pretrain.py

import os
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
import time
import math
import pickle
from contextlib import nullcontext
import numpy as np
import torch
from model import Transformer, ModelArgs
from torch.distributed import destroy_process_group, init_process_group
from torch.nn.parallel import DistributedDataParallel as DDP

from dataset import PretrainDataset
import logging

#To run with DDP on 4 gpus on 1 node, example:
# torchrun --standalone --nproc_per_node=4 pretrain.py OR python -m torch.distributed.launch --nproc_per_node=4 pretrain.py
        
def get_logger(filename, verbosity=1, name=None):
    level_dict = {0: logging.DEBUG, 1: logging.INFO, 2: logging.WARNING}
    formatter = logging.Formatter(
        "[%(asctime)s][%(filename)s][%(levelname)s] %(message)s"
    )
    logger = logging.getLogger(name)
    logger.setLevel(level_dict[verbosity])

    fh = logging.FileHandler(filename, "w")
    fh.setFormatter(formatter)
    logger.addHandler(fh)

    sh = logging.StreamHandler()
    sh.setFormatter(formatter)
    logger.addHandler(sh)
    return logger
# -----------------------------------------------------------------------------
def get_lr(it):
    # 1) linear warmup for warmup_iters steps
    if it < warmup_iters:
        return learning_rate * it / warmup_iters
    # 2) if it > lr_decay_iters, return min learning rate
    if it > lr_decay_iters:
        return min_lr
    # 3) in between, use cosine decay down to min learning rate
    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)) # coeff ranges 0..1
    return min_lr + coeff * (learning_rate - min_lr)

def train_epoch(epoch):
    start_time=time.time()
    for step, (X, Y) in enumerate(train_loader):
        X=X.to(device)
        Y=Y.to(device)
        lr = get_lr(epoch*iter_per_epoch+step) if decay_lr else learning_rate
        for param_group in optimizer.param_groups:
            param_group['lr'] = lr
        # and using the GradScaler if data type is float16
        #for micro_step in range(gradient_accumulation_steps):
        if ddp:
            # in DDP training we only need to sync gradients at the last micro step.
            # the official way to do this is with model.no_sync() context manager, but
            # I really dislike that this bloats the code and forces us to repeat code
            # looking at the source of that context manager, it just toggles this variable
            model.require_backward_grad_sync = 0 == gradient_accumulation_steps - 1
        with ctx:
            logits = model(X, Y)
            loss = raw_model.last_loss
            #loss = loss / gradient_accumulation_steps
        # immediately async prefetch next batch while model is doing the forward pass on the GPU
        # backward pass, with gradient scaling if training in fp16
        scaler.scale(loss).backward()
        #
        # clip the gradient
        if grad_clip != 0.0:
            scaler.unscale_(optimizer)
            torch.nn.utils.clip_grad_norm_(model.parameters(), grad_clip)
        # step the optimizer and scaler if training in fp16
        scaler.step(optimizer)
        scaler.update()
        # flush the gradients as soon as we can, no need for this memory anymore
        optimizer.zero_grad(set_to_none=True)
        #打印日志
        if step % log_interval == 0:
            spend_time=time.time()-start_time
            logger.info(
                    'Epoch:[{}/{}]({}/{}) loss:{:.3f} lr:{:.7f} epoch_Time:{}min:'.format(
                        epoch,
                        max_epoch, 
                        step, 
                        iter_per_epoch,
                        loss.item(), 
                        optimizer.param_groups[-1]['lr'],
                        spend_time / (step+1) * iter_per_epoch // 60 - spend_time // 60))

@torch.no_grad()
def valid_epoch(epoch):
    global best_val_loss
    losses = []
    model.eval()
    for _, (X, Y) in enumerate(val_loader):
        X=X.to(device)
        Y=Y.to(device)
        with ctx:
            logits, loss = model(X, Y)
        losses.append(loss.item())
    model.train()
    val_loss=np.mean(losses)
    #
    logger.info('valid loss = {:.4f}'.format(val_loss))
    if val_loss < best_val_loss:
        best_val_loss = val_loss
        logger.info('best val_loss: {} best_epoch: {} '.format(best_val_loss,epoch))
        torch.save(raw_model.state_dict(),'{}/best.pth'.format(save_dir))
    #
    return val_loss

def init_model():
    # model init
    # model init
    model_args = dict(
        dim=dim,
        n_layers=n_layers,
        n_heads=n_heads,
        n_kv_heads=n_heads,
        vocab_size=64793,
        multiple_of=multiple_of,
        max_seq_len=max_seq_len,
        dropout=dropout,
    )  # start with model_args from command line
    if init_from == "scratch":
        # init a new model from scratch
        print("Initializing a new model from scratch")
        gptconf = ModelArgs(**model_args)
        model = Transformer(gptconf)
    elif init_from == "resume":
        print(f"Resuming training from {out_dir}")
        # resume training from a checkpoint.
        ckpt_path = os.path.join(out_dir, "ckpt.pt")
        checkpoint = torch.load(ckpt_path, map_location=device)
        checkpoint_model_args = checkpoint["model_args"]
        # force these config attributes to be equal otherwise we can't even resume training
        # the rest of the attributes (e.g. dropout) can stay as desired from command line
        for k in ["dim", "n_layers", "n_heads", "n_kv_heads", "vocab_size", "multiple_of", "max_seq_len"]:
            model_args[k] = checkpoint_model_args[k]
        # create the model
        gptconf = ModelArgs(**model_args)
        model = Transformer(gptconf)
        state_dict = checkpoint["model"]
        # fix the keys of the state dictionary :(
        # honestly no idea how checkpoints sometimes get this prefix, have to debug more
        unwanted_prefix = "_orig_mod."
        for k, v in list(state_dict.items()):
            if k.startswith(unwanted_prefix):
                state_dict[k[len(unwanted_prefix) :]] = state_dict.pop(k)
        model.load_state_dict(state_dict)
        iter_num = checkpoint["iter_num"]
        best_val_loss = checkpoint["best_val_loss"]
    return model
# I/O
if __name__=="__main__":
    out_dir = 'out'
    max_epoch = 2
    eval_interval = 1
    log_interval = 100
    eval_iters = 200
    eval_only = False # if True, script exits right after the first eval
    always_save_checkpoint = True # if True, always save a checkpoint after each eval
    init_from = 'scratch' # 'scratch' or 'resume' or 'gpt2*'
    #
    gradient_accumulation_steps = 1 # used to simulate larger batch sizes
    batch_size = 32  # if gradient_accumulation_steps > 1, this is the micro-batch size
    # model 根据需要更改 
    max_seq_len = 512
    dim = 1024
    n_layers = 12
    n_heads = 8
    multiple_of = 32
    dropout = 0.0 # for pretraining 0 is good, for finetuning try 0.1+
    bias = False # do we use bias inside LayerNorm and Linear layers?
    # adamw optimizer
    learning_rate = 3e-4 # max learning rate
    weight_decay = 1e-1
    beta1 = 0.9
    beta2 = 0.95
    grad_clip = 1.0 # clip gradients at this value, or disable if == 0.0
    # learning rate decay settings
    decay_lr = True # whether to decay the learning rate
    warmup_iters = 1000 # how many steps to warm up for
    lr_decay_iters = 80000 # should be ~= max_iters per Chinchilla
    min_lr = 1e-5 # minimum learning rate, should be ~= learning_rate/10 per Chinchilla
    # DDP settings
    backend = 'nccl' # 'nccl', 'gloo', etc.
    # system
    device = 'cuda' # examples: 'cpu', 'cuda', 'cuda:0', 'cuda:1' etc., or try 'mps' on macbooks
    dtype = 'float16' # 'float32', 'bfloat16', or 'float16', the latter will auto implement a GradScaler
    compile = False # use PyTorch 2.0 to compile the model to be faster
    # -----------------------------------------------------------------------------
    config_keys = [
        k
        for k, v in globals().items()
        if not k.startswith("_") and isinstance(v, (int, float, bool, str))
    ]
    # exec(open("configurator.py").read())  # overrides from command line or config file
    # config = {k: globals()[k] for k in config_keys}  # will be useful for logging
    # -----------------------------------------------------------------------------

    save_dir =os.path.join(out_dir , '20230815_baike_pretrain')
    if not os.path.exists(save_dir): os.makedirs(save_dir)
    logger = get_logger(os.path.join(save_dir,'log.log'))
    # various inits, derived attributes, I/O setup
   # various inits, derived attributes, I/O setup
    ddp = int(os.environ.get("RANK", -1)) != -1  # is this a ddp run?
    
    if ddp:
        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)
        master_process = ddp_rank == 0  # this process will do logging, checkpointing etc.
        seed_offset = ddp_rank  # each process gets a different seed
        # world_size number of processes will be training simultaneously, so we can scale
        # down the desired gradient accumulation iterations per process proportionally
        #assert gradient_accumulation_steps % ddp_world_size == 0
        #gradient_accumulation_steps //= ddp_world_size
    else:
        # if not ddp, we are running on a single gpu, and one process
        master_process = True
        seed_offset = 0
        ddp_world_size = 1
    tokens_per_iter = gradient_accumulation_steps * ddp_world_size * batch_size * max_seq_len
    if master_process:
        print(f"tokens per iteration will be: {tokens_per_iter:,}")
        print(f"breaks down as: {gradient_accumulation_steps} grad accum steps * {ddp_world_size} processes * {batch_size} batch size * {max_seq_len} max seq len")

    if master_process:
        os.makedirs(out_dir, exist_ok=True)
    torch.manual_seed(1337 + seed_offset)
    torch.backends.cuda.matmul.allow_tf32 = True  # allow tf32 on matmul
    torch.backends.cudnn.allow_tf32 = True  # allow tf32 on cudnn
    device_type = "cuda" if "cuda" in device else "cpu"  # for later use in torch.autocast
    # note: float16 data type will automatically use a GradScaler
    ptdtype = {"float32": torch.float32, "bfloat16": torch.bfloat16, "float16": torch.float16}[dtype]
    ctx = (
        nullcontext()
        if device_type == "cpu"
        else torch.cuda.amp.autocast()
    )
    #
    best_val_loss = 1e9
    #
    #-----init dataloader------
    data_path_list=[
        './data/pretrain_data.bin'
        #'./data/baidubaike_563w.bin',
        #'./data/medical_book.bin',
        # './data/medical_encyclopedia.bin',
        # './data/medical_qa.bin',
        # './data/wiki.bin'
    ]
    train_ds = PretrainDataset(data_path_list, max_length=max_seq_len,memmap=True)
    train_sampler = torch.utils.data.distributed.DistributedSampler(train_ds)
    train_loader = torch.utils.data.DataLoader(
        train_ds,
        batch_size=batch_size,
        pin_memory=False,
        drop_last=False,
        shuffle=False,        
        num_workers=4,
        sampler=train_sampler
    )
    # val_ds = PretrainDataset(data_path_list, max_length=256)
    # val_loader = torch.utils.data.DataLoader(
    #     val_ds,
    #     batch_size=batch_size,
    #     pin_memory=False,
    #     drop_last=False,
    #     shuffle=False,        
    #     num_workers=0,
    # )
    #init model
    model=init_model()
    model.to(device)
    # initialize a GradScaler. If enabled=False scaler is a no-op
    scaler = torch.cuda.amp.GradScaler(enabled=(dtype == 'float16'))
    # optimizer
    optimizer = model.configure_optimizers(weight_decay, learning_rate, (beta1, beta2), device_type)
    #
    scheduler = torch.optim.lr_scheduler.CosineAnnealingWarmRestarts(optimizer, T_0=max_epoch, T_mult=1, eta_min=1e-6, last_epoch=-1)
    iter_per_epoch=len(train_loader)
    warmup_epoch=1
    
    # compile the model
    if compile:
        print("compiling the model... (takes a ~minute)")
        unoptimized_model = model
        model = torch.compile(model) # requires PyTorch 2.0
    # wrap model into DDP container
    if ddp:
        # Ignore the `freqs_cis` buffer so that DDP does not broadcast it at
        # construction time since NCCL does not support `ComplexFloat`
        prefix = "_orig_mod." if compile else ""
        model._ddp_params_and_buffers_to_ignore = {prefix + "freqs_cis"}
        model = DDP(model, device_ids=[ddp_local_rank])
        #
    raw_model = model.module if ddp else model # unwrap DDP container if needed
    # training loop
    for epoch in range(max_epoch):
        train_epoch(epoch)
        #val_loss=valid_epoch(epoch)
        if torch.distributed.get_rank() == 0:  #一般用0，当然，可以选任意的rank保存。
            #
            torch.save(raw_model.state_dict(),'{}/epoch_{}.pth'.format(save_dir,epoch))
    if ddp:
        destroy_process_group()