(beta) PyTorch 中使用 Eager 模式的静态量化 ¶
译者:片刻小哥哥
项目地址:https://pytorch.apachecn.org/2.0/tutorials/advanced/static_quantization_tutorial
原始地址:https://pytorch.org/tutorials/advanced/static_quantization_tutorial.html
作者 : Raghuraman Krishnamoorthi 编辑 : Seth Weidman , 张杰
本教程展示如何进行训练后静态量化,并说明 两种更先进的技术 - 每通道量化和量化感知训练 - 以进一步提高模型’s 的准确性。请注意,目前仅支持 CPU 量化,因此我们在本教程中不会使用 GPU/CUDA。 在本教程结束时,您将看到 PyTorch 中的量化 如何导致模型大小显着减小,同时增加模型大小速度。此外,您’将了解如何 轻松应用此处 所示的一些高级量化技术 ,以便您的量化模型 花费更少
警告:我们使用其他 PyTorch 存储库中的大量样板代码来定义
MobileNetV2
模型架构、定义数据加载器等。我们当然鼓励您阅读它;但如果您想了解量化功能,请随意跳到 “4。训练后静态量化” 部分。
我们’ 将首先进行必要的导入:
import os
import sys
import time
import numpy as np
import torch
import torch.nn as nn
from torch.utils.data import DataLoader
import torchvision
from torchvision import datasets
import torchvision.transforms as transforms
# Set up warnings
import warnings
warnings.filterwarnings(
action='ignore',
category=DeprecationWarning,
module=r'.*'
)
warnings.filterwarnings(
action='default',
module=r'torch.ao.quantization'
)
# Specify random seed for repeatable results
torch.manual_seed(191009)
1. 模型架构 ¶
我们首先定义 MobileNetV2 模型架构,并进行一些显着的修改 以启用量化:
- 将加法替换为
nn.quantized.FloatFunctional
- 在网络的开头和结尾插入
QuantStub
和DeQuantStub
。 - 将 ReLU6 替换为 ReLU
注意:此代码取自 此处 .
from torch.ao.quantization import QuantStub, DeQuantStub
def _make_divisible(v, divisor, min_value=None):
"""
This function is taken from the original tf repo.
It ensures that all layers have a channel number that is divisible by 8
It can be seen here:
https://github.com/tensorflow/models/blob/master/research/slim/nets/mobilenet/mobilenet.py
:param v:
:param divisor:
:param min_value:
:return:
"""
if min_value is None:
min_value = divisor
new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
# Make sure that round down does not go down by more than 10%.
if new_v < 0.9 * v:
new_v += divisor
return new_v
类 ConvBNReLU(nn.Sequential):
def __init__(self, in_planes, out_planes, kernel_size=3, stride=1, groups=1 ):
padding = (kernel_size - 1) //2
super(ConvBNReLU, self).__init__(
nn.Conv2d(in_planes, out_planes, kernel_size, stride, padding, groups=groups,bias=False),
nn.BatchNorm2d(out_planes, Momentum=0.1),
# 替换为 ReLU
nn.ReLU (原地=假)
)
class InvertedResidual(nn.Module):
def __init__(self, inp, oup, stride, Expand_ratio):
super(InvertedResidual, self)._ _init__()
self.stride = stride
在 [1, 2] 中断言步幅
hide_dim = int(round(inp * Expand_ratio) )
self.use_res_connect = self.stride == 1 且 inp == oup
层 = []
if Expand_ratio != 1:
# pw
层.append(ConvBNReLU(inp,hidden_dim,kernel_size=1))
层.extend([
# dw
ConvBNReLU(hidden_dim,hidden_dim,stride=stride, groups =hidden_dim),
# pw-线性
nn.Conv2d(hidden_dim, oup, 1, 1, 0, 偏差=False),
nn.BatchNorm2d(oup, 动量=0.1),
])
self.conv = nn.Sequential(*layers)
# 将 torch.add 替换为 floatFunctional
self.skip_add = nn.quantized.FloatFunctional()
defforward (self, x):
如果 self.use_res_connect:
返回 self.skip_add.add(x, self.conv(x))
else:
返回 self.conv (X)
class MobileNetV2(nn.Module):
def __init__(self, num_classes=1000, width_mult=1.0, inverted_residual_setting=None, round_nearest=8):
"""
MobileNet V2 main class
Args:
num_classes (int): Number of classes
width_mult (float): Width multiplier - adjusts number of channels in each layer by this amount
inverted_residual_setting: Network structure
round_nearest (int): Round the number of channels in each layer to be a multiple of this number
Set to 1 to turn off rounding
"""
super(MobileNetV2, self).__init__()
block = InvertedResidual
input_channel = 32
last_channel = 1280
if inverted_residual_setting is None:
inverted_residual_setting = [
# t, c, n, s
[1, 16, 1, 1],
[6, 24, 2, 2],
[6, 32, 3, 2],
[6, 64, 4, 2],
[6, 96, 3, 1],
[6, 160, 3, 2],
[6, 320, 1, 1],
]
# only check the first element, assuming user knows t,c,n,s are required
if len(inverted_residual_setting) == 0 or len(inverted_residual_setting[0]) != 4:
raise ValueError("inverted_residual_setting should be non-empty "
"or a 4-element list, got {}".format(inverted_residual_setting))
# building first layer
input_channel = _make_divisible(input_channel * width_mult, round_nearest)
self.last_channel = _make_divisible(last_channel * max(1.0, width_mult), round_nearest)
features = [ConvBNReLU(3, input_channel, stride=2)]
# building inverted residual blocks
for t, c, n, s in inverted_residual_setting:
output_channel = _make_divisible(c * width_mult, round_nearest)
for i in range(n):
stride = s if i == 0 else 1
features.append(block(input_channel, output_channel, stride, expand_ratio=t))
input_channel = output_channel
# building last several layers
features.append(ConvBNReLU(input_channel, self.last_channel, kernel_size=1))
# make it nn.Sequential
self.features = nn.Sequential(*features)
self.quant = QuantStub()
self.dequant = DeQuantStub()
# building classifier
self.classifier = nn.Sequential(
nn.Dropout(0.2),
nn.Linear(self.last_channel, num_classes),
)
# weight initialization
for m in self.modules():
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight, mode='fan_out')
if m.bias is not None:
nn.init.zeros_(m.bias)
elif isinstance(m, nn.BatchNorm2d):
nn.init.ones_(m.weight)
nn.init.zeros_(m.bias)
elif isinstance(m, nn.Linear):
nn.init.normal_(m.weight, 0, 0.01)
nn.init.zeros_(m.bias)
def forward(self, x):
x = self.quant(x)
x = self.features(x)
x = x.mean([2, 3])
x = self.classifier(x)
x = self.dequant(x)
return x
# Fuse Conv+BN and Conv+BN+Relu modules prior to quantization
# This operation does not change the numerics
def fuse_model(self, is_qat=False):
fuse_modules = torch.ao.quantization.fuse_modules_qat if is_qat else torch.ao.quantization.fuse_modules
for m in self.modules():
if type(m) == ConvBNReLU:
fuse_modules(m, ['0', '1', '2'], inplace=True)
if type(m) == InvertedResidual:
for idx in range(len(m.conv)):
if type(m.conv[idx]) == nn.Conv2d:
fuse_modules(m.conv, [str(idx), str(idx + 1)], inplace=True)
2. 辅助函数 ¶
-
接下来我们定义几个辅助函数来帮助模型评估。这些大多来自 此处 。
class AverageMeter(object):
"""Computes and stores the average and current value"""
def __init__(self, name, fmt=':f'):
self.name = name
self.fmt = fmt
self.reset()
def reset(self):
self.val = 0
self.avg = 0
self.sum = 0
self.count = 0
def update(self, val, n=1):
self.val = val
self.sum += val * n
self.count += n
self.avg = self.sum / self.count
def __str__(self):
fmtstr = '{name} {val' + self.fmt + '} ({avg' + self.fmt + '})'
return fmtstr.format(**self.__dict__)
def precision(output, target, topk=(1,)):
"""使用 torch.no_grad() 计算指定 k 值的 k 个前预测的准确度"""
:\ n maxk = max(topk)
batch_size = target.size(0)
_, pred = output.topk(maxk, 1, True, True)
pred = pred.t( )
正确 = pred.eq(target.view(1, -1).expand_as(pred))
res = []
for k in topk:
正确_k = 正确[:k].reshape(-1).float().sum(0, keepdim=True)
res.append(正确_k.mul_(100.0 /batch_size))
return资源
def evaluate(model, criterion, data_loader, neval_batches):
model.eval()
top1 = AverageMeter('Acc@1', ':6.2f')
top5 = AverageMeter('Acc@5', ':6.2f')
cnt = 0
with torch.no_grad():
for image, target in data_loader:
output = model(image)
loss = criterion(output, target)
cnt += 1
acc1, acc5 = accuracy(output, target, topk=(1, 5))
print('.', end = '')
top1.update(acc1[0], image.size(0))
top5.update(acc5[0], image.size(0))
if cnt >= neval_batches:
return top1, top5
return top1, top5
def load_model(model_file):
model = MobileNetV2()
state_dict = torch.load(model_file)
model.load_state_dict(state_dict)
model.to('cpu')
return model
def print_size_of_model(model):
torch.save(model.state_dict(), "temp.p")
print('Size (MB):', os.path.getsize("temp.p")/1e6)
os.remove('temp.p')
3. 定义数据集和数据加载器 ¶
作为最后一个主要设置步骤,我们为训练和测试集定义数据加载器。
ImageNet 数据 ¶
要使用整个 ImageNet 数据集运行本教程中的代码,请首先按照此处的说明下载 imagenet ImageNet Data 。将下载的文件解压缩到 ‘data_path’ 文件夹中。
下载数据后,我们将显示下面的函数,这些函数定义我们’将用来读取此数据的数据加载器。这些函数主要来自 此处 。
def prepare_data_loaders(data_path):
normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
dataset = torchvision.datasets.ImageNet(
data_path, split="train", transform=transforms.Compose([
transforms.RandomResizedCrop(224),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
normalize,
]))
dataset_test = torchvision.datasets.ImageNet(
data_path, split="val", transform=transforms.Compose([
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
normalize,
]))
train_sampler = torch.utils.data.RandomSampler(dataset)
test_sampler = torch.utils.data.SequentialSampler(dataset_test)
data_loader = torch.utils.data.DataLoader(
dataset, batch_size=train_batch_size,
sampler=train_sampler)
data_loader_test = torch.utils.data.DataLoader(
dataset_test, batch_size=eval_batch_size,
sampler=test_sampler)
return data_loader, data_loader_test
接下来,我们’ 将加载预训练的 MobileNetV2 模型。我们提供下载模型的 URL 此处 .
data_path = '~/.data/imagenet'
saved_model_dir = 'data/'
float_model_file = 'mobilenet_pretrained_float.pth'
scripted_float_model_file = 'mobilenet_quantization_scripted.pth'
scripted_quantized_model_file = 'mobilenet_quantization_scripted_quantized.pth'
train_batch_size = 30
eval_batch_size = 50
data_loader, data_loader_test = prepare_data_loaders(data_path)
criterion = nn.CrossEntropyLoss()
float_model = load_model(saved_model_dir + float_model_file).to('cpu')
# Next, we'll "fuse modules"; this can both make the model faster by saving on memory access
# while also improving numerical accuracy. While this can be used with any model, this is
# especially common with quantized models.
print(' Inverted Residual Block: Before fusion ', float_model.features[1].conv)
float_model.eval()
# Fuses modules
float_model.fuse_model()
# Note fusion of Conv+BN+Relu and Conv+Relu
print(' Inverted Residual Block: After fusion',float_model.features[1].conv)
最后,为了获得 “baseline” 精度,让 ’s 查看我们的未量化模型 与融合模块的精度
num_eval_batches = 1000
print("Size of baseline model")
print_size_of_model(float_model)
top1, top5 = evaluate(float_model, criterion, data_loader_test, neval_batches=num_eval_batches)
print('Evaluation accuracy on %d images, %2.2f'%(num_eval_batches * eval_batch_size, top1.avg))
torch.jit.save(torch.jit.script(float_model), saved_model_dir + scripted_float_model_file)
在整个模型上,我们在 50,000 张图像的评估数据集上获得了 71.9% 的准确率。
这将是我们进行比较的基准。接下来让’s尝试不同的量化方法
4. 训练后静态量化 ¶
训练后静态量化不仅涉及动态量化中将权重从 float 转换为 int, 还需要执行额外的步骤:首先通过网络馈送 数据并计算不同激活的结果分布 (具体来说,这是通过在记录此数据的不同点插入观察者模块来完成的)。然后,使用这些分布来确定在推理时应如何具体量化不同的激活(一种简单的技术是将整个激活范围简单地分为 256 个级别,但我们也支持更复杂的方法)。重要的是,这个额外的步骤允许我们在操作之间传递量化值,而不是在每个操作之间将这些值转换为浮点数,然后再转换回整数,从而显着提高速度。
num_calibration_batches = 32
myModel = load_model(saved_model_dir + float_model_file).to('cpu')
myModel.eval()
# Fuse Conv, bn and relu
myModel.fuse_model()
# Specify quantization configuration
# Start with simple min/max range estimation and per-tensor quantization of weights
myModel.qconfig = torch.ao.quantization.default_qconfig
print(myModel.qconfig)
torch.ao.quantization.prepare(myModel, inplace=True)
# Calibrate first
print('Post Training Quantization Prepare: Inserting Observers')
print(' Inverted Residual Block:After observer insertion ', myModel.features[1].conv)
# Calibrate with the training set
evaluate(myModel, criterion, data_loader, neval_batches=num_calibration_batches)
print('Post Training Quantization: Calibration done')
# Convert to quantized model
torch.ao.quantization.convert(myModel, inplace=True)
print('Post Training Quantization: Convert done')
print(' Inverted Residual Block: After fusion and quantization, note fused modules: ',myModel.features[1].conv)
print("Size of model after quantization")
print_size_of_model(myModel)
top1, top5 = evaluate(myModel, criterion, data_loader_test, neval_batches=num_eval_batches)
print('Evaluation accuracy on %d images, %2.2f'%(num_eval_batches * eval_batch_size, top1.avg))
对于这个量化模型,我们在评估数据集上看到的准确率为 56.7%。这是因为我们使用简单的最小/最大观察器来确定量化参数。尽管如此,我们确实将模型的大小减少到略低于 3.6 MB,几乎减少了 4 倍。
此外,我们只需使用不同的量化配置即可显着提高准确性。我们使用 x86 架构量化的推荐配置重复相同的练习。此配置执行以下操作:
- 基于每个通道量化权重
- 使用直方图观察器收集激活直方图,然后以最佳方式选择 量化参数。
per_channel_quantized_model = load_model(saved_model_dir + float_model_file)
per_channel_quantized_model.eval()
per_channel_quantized_model.fuse_model()
# The old 'fbgemm' is still available but 'x86' is the recommended default.
per_channel_quantized_model.qconfig = torch.ao.quantization.get_default_qconfig('x86')
print(per_channel_quantized_model.qconfig)
torch.ao.quantization.prepare(per_channel_quantized_model, inplace=True)
evaluate(per_channel_quantized_model,criterion, data_loader, num_calibration_batches)
torch.ao.quantization.convert(per_channel_quantized_model, inplace=True)
top1, top5 = evaluate(per_channel_quantized_model, criterion, data_loader_test, neval_batches=num_eval_batches)
print('Evaluation accuracy on %d images, %2.2f'%(num_eval_batches * eval_batch_size, top1.avg))
torch.jit.save(torch.jit.script(per_channel_quantized_model), saved_model_dir + scripted_quantized_model_file)
仅更改此量化配置方法即可将准确度提高到 超过 67.3%!尽管如此,这仍然比上面达到的 71.9% 的基线差了 4%。 所以让我们尝试量化感知训练。
5. 量化感知训练 ¶
量化感知训练 (QAT) 是通常能获得最高准确度的量化方法。 使用 QAT,所有权重和激活在前向和后向传播过程中均被“fake 量化”训练:也就是说,浮点值被四舍五入以模拟 int8 值,但所有计算仍然使用浮点数完成。因此,训练期间的所有权重调整都是在模型最终被量化的情况下进行的;因此,在量化之后,此方法通常会 比动态量化或训练后静态量化产生更高的精度。
实际执行 QAT 的整体工作流程与之前非常相似:
- 我们可以使用与之前相同的模型:量化感知 训练不需要额外的准备。
- 我们需要使用
qconfig
指定在后面插入哪种假量化权重 和激活,而不是指定观察者
我们首先定义一个训练函数:
def train_one_epoch(model, criterion, optimizer, data_loader, device, ntrain_batches):
model.train()
top1 = AverageMeter('Acc@1', ':6.2f')
top5 = AverageMeter('Acc@5', ':6.2f')
avgloss = AverageMeter('Loss', '1.5f')
cnt = 0
for image, target in data_loader:
start_time = time.time()
print('.', end = '')
cnt += 1
image, target = image.to(device), target.to(device)
output = model(image)
loss = criterion(output, target)
optimizer.zero_grad()
loss.backward()
optimizer.step()
acc1, acc5 = accuracy(output, target, topk=(1, 5))
top1.update(acc1[0], image.size(0))
top5.update(acc5[0], image.size(0))
avgloss.update(loss, image.size(0))
if cnt >= ntrain_batches:
print('Loss', avgloss.avg)
print('Training: * Acc@1 {top1.avg:.3f} Acc@5 {top5.avg:.3f}'
.format(top1=top1, top5=top5))
return
print('Full imagenet train set: * Acc@1 {top1.global_avg:.3f} Acc@5 {top5.global_avg:.3f}'
.format(top1=top1, top5=top5))
return
我们像以前一样融合模块
qat_model = load_model(saved_model_dir + float_model_file)
qat_model.fuse_model(is_qat=True)
optimizer = torch.optim.SGD(qat_model.parameters(), lr = 0.0001)
# The old 'fbgemm' is still available but 'x86' is the recommended default.
qat_model.qconfig = torch.ao.quantization.get_default_qat_qconfig('x86')
最后,
prepare_qat
执行 “fake 量化”,为量化感知训练准备模型
torch.ao.quantization.prepare_qat(qat_model, inplace=True)
print('Inverted Residual Block: After preparation for QAT, note fake-quantization modules ',qat_model.features[1].conv)
训练高精度量化模型需要在推理时对数值进行精确建模。因此,对于量化感知训练,我们通过以下方式修改训练循环:
- 在训练结束时切换批量归一化以使用运行均值和方差,以更好 匹配推理数值。
- 我们还冻结量化器参数(比例和零点)并微调权重。
num_train_batches = 20
# QAT takes time and one needs to train over a few epochs.
# Train and check accuracy after each epoch
for nepoch in range(8):
train_one_epoch(qat_model, criterion, optimizer, data_loader, torch.device('cpu'), num_train_batches)
if nepoch > 3:
# Freeze quantizer parameters
qat_model.apply(torch.ao.quantization.disable_observer)
if nepoch > 2:
# Freeze batch norm mean and variance estimates
qat_model.apply(torch.nn.intrinsic.qat.freeze_bn_stats)
# Check the accuracy after each epoch
quantized_model = torch.ao.quantization.convert(qat_model.eval(), inplace=False)
quantized_model.eval()
top1, top5 = evaluate(quantized_model,criterion, data_loader_test, neval_batches=num_eval_batches)
print('Epoch %d :Evaluation accuracy on %d images, %2.2f'%(nepoch, num_eval_batches * eval_batch_size, top1.avg))
量化感知训练在整个 imagenet 数据集上的准确率超过 71.5%,接近浮点准确率 71.9%。
有关量化感知训练的更多信息:
- QAT 是训练后量化技术的超集,允许进行更多调试。 例如,我们可以分析模型的准确性是否受到权重或激活 量化的限制。
- 我们还可以模拟浮点量化模型,因为 我们使用假量化来对实际量化算术的数值进行建模。 *我们也可以轻松模仿训练后量化。
量化加速 ¶
最后,让’s 确认一下我们上面提到的事情:我们的量化模型实际上 执行推理速度更快吗?让’s 测试:
def run_benchmark(model_file, img_loader):
elapsed = 0
model = torch.jit.load(model_file)
model.eval()
num_batches = 5
# Run the scripted model on a few batches of images
for i, (images, target) in enumerate(img_loader):
if i < num_batches:
start = time.time()
output = model(images)
end = time.time()
elapsed = elapsed + (end-start)
else:
break
num_images = images.size()[0] * num_batches
print('Elapsed time: %3.0f ms' % (elapsed/num_images*1000))
return elapsed
run_benchmark(saved_model_dir + scripted_float_model_file, data_loader_test)
run_benchmark(saved_model_dir + scripted_quantized_model_file, data_loader_test)
在 MacBook Pro 上本地运行该模型,常规模型需要 61 毫秒, 量化模型只需 20 毫秒,这说明了与浮点模型相比,我们看到量化模型典型的 2-4 倍加速。
结论 ¶
在本教程中,我们展示了两种量化方法 - 训练后静态量化和量化感知训练 - 描述了它们在底层的用途 “” 以及如何在 中使用它们PyTorch。
感谢您的阅读!一如既往,我们欢迎任何反馈,因此请在此处 创建问题 (如果有任何反馈)。