArguments, ModelArguments, DenseTrainingArguments
from model import GaLMModel, DPGaLMModel, GaLMModelforNCC
from trainer import get_galm_trainer
from utils import compute_metrics, calculate_ncc_metrics

LABEL_FILE_NAME = "label_name.txt"

class CustomCallback(TrainerCallback):
    def __init__(self, trainer) -> None:
        super().__init__()
        self._trainer = trainer

    def on_epoch_end(self, args, state, control, **kwargs):
        if control.should_evaluate:
            control_copy = deepcopy(control)
            self._trainer.evaluate(eval_dataset=self._trainer.train_dataset, metric_key_prefix="train")
            return control_copy

def get_lora_model(model_checkpoints, model, rank=4, alpha=16, lora_dropout=0.1, bias='none'):
    if model_checkpoints == 'mistralai/Mistral-7B-v0.1' or model_checkpoints == 'meta-llama/Llama-2-7b-hf': 
        peft_config = LoraConfig(
            r=rank, lora_alpha=alpha, lora_dropout=lora_dropout, bias=bias, 
            target_modules=[
                "q_proj",
                "v_proj",
            ],
    )
    else: 
        peft_config = LoraConfig(
            task_type=TaskType.SEQ_CLS, r=rank, lora_alpha=alpha, lora_dropout=lora_dropout, bias=bias,
        )
    model.lm = get_peft_model(model.lm, peft_config)
    print(model.lm.print_trainable_parameters())

    return model

logger = logging.getLogger(__name__)

def main():
    """
    Training function
    """
    parser = HfArgumentParser((ModelArguments, DataArguments, DenseTrainingArguments))

    if len(sys.argv) == 2 and sys.argv[1].endswith(".json"):
        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()
        model_args: ModelArguments
        data_args: DataArguments
        training_args: TrainingArguments

    if (
            os.path.exists(training_args.output_dir)
            and os.listdir(training_args.output_dir)
            and training_args.do_train
            and not training_args.overwrite_output_dir
    ):
        raise ValueError(
            f"Output directory ({training_args.output_dir}) already exists and is not empty. Use --overwrite_output_dir to overcome."
        )

    # Setup logging
    logging.basicConfig(
        format="%(asctime)s - %(levelname)s - %(name)s -   %(message)s",
        datefmt="%m/%d/%Y %H:%M:%S",
        level=logging.INFO if training_args.local_rank in [-1, 0] else logging.WARN,
    )
    logger.warning(
        "Process rank: %s, device: %s, n_gpu: %s, distributed training: %s, 16-bits training: %s",
        training_args.local_rank,
        training_args.device,
        training_args.n_gpu,
        bool(training_args.local_rank != -1),
        training_args.fp16,
    )
    logger.info("Training/evaluation parameters %s", training_args)
    logger.info("MODEL parameters %s", model_args)

    set_seed(training_args.seed)
    training_args.lr_scheduler_type= "cosine"

    config = AutoConfig.from_pretrained(
        model_args.config_name if model_args.config_name else model_args.model_name_or_path,
        num_labels=1,
        cache_dir=model_args.cache_dir,
    )

    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,
        add_prefix_space=True,
        use_fast=False,
    )

    if data_args.set_pad_id:
        tokenizer.pad_token = tokenizer.eos_token

    pt_model = GaLMModelforNCC.build(
        model_args,
        data_args,
        training_args,
        config=config,
        cached_dir=model_args.cache_dir,
    )

    if training_args.use_peft:
        model = get_lora_model(
            model_args.model_name_or_path,
            pt_model,
            rank=training_args.lora_rank,
            alpha=training_args.lora_alpha,
            lora_dropout=training_args.lora_dropout,
            bias=training_args.lora_bias
        )
    else:
        model = pt_model

    if data_args.set_pad_id: 
        model.lm.config.pad_token_id = model.lm.config.eos_token_id

    # move model to GPU device
    if model.lm.device.type != 'cuda':
        model.lm = model.lm.to('cuda')
        model.classifier = model.classifier.to('cuda')

    # fix parameters or not
    if training_args.fix_bert:
        params =  model.lm.bert.parameters() if 'bert' in model_args.model_type else model.lm.parameters()
        for param in params:
            param.requires_grad = False

    train_dataset = TrainNCCDataset(
        tokenizer,
        data_args,
        shuffle_seed=training_args.seed,
        cache_dir=data_args.data_cache_dir or model_args.cache_dir,
    )
    eval_dataset = (
        EvalNCCDataset(
            tokenizer,
            data_args,
            shuffle_seed=training_args.seed,
            cache_dir=data_args.data_cache_dir or model_args.cache_dir,
        )
        if data_args.eval_path is not None
        else None
    )

    test_dataset = (
        EvalNCCDataset(
            tokenizer,
            data_args,
            shuffle_seed=training_args.seed,
            cache_dir=data_args.data_cache_dir or model_args.cache_dir,
            eval=False
        )
        if data_args.eval_path is not None
        else None
    )
    
    galm_trainer = get_galm_trainer(private=False)
    
    trainer = galm_trainer(
        model=model,
        args=training_args,
        train_dataset=train_dataset,
        eval_dataset=eval_dataset,
        data_collator=TrainNCCCollator(
            tokenizer,
            max_len=data_args.max_len,
        ),
        compute_metrics=calculate_ncc_metrics
    )
    trainer.add_callback(CustomCallback(trainer))
    train_dataset.trainer = trainer
    trainer.train()
    metrics = trainer.evaluate(eval_dataset=test_dataset)
    print(metrics)
    trainer.state.save_to_json(os.path.join(training_args.output_dir, 'final_state.json'))


if __name__ == "__main__":
    main()