淘先锋技术网

首页 1 2 3 4 5 6 7

项目场景:

Linux系统,pytorch环境


问题描述

使用的服务器有两张显卡,感觉一张显卡跑代码比较慢,想配置两张显卡同时跑代码,只需要在你的代码中添加几行,就可以使用双显卡,亲测有效。


解决方案:

提示:这里填写该问题的具体解决方案:

先看以下官方示例代码,插入添加的地方是需要我们添加在代码中的代码行

import torch
import torch.nn as nn
from torch.utils.data import Dataset, DataLoader
import os 
#######添加
os.environ['CUDA_VISIBLE_DEVICES'] = '0,1' # 这里输入你的GPU_id

# Parameters and DataLoaders
input_size = 5
output_size = 2

batch_size = 30
data_size = 100
#######添加
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

# Dummy DataSet
class RandomDataset(Dataset):

    def __init__(self, size, length):
        self.len = length
        self.data = torch.randn(length, size)

    def __getitem__(self, index):
        return self.data[index]

    def __len__(self):
        return self.len

rand_loader = DataLoader(dataset=RandomDataset(input_size, data_size),
                         batch_size=batch_size, shuffle=True)

# Simple Model
class Model(nn.Module):
    # Our model

    def __init__(self, input_size, output_size):
        super(Model, self).__init__()
        self.fc = nn.Linear(input_size, output_size)

    def forward(self, input):
        output = self.fc(input)
        print("\tIn Model: input size", input.size(),
              "output size", output.size())

        return output
################添加
# Create Model and DataParallel
model = Model(input_size, output_size)
if torch.cuda.device_count() > 1:
  print("Let's use", torch.cuda.device_count(), "GPUs!")
  model = nn.DataParallel(model)
model.to(device)


#Run the Model
for data in rand_loader:
    input = data.to(device)
    output = model(input)
    print("Outside: input size", input.size(),
          "output_size", output.size())

其中我将model = nn.DataParallel(model)修改为model = nn.DataParallel(model.cuda()),这一步直接参照网上修改的,因此这一步没有报错。

比如我自己在我代码中添加如下:

from model.hash_model import DCMHT as DCMHT
import os
from tqdm import tqdm
import torch
import torch.nn as nn
from torch.utils.data import DataLoader
import scipy.io as scio


from .base import TrainBase
from model.optimization import BertAdam
from utils import get_args, calc_neighbor, cosine_similarity, euclidean_similarity
from utils.calc_utils import calc_map_k_matrix as calc_map_k
from dataset.dataloader import dataloader
###############添加
os.environ['CUDA_VISIBLE_DEVICES'] = '0,1' # 这里输入你的GPU_id
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

class Trainer(TrainBase):

    def __init__(self,
                rank=0):
        args = get_args()
        super(Trainer, self).__init__(args, rank)
        self.logger.info("dataset len: {}".format(len(self.train_loader.dataset)))
        self.run()

    def _init_model(self):
        self.logger.info("init model.")
        linear = False
        if self.args.hash_layer == "linear":
            linear = True

        self.logger.info("ViT+GPT!")
        HashModel = DCMHT
        self.model = HashModel(outputDim=self.args.output_dim, clipPath=self.args.clip_path,
                            writer=self.writer, logger=self.logger, is_train=self.args.is_train, linear=linear).to(self.rank)
####################################添加
        self.model= nn.DataParallel(self.model.cuda())
        if torch.cuda.device_count() >1:
            print("Lets use",torch.cuda.device_count(),"GPUs!")
        self.model.to(device)

        if self.args.pretrained != "" and os.path.exists(self.args.pretrained):
            self.logger.info("load pretrained model.")
            self.model.load_state_dict(torch.load(self.args.pretrained, map_location=f"cuda:{self.rank}"))
        
        self.model.float()
        self.optimizer = BertAdam([
                    {'params': self.model.clip.parameters(), 'lr': self.args.clip_lr},
                    {'params': self.model.image_hash.parameters(), 'lr': self.args.lr},
                    {'params': self.model.text_hash.parameters(), 'lr': self.args.lr}
                    ], lr=self.args.lr, warmup=self.args.warmup_proportion, schedule='warmup_cosine', 
                    b1=0.9, b2=0.98, e=1e-6, t_total=len(self.train_loader) * self.args.epochs,
                    weight_decay=self.args.weight_decay, max_grad_norm=1.0)
                
        print(self.model)

 添加以上代码后一般还会报如下错误:

“AttributeError: ‘DataParallel’ object has no attribute ‘xxx’”,解决办法为先在dataparallel后的model调用module模块,然后再调用xxx。

 比如在上述我自己的代码中会报错“AttributeError: ‘DataParallel’ object has no attribute ‘clip’”,

解决办法是将model,修改为model.module.,后续报错大致相同,将你的代码中涉及到model.的地方修改为model.module.即可。

self.optimizer = BertAdam([
                    {'params': self.model.module.clip.parameters(), 'lr': self.args.clip_lr},
                    {'params': self.model.module.image_hash.parameters(), 'lr': self.args.lr},
                    {'params': self.model.module.text_hash.parameters(), 'lr': self.args.lr}
                    ], lr=self.args.lr, warmup=self.args.warmup_proportion, 

 检查显卡使用情况

打开终端,在终端输入nvidia-smi命令可查看显卡使用情况

成功使用双显卡跑代码!