Deep learning Pytorch Optimizer不'；t更新参数
我制作了我的定制模型AlexNetQIL（Alexnet带有QIL层） “QIL”表示量化学习间隔Deep learning Pytorch Optimizer不'；t更新参数,deep-learning,pytorch,backpropagation,quantization,Deep Learning,Pytorch,Backpropagation,Quantization,我制作了我的定制模型AlexNetQIL（Alexnet带有QIL层） “QIL”表示量化学习间隔 我训练了我的模型，损失值并没有减少，我发现我的模型中的参数并没有更新，因为我添加了QIL层 我附上了我的代码AlexNetQil和Kil 请有人告诉我我的密码有什么问题 AlexNetQIL import torch import torch.nn as nn from qil import * class AlexNetQIL(nn.Module): #def __init__(se
import torch
import torch.nn as nn
from qil import *
class AlexNetQIL(nn.Module):
#def __init__(self, num_classes=1000): for imagenet
def __init__(self, num_classes=10): # for cifar-10
super(AlexNetQIL, self).__init__()
self.conv1 = nn.Conv2d(3, 64, kernel_size=3, stride=2, padding=1)
self.bn1 = nn.BatchNorm2d(64)
self.relu2 = nn.ReLU(inplace=True)
self.maxpool1 = nn.MaxPool2d(kernel_size=2)
self.qil2 = Qil()
self.conv2 = nn.Conv2d(64, 192, kernel_size=3, padding=1)
self.bn2 = nn.BatchNorm2d(192)
self.relu2 = nn.ReLU(inplace=True)
self.maxpool2 = nn.MaxPool2d(kernel_size=2)
self.qil3 = Qil()
self.conv3 = nn.Conv2d(192, 384, kernel_size=3, padding=1)
self.bn3 = nn.BatchNorm2d(384)
self.relu3 = nn.ReLU(inplace=True)
self.qil4 = Qil()
self.conv4 = nn.Conv2d(384, 256, kernel_size=3, padding=1)
self.bn4 = nn.BatchNorm2d(256)
self.relu4 = nn.ReLU(inplace=True)
self.qil5 = Qil()
self.conv5 = nn.Conv2d(256, 256, kernel_size=3, padding=1)
self.bn5 = nn.BatchNorm2d(256)
self.relu5 = nn.ReLU(inplace=True)
self.maxpool5 = nn.MaxPool2d(kernel_size=2)
self.classifier = nn.Sequential(
nn.Linear(256 * 2 * 2, 4096),
nn.ReLU(inplace=True),
nn.Linear(4096, 4096),
nn.ReLU(inplace=True),
nn.Linear(4096, num_classes),
)
def forward(self,x,inference = False):
x = self.conv1(x)
x = self.bn1(x)
x = self.relu2(x)
x = self.maxpool1(x)
x,self.conv2.weight = self.qil2(x,self.conv2.weight,inference ) # if I remove this line, No problem
x = self.conv2(x)
x = self.bn2(x)
x = self.relu2(x)
x = self.maxpool2(x)
x,self.conv3.weight = self.qil3(x,self.conv3.weight,inference ) # if I remove this line, No problem
x = self.conv3(x)
x = self.bn3(x)
x = self.relu3(x)
x,self.conv4.weight = self.qil4(x,self.conv4.weight,inference ) # if I remove this line, No problem
x = self.conv4(x)
x = self.bn4(x)
x = self.relu4(x)
x,self.conv5.weight = self.qil5(x,self.conv5.weight,inference ) # if I remove this line, No problem
x = self.conv5(x)
x = self.bn5(x)
x = self.relu5(x)
x = self.maxpool5(x)
x = x.view(x.size(0),256 * 2 * 2)
x = self.classifier(x)
return x
import torch
import torch.nn as nn
from qil import *
class AlexNetQIL(nn.Module):
#def __init__(self, num_classes=1000): for imagenet
def __init__(self, num_classes=10): # for cifar-10
super(AlexNetQIL, self).__init__()
self.conv1 = nn.Conv2d(3, 64, kernel_size=3, stride=2, padding=1)
self.bn1 = nn.BatchNorm2d(64)
self.relu2 = nn.ReLU(inplace=True)
self.maxpool1 = nn.MaxPool2d(kernel_size=2)
self.qil2 = Qil()
self.conv2 = nn.Conv2d(64, 192, kernel_size=3, padding=1)
self.bn2 = nn.BatchNorm2d(192)
self.relu2 = nn.ReLU(inplace=True)
self.maxpool2 = nn.MaxPool2d(kernel_size=2)
self.qil3 = Qil()
self.conv3 = nn.Conv2d(192, 384, kernel_size=3, padding=1)
self.bn3 = nn.BatchNorm2d(384)
self.relu3 = nn.ReLU(inplace=True)
self.qil4 = Qil()
self.conv4 = nn.Conv2d(384, 256, kernel_size=3, padding=1)
self.bn4 = nn.BatchNorm2d(256)
self.relu4 = nn.ReLU(inplace=True)
self.qil5 = Qil()
self.conv5 = nn.Conv2d(256, 256, kernel_size=3, padding=1)
self.bn5 = nn.BatchNorm2d(256)
self.relu5 = nn.ReLU(inplace=True)
self.maxpool5 = nn.MaxPool2d(kernel_size=2)
self.classifier = nn.Sequential(
nn.Linear(256 * 2 * 2, 4096),
nn.ReLU(inplace=True),
nn.Linear(4096, 4096),
nn.ReLU(inplace=True),
nn.Linear(4096, num_classes),
)
def forward(self,x,inference = False):
x = self.conv1(x)
x = self.bn1(x)
x = self.relu2(x)
x = self.maxpool1(x)
x,self.conv2.weight = self.qil2(x,self.conv2.weight,inference ) # if I remove this line, No problem
x = self.conv2(x)
x = self.bn2(x)
x = self.relu2(x)
x = self.maxpool2(x)
x,self.conv3.weight = self.qil3(x,self.conv3.weight,inference ) # if I remove this line, No problem
x = self.conv3(x)
x = self.bn3(x)
x = self.relu3(x)
x,self.conv4.weight = self.qil4(x,self.conv4.weight,inference ) # if I remove this line, No problem
x = self.conv4(x)
x = self.bn4(x)
x = self.relu4(x)
x,self.conv5.weight = self.qil5(x,self.conv5.weight,inference ) # if I remove this line, No problem
x = self.conv5(x)
x = self.bn5(x)
x = self.relu5(x)
x = self.maxpool5(x)
x = x.view(x.size(0),256 * 2 * 2)
x = self.classifier(x)
return x
- 通过两个步骤量化权重和输入激活
- 变压器（参数）->离散化器（参数）
导入火炬
导入torch.nn作为nn
将numpy作为np导入
导入副本
#Qil（量化区间学习）
类别Qil（nn.模块）：
离散化水平=32
定义初始化（自）：
超级（齐尔，自我）。\uuuu初始
self.cw=nn.Parameter（torch.rand（1））#我必须训练这个间隔参数
self.dw=nn.Parameter（torch.rand（1））#我必须训练这个间隔参数
self.cx=nn.Parameter（torch.rand（1））#我必须训练这个间隔参数
self.dx=nn.Parameter（torch.rand（1））#我必须训练这个间隔参数
self.gamma=nn.Parameter（火炬张量（1.0））#我必须训练这个变压器参数
self.a=齐尔离散化水平
def forward（自身、x、权重、推断=False）：
如果不是推断：
重量=自身变压器重量（重量）
重量=自离散化器（重量）
x=自变压器激活（x）
x=自离散化器（x）
返回火炬nn.参数（x），火炬nn.参数（重量）
def变压器重量（自身重量）：
设备=重量。设备
aw，bw=（0.5/self.dw），（-0.5*self.cw/self.dw+0.5）
重量=火炬，其中（abs（重量）self.cw+self.dw，
weights.sign（），weights）
重量=火炬，其中（（abs（重量）>=self.cw-self.dw）和（abs（重量）self.cx+self.dx，
火炬张量（1.0）.至（装置），x）
x=torch.where（（abs（x）>=self.cx-self.dx）&（abs（x）你是怎么训练的？你是说loss.backward吗？你确定.round
是可微的吗？是的，如果我删除使用.round的离散化函数，它会工作……但是我需要使用.round函数来量化我的参数。