221 lines
8.8 KiB
Python
221 lines
8.8 KiB
Python
# 导入必要的库
|
||
import os
|
||
import sys
|
||
__package__ = "trainer" # 设置包名
|
||
# 将项目根目录添加到系统路径
|
||
sys.path.append(os.path.abspath(os.path.join(os.path.dirname(__file__), '..')))
|
||
|
||
import argparse
|
||
import time
|
||
import math
|
||
import warnings
|
||
import torch
|
||
import torch.distributed as dist
|
||
from torch import optim, nn
|
||
from torch.nn.parallel import DistributedDataParallel
|
||
from torch.utils.data import DataLoader, DistributedSampler
|
||
from contextlib import nullcontext
|
||
from transformers import AutoTokenizer
|
||
from model.model_minimind import MiniMindConfig, MiniMindForCausalLM
|
||
from dataset.lm_dataset import PretrainDataset
|
||
|
||
warnings.filterwarnings('ignore')
|
||
|
||
|
||
# 日志打印函数
|
||
# 在分布式训练时只在主进程(rank=0)上打印日志
|
||
def Logger(content):
|
||
if not ddp or dist.get_rank() == 0:
|
||
print(content)
|
||
|
||
|
||
# 余弦学习率调度器
|
||
# 在训练过程中逐渐降低学习率，最终降到初始值的1/10
|
||
def get_lr(current_step, total_steps, lr):
|
||
return lr / 10 + 0.5 * lr * (1 + math.cos(math.pi * current_step / total_steps))
|
||
|
||
|
||
# 训练一个epoch
|
||
def train_epoch(epoch, wandb):
|
||
# 使用交叉熵损失函数，reduction='none'以便后续通过mask处理填充token
|
||
loss_fct = nn.CrossEntropyLoss(reduction='none')
|
||
start_time = time.time()
|
||
for step, (X, Y, loss_mask) in enumerate(train_loader):
|
||
X = X.to(args.device)
|
||
Y = Y.to(args.device)
|
||
loss_mask = loss_mask.to(args.device)
|
||
|
||
lr = get_lr(epoch * iter_per_epoch + step, args.epochs * iter_per_epoch, args.learning_rate)
|
||
for param_group in optimizer.param_groups:
|
||
param_group['lr'] = lr
|
||
|
||
with ctx:
|
||
res = model(X)
|
||
loss = loss_fct(
|
||
res.logits.view(-1, res.logits.size(-1)),
|
||
Y.view(-1)
|
||
).view(Y.size())
|
||
loss = (loss * loss_mask).sum() / loss_mask.sum()
|
||
loss += res.aux_loss
|
||
loss = 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:{:.12f} epoch_Time:{}min:'.format(
|
||
epoch + 1,
|
||
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}/pretrain_{lm_config.hidden_size}{moe_path}.pth'
|
||
|
||
if isinstance(model, torch.nn.parallel.DistributedDataParallel):
|
||
state_dict = model.module.state_dict()
|
||
else:
|
||
state_dict = model.state_dict()
|
||
|
||
state_dict = {k: v.half() for k, v in state_dict.items()} # 半精度保存
|
||
torch.save(state_dict, ckp)
|
||
model.train()
|
||
|
||
|
||
# 初始化模型和分词器
|
||
def init_model(lm_config):
|
||
# 加载预训练的分词器
|
||
tokenizer = AutoTokenizer.from_pretrained('../model/')
|
||
# 初始化因果语言模型并移动到指定设备
|
||
model = MiniMindForCausalLM(lm_config).to(args.device)
|
||
# 打印模型可训练参数总量
|
||
Logger(f'LLM可训练总参数量：{sum(p.numel() for p in model.parameters() if p.requires_grad) / 1e6:.3f} 百万')
|
||
return model, tokenizer
|
||
|
||
|
||
# 初始化分布式训练环境
|
||
def init_distributed_mode():
|
||
if not ddp: return
|
||
global ddp_local_rank, DEVICE
|
||
|
||
# 初始化分布式进程组，使用NCCL后端
|
||
dist.init_process_group(backend="nccl")
|
||
# 获取分布式训练的相关参数
|
||
ddp_rank = int(os.environ["RANK"]) # 全局进程编号
|
||
ddp_local_rank = int(os.environ["LOCAL_RANK"]) # 本地进程编号
|
||
ddp_world_size = int(os.environ["WORLD_SIZE"]) # 总进程数
|
||
# 设置当前进程使用的GPU
|
||
DEVICE = f"cuda:{ddp_local_rank}"
|
||
torch.cuda.set_device(DEVICE)
|
||
|
||
|
||
# torchrun --nproc_per_node 2 1-pretrain.py
|
||
if __name__ == "__main__":
|
||
parser = argparse.ArgumentParser(description="MiniMind Pretraining")
|
||
parser.add_argument("--out_dir", type=str, default="../out")
|
||
# 若要以最快速度实现zero则epochs设置为1轮；否则应当利用有限的数据训练2~6个epochs。
|
||
parser.add_argument("--epochs", type=int, default=1)
|
||
parser.add_argument("--batch_size", type=int, default=32)
|
||
parser.add_argument("--learning_rate", type=float, default=5e-4)
|
||
parser.add_argument("--device", type=str, default="cuda:0" if torch.cuda.is_available() else "cpu")
|
||
parser.add_argument("--dtype", type=str, default="bfloat16")
|
||
parser.add_argument("--use_wandb", action="store_true")
|
||
parser.add_argument("--wandb_project", type=str, default="MiniMind-Pretrain")
|
||
parser.add_argument("--num_workers", type=int, default=1)
|
||
parser.add_argument("--ddp", action="store_true")
|
||
parser.add_argument("--accumulation_steps", type=int, default=8)
|
||
parser.add_argument("--grad_clip", type=float, default=1.0)
|
||
parser.add_argument("--warmup_iters", type=int, default=0)
|
||
parser.add_argument("--log_interval", type=int, default=100)
|
||
parser.add_argument("--save_interval", type=int, default=100)
|
||
parser.add_argument('--local_rank', type=int, default=-1)
|
||
parser.add_argument('--hidden_size', default=512, type=int)
|
||
parser.add_argument('--num_hidden_layers', default=8, type=int)
|
||
parser.add_argument('--max_seq_len', default=512, type=int)
|
||
parser.add_argument('--use_moe', default=False, type=bool)
|
||
parser.add_argument("--data_path", type=str, default="../dataset/pretrain_hq.jsonl")
|
||
parser.add_argument("--wandb_api_key", type=str, default=None, help="WandB API Key，用于无交互环境自动登录")
|
||
args = parser.parse_args()
|
||
|
||
lm_config = MiniMindConfig(hidden_size=args.hidden_size, num_hidden_layers=args.num_hidden_layers, use_moe=args.use_moe)
|
||
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)
|
||
tokens_per_iter = args.batch_size * args.max_seq_len
|
||
device_type = "cuda" if "cuda" in args.device else "cpu"
|
||
|
||
args.wandb_run_name = f"MiniMind-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"
|
||
|
||
base_seed = 1337
|
||
torch.manual_seed(base_seed)
|
||
torch.cuda.manual_seed(base_seed)
|
||
|
||
if ddp:
|
||
init_distributed_mode()
|
||
args.device = torch.device(DEVICE)
|
||
rank = dist.get_rank()
|
||
torch.manual_seed(base_seed + rank)
|
||
# 同时设置 CUDA 的随机种子
|
||
torch.cuda.manual_seed(base_seed + rank)
|
||
|
||
if args.use_wandb and (not ddp or ddp_local_rank == 0):
|
||
import wandb
|
||
# 优先使用命令行参数，其次环境变量
|
||
api_key = args.wandb_api_key or os.environ.get("WANDB_API_KEY", None)
|
||
if api_key is not None:
|
||
# 自动登录，适用于无交互环境
|
||
wandb.login(key=api_key)
|
||
# 初始化 wandb 任务
|
||
wandb.init(project=args.wandb_project, name=args.wandb_run_name)
|
||
else:
|
||
wandb = None
|
||
|
||
model, tokenizer = init_model(lm_config)
|
||
train_ds = PretrainDataset(args.data_path, tokenizer, max_length=args.max_seq_len)
|
||
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
|
||
)
|
||
|
||
scaler = torch.cuda.amp.GradScaler(enabled=(args.dtype in ['float16', 'bfloat16']))
|
||
optimizer = optim.AdamW(model.parameters(), lr=args.learning_rate)
|
||
|
||
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)
|