pretrain.py

import os
import platform
import argparse
import time
import math
import warnings
import torch
import torch.distributed as dist
from torch import optim
from torch.nn.parallel import DistributedDataParallel
from torch.optim.lr_scheduler import CosineAnnealingLR
from torch.utils.data import DataLoader, DistributedSampler
from contextlib import nullcontext
from models.model_llama import Transformer   ## 从 /models/ 这个文件夹中的 model_llama.py 中，导入 Transfomrer 这个 class
from models.LMConfig import LMConfig         ## 从 /models/ 这个文件夹中的 LMconfig.py 中，导入 LMConfig 这个 class
from models.dataset import PretrainDataset   ## 从 /models/ 这个文件夹中的 dataset.py 中，导入 PretrainDataset 这个 class

warnings.filterwarnings('ignore')


def Logger(content):
    if not ddp or dist.get_rank() == 0:
        print(content)


def get_lr(it, all):
    warmup_iters = args.warmup_iters
    lr_decay_iters = all
    min_lr = args.learning_rate / 10

    if it < warmup_iters:
        return args.learning_rate * it / warmup_iters
    if it > lr_decay_iters:
        return min_lr
    decay_ratio = (it - warmup_iters) / (lr_decay_iters - warmup_iters)
    assert 0 <= decay_ratio <= 1
    coeff = 0.5 * (1.0 + math.cos(math.pi * decay_ratio))
    return min_lr + coeff * (args.learning_rate - min_lr)


def train_epoch(epoch, wandb):
    start_time = time.time()
    for step, (X, Y) in enumerate(train_loader):
        X = X.to(args.device)
        Y = Y.to(args.device)

        lr = get_lr(epoch * iter_per_epoch + step, args.epochs * iter_per_epoch)
        for param_group in optimizer.param_groups:
            param_group['lr'] = lr

        with ctx:
            out = model(X, Y)
            loss = out.last_loss / args.accumulation_steps

        scaler.scale(loss).backward()

        if (step + 1) % args.accumulation_steps == 0:
            scaler.unscale_(optimizer)
            torch.nn.utils.clip_grad_norm_(model.parameters(), args.grad_clip)

            scaler.step(optimizer)
            scaler.update()

            optimizer.zero_grad(set_to_none=True)

        if step % args.log_interval == 0:
            spend_time = time.time() - start_time
            Logger(
                'Epoch:[{}/{}]({}/{}) loss:{:.3f} lr:{:.7f} epoch_Time:{}min:'.format(
                    epoch,
                    args.epochs,
                    step,
                    iter_per_epoch,
                    loss.item() * args.accumulation_steps,
                    optimizer.param_groups[-1]['lr'],
                    spend_time / (step + 1) * iter_per_epoch // 60 - spend_time // 60))

            if (wandb is not None) and (not ddp or dist.get_rank() == 0):
                wandb.log({"loss": loss.item() * args.accumulation_steps,
                           "lr": optimizer.param_groups[-1]['lr'],
                           "epoch_Time": spend_time / (step + 1) * iter_per_epoch // 60 - spend_time // 60})

        if (step + 1) % args.save_interval == 0 and (not ddp or dist.get_rank() == 0):
            model.eval()
            moe_path = '_moe' if lm_config.use_moe else ''
            ckp = f'{args.save_dir}/{args.model_name}.pth'  # 使用用户提供的模型名称

            if isinstance(model, torch.nn.parallel.DistributedDataParallel):
                state_dict = model.module.state_dict()
            else:
                state_dict = model.state_dict()

            torch.save(state_dict, ckp)
            Logger(f"保存模型到 {ckp}")
            optimizer_state_path = f'{args.save_dir}/{args.model_name}_optimizer.pth'
            torch.save(optimizer.state_dict(), optimizer_state_path)
            Logger(f"保存优化器状态到 {optimizer_state_path}")
            
            model.train()


def init_model():
    def count_parameters(model):
        return sum(p.numel() for p in model.parameters() if p.requires_grad)

    model = Transformer(lm_config).to(args.device)
    moe_path = '_moe' if lm_config.use_moe else ''
    
    # 加入恢复训练的逻辑
    checkpoint_path = f'{args.save_dir}/{args.model_name}.pth'  # 使用用户提供的模型名称
    if os.path.exists(checkpoint_path):
        Logger(f"加载模型检查点 {checkpoint_path}")
        model.load_state_dict(torch.load(checkpoint_path, map_location=args.device))
    else:
        Logger(f"没有找到模型检查点，开始从头训练")
    
    Logger(f'LLM总参数量：{count_parameters(model) / 1e6:.3f} 百万')
    return model



def init_distributed_mode():
    if not ddp: return
    global ddp_local_rank, DEVICE

    dist.init_process_group(backend="nccl")
    ddp_rank = int(os.environ["RANK"])
    ddp_local_rank = int(os.environ["LOCAL_RANK"])
    ddp_world_size = int(os.environ["WORLD_SIZE"])
    DEVICE = f"cuda:{ddp_local_rank}"
    torch.cuda.set_device(DEVICE)


# torchrun --nproc_per_node 2 pretrain.py
if __name__ == "__main__":
    parser = argparse.ArgumentParser(description="Dylan_LLM Pretraining")
    parser.add_argument("--out_dir", type=str, default="out", help="Output directory")
    parser.add_argument("--epochs", type=int, default=20, help="Number of epochs")
    parser.add_argument("--batch_size", type=int, default=32, help="Batch size")
    parser.add_argument("--learning_rate", type=float, default=2e-4, help="Learning rate")
    parser.add_argument("--device", type=str, default="cuda:0" if torch.cuda.is_available() else "cpu", help="Device to use")
    parser.add_argument("--dtype", type=str, default="bfloat16", help="Data type")
    parser.add_argument("--use_wandb", action="store_true", help="Use Weights & Biases")
    parser.add_argument("--wandb_project", type=str, default="Dylan-LLM-Pretrain", help="Weights & Biases project name")
    parser.add_argument("--num_workers", type=int, default=8, help="Number of workers for data loading")
    parser.add_argument("--data_path", type=str, default="./data/processed/pretrain_data.bin", help="Path to training data")
    parser.add_argument("--ddp", action="store_true", help="Use DistributedDataParallel")
    parser.add_argument("--accumulation_steps", type=int, default=8, help="Gradient accumulation steps")
    parser.add_argument("--grad_clip", type=float, default=1.0, help="Gradient clipping threshold")
    parser.add_argument("--warmup_iters", type=int, default=0, help="Number of warmup iterations")
    parser.add_argument("--log_interval", type=int, default=10, help="Logging interval")
    parser.add_argument("--save_interval", type=int, default=20, help="Model saving interval")
    parser.add_argument("--model_name", type=str, default="pretrain_512", help="模型名称，用于保存和加载检查点")

    parser.add_argument("--local_rank", type=int, default=-1, help="Local rank for distributed training")

    args = parser.parse_args()

    lm_config = LMConfig()
    max_seq_len = lm_config.max_seq_len
    args.save_dir = os.path.join(args.out_dir)
    os.makedirs(args.save_dir, exist_ok=True)
    os.makedirs(args.out_dir, exist_ok=True)
    checkpoint_path = f'{args.save_dir}/{args.model_name}.pth'
    
    tokens_per_iter = args.batch_size * max_seq_len
    torch.manual_seed(1337)
    device_type = "cuda" if "cuda" in args.device else "cpu"

    args.wandb_run_name = f"Dylan-LLM-Pretrain-Epoch-{args.epochs}-BatchSize-{args.batch_size}-LearningRate-{args.learning_rate}"

    ctx = nullcontext() if device_type == "cpu" else torch.cuda.amp.autocast()

    ddp = int(os.environ.get("RANK", -1)) != -1  # is this a ddp run?
    ddp_local_rank, DEVICE = 0, "cuda:0"
    if ddp:
        init_distributed_mode()
        args.device = torch.device(DEVICE)

    if args.use_wandb and (not ddp or ddp_local_rank == 0):
        import wandb
        wandb.init(project=args.wandb_project, name=args.wandb_run_name)
    else:
        wandb = None

    data_path_list = [args.data_path]
    train_ds = PretrainDataset(data_path_list, max_length=max_seq_len, memmap=True)
    train_sampler = DistributedSampler(train_ds) if ddp else None
    train_loader = DataLoader(
        train_ds,
        batch_size=args.batch_size,
        pin_memory=True,
        drop_last=False,
        shuffle=False,
        num_workers=args.num_workers,
        sampler=train_sampler
    )

    model = init_model()

    scaler = torch.cuda.amp.GradScaler(enabled=(args.dtype in ['float16', 'bfloat16']))
    optimizer = optim.Adam(model.parameters(), lr=args.learning_rate)

    # 恢复优化器状态
    optimizer_state_path = f"{args.save_dir}/{args.model_name}_optimizer.pth"
    if os.path.exists(checkpoint_path):
        optimizer_state_path = f"{args.save_dir}/{args.model_name}_optimizer.pth"
        if os.path.exists(optimizer_state_path):
            Logger(f"加载优化器状态 {optimizer_state_path}")
            optimizer.load_state_dict(torch.load(optimizer_state_path, map_location=args.device))
        else:
            Logger(f"没有找到优化器状态，使用新优化器")

    if False and platform.system() != "Windows" and float(torch.__version__.split('.')[0]) >= 2:
        Logger("compiling the model... (takes a ~minute)")
        unoptimized_model = model
        model = torch.compile(model)

    if ddp:
        model._ddp_params_and_buffers_to_ignore = {"pos_cis"}
        model = DistributedDataParallel(model, device_ids = [ddp_local_rank])

    iter_per_epoch = len(train_loader)
    for epoch in range(args.epochs):
        train_epoch(epoch, wandb)