写在前面
限于财力不足，本机上只有一个 GPU 可供使用，因此这部分的代码只能够稍作了解，能够使用的 GPU 也只有一个。
多 GPU 的数据并行：有几张卡，对一个小批量数据，有几张卡就分成几块，每个 GPU 分别计算梯度，然后加起来做并行。
从零开始实现
%matplotlib inline
import torch
from torch import nn
from torch.nn import functional as F
from d2l import torch as d2l
简单网络
# 初始化模型参数
scale = 0.01
W1 = torch.randn(size=(20, 1, 3, 3)) * scale
b1 = torch.zeros(20)
W2 = torch.randn(size=(50, 20, 5, 5)) * scale
b2 = torch.zeros(50)
W3 = torch.randn(size=(800, 128)) * scale
b3 = torch.zeros(128)
W4 = torch.randn(size=(128, 10)) * scale
b4 = torch.zeros(10)
params = [W1, b1, W2, b2, W3, b3, W4, b4]
# 定义模型
def lenet(X, params):
h1_conv = F.conv2d(input=X, weight=params[0], bias=params[1])
h1_activation = F.relu(h1_conv)
h1 = F.avg_pool2d(input=h1_activation, kernel_size=(2, 2), stride=(2, 2))
h2_conv = F.conv2d(input=h1, weight=params[2], bias=params[3])
h2_activation = F.relu(h2_conv)
h2 = F.avg_pool2d(input=h2_activation, kernel_size=(2, 2), stride=(2, 2))
h2 = h2.reshape(h2.shape[0], -1)
h3_linear = torch.mm(h2, params[4]) + params[5]
h3 = F.relu(h3_linear)
y_hat = torch.mm(h3, params[6]) + params[7]
return y_hat
# 交叉熵损失函数
loss = nn.CrossEntropyLoss(reduction='none')
向多个设备分发参数，并通过将模型参数复制到一个GPU：
def get_params(params, device): # 把一个参数复制到另外一个GPU上去
new_params = [p.to(device) for p in params]
for p in new_params:
p.requires_grad_() #对每一个参数都需要计算梯度
return new_params
new_params = get_params(params, d2l.try_gpu(0))
print('b1 权重:', new_params[1])
print('b1 梯度:', new_params[1].grad)
allreduce函数将所有向量相加，并将结果广播给所有GPU
def allreduce(data):
for i in range(1, len(data)):
data[0][:] += data[i].to(data[0].device)
for i in range(1, len(data)):
data[i][:] = data[0].to(data[i].device)
data = [torch.ones((1, 2), device=d2l.try_gpu(i)) * (i + 1) for i in range(2)]
print('allreduce之前：\n', data[0], '\n', data[1])
allreduce(data)
print('allreduce之后：\n', data[0], '\n', data[1])
将一个小批量数据均匀地分布在多个 GPU 上
data = torch.arange(20).reshape(4, 5)
devices = [torch.device('cuda:0'), torch.device('cuda:1')]
split = nn.parallel.scatter(data, devices)
print('input :', data)
print('load into', devices)
print('output:', split)
#@save
def split_batch(X, y, devices):
"""将X和y拆分到多个设备上"""
assert X.shape[0] == y.shape[0]
return (nn.parallel.scatter(X, devices),
nn.parallel.scatter(y, devices))
在一个小批量上实现多GPU训练
def train_batch(X, y, device_params, devices, lr):
X_shards, y_shards = split_batch(X, y, devices)
# 在每个GPU上分别计算损失
ls = [loss(lenet(X_shard, device_W), y_shard).sum()
for X_shard, y_shard, device_W in zip(
X_shards, y_shards, device_params)]
for l in ls: # 反向传播在每个GPU上分别执行
l.backward()
# 将每个GPU的所有梯度相加，并将其广播到所有GPU
with torch.no_grad():
for i in range(len(device_params[0])):
allreduce(
[device_params[c][i].grad for c in range(len(devices))])
# 在每个GPU上分别更新模型参数
for param in device_params:
d2l.sgd(param, lr, X.shape[0]) # 在这里，我们使用全尺寸的小批量
定义训练模型：
def train(num_gpus, batch_size, lr):
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size)
devices = [d2l.try_gpu(i) for i in range(num_gpus)]
# 将模型参数复制到num_gpus个GPU
device_params = [get_params(params, d) for d in devices]
num_epochs = 10
animator = d2l.Animator('epoch', 'test acc', xlim=[1, num_epochs])
timer = d2l.Timer()
for epoch in range(num_epochs):
timer.start()
for X, y in train_iter:
# 为单个小批量执行多GPU训练
train_batch(X, y, device_params, devices, lr)
torch.cuda.synchronize()
timer.stop()
# 在GPU0上评估模型
animator.add(epoch + 1, (d2l.evaluate_accuracy_gpu(
lambda x: lenet(x, device_params[0]), test_iter, devices[0]),))
print(f'测试精度：{animator.Y[0][-1]:.2f}，{timer.avg():.1f}秒/轮，'
f'在{str(devices)}')
在单个 GPU 上运行：
增加为 2 个 GPU
并行后并没有变快，可能有以下原因：
- Data 读取比较慢
- GPU 增加了，但是 batch_size 没有增加
多 GPU 的简洁实现
import torch
from torch import nn
from d2l import torch as d2l
简单网络
#@save
def resnet18(num_classes, in_channels=1):
"""稍加修改的ResNet-18模型"""
def resnet_block(in_channels, out_channels, num_residuals,
first_block=False):
blk = []
for i in range(num_residuals):
if i == 0 and not first_block:
blk.append(d2l.Residual(in_channels, out_channels,
use_1x1conv=True, strides=2))
else:
blk.append(d2l.Residual(out_channels, out_channels))
return nn.Sequential(*blk)
# 该模型使用了更小的卷积核、步长和填充，而且删除了最大汇聚层
net = nn.Sequential(
nn.Conv2d(in_channels, 64, kernel_size=3, stride=1, padding=1),
nn.BatchNorm2d(64),
nn.ReLU())
net.add_module("resnet_block1", resnet_block(
64, 64, 2, first_block=True))
net.add_module("resnet_block2", resnet_block(64, 128, 2))
net.add_module("resnet_block3", resnet_block(128, 256, 2))
net.add_module("resnet_block4", resnet_block(256, 512, 2))
net.add_module("global_avg_pool", nn.AdaptiveAvgPool2d((1,1)))
net.add_module("fc", nn.Sequential(nn.Flatten(),
nn.Linear(512, num_classes)))
return net
net = resnet18(10)
# 获取GPU列表
devices = d2l.try_all_gpus()
# 我们将在训练代码实现中初始化网络
训练
def train(net, num_gpus, batch_size, lr):
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size)
devices = [d2l.try_gpu(i) for i in range(num_gpus)]
def init_weights(m):
if type(m) in [nn.Linear, nn.Conv2d]:
nn.init.normal_(m.weight, std=0.01)
net.apply(init_weights)
# 在多个GPU上设置模型
net = nn.DataParallel(net, device_ids=devices)
trainer = torch.optim.SGD(net.parameters(), lr)
loss = nn.CrossEntropyLoss()
timer, num_epochs = d2l.Timer(), 10
animator = d2l.Animator('epoch', 'test acc', xlim=[1, num_epochs])
for epoch in range(num_epochs):
net.train()
timer.start()
for X, y in train_iter:
trainer.zero_grad()
X, y = X.to(devices[0]), y.to(devices[0])
l = loss(net(X), y)
l.backward()
trainer.step()
timer.stop()
animator.add(epoch + 1, (d2l.evaluate_accuracy_gpu(net, test_iter),))
print(f'测试精度：{animator.Y[0][-1]:.2f}，{timer.avg():.1f}秒/轮，'
f'在{str(devices)}')
在单个 GPU 上训练网络
train(net, num_gpus=1, batch_size=256, lr=0.1)
使用2个GPU进行训练
train(net, num_gpus=2, batch_size=512, lr=0.2)
QA 思考
Q1：验证集准确率震荡较大是哪个参数影响最大呢？
A1：lr
Q2：为什么batch_size调的比较小，比如8，精度会一直在0.1左右，一直不怎么变化
A2：因为batch_size调的比较小的时候，lr 不能太大。
