python/modelQuantizationPreTrained.py at master · anantguptadbl/python · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
import os
import numpy as np
import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
from torch.autograd import Variable
import torch.utils.data as data
import torchvision
from torchvision import transforms
import torchvision.models as models

# VGG Model
vgg16 = models.vgg16(pretrained=True)

# Requires Grad False
# Freezing the first few layers of the alexnet model
for moduleIndex in range(30):
    for param in vgg16.features._modules[str(moduleIndex)].parameters():
        param.requires_grad=False
        print(param.requires_grad)

transform_train = transforms.Compose([
    transforms.RandomResizedCrop(224),
    transforms.ToTensor(),
    transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
   ])

transform_test = transforms.Compose([
    transforms.RandomResizedCrop(224),
    transforms.ToTensor(),
    transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
])

class LimitDataset(data.Dataset):
    def __init__(self, dataset, n):
        self.dataset = dataset
        self.n = n

    def __len__(self):
        return self.n

    def __getitem__(self, i):
        return self.dataset[i]


trainset=LimitDataset(torchvision.datasets.CIFAR10(root='./data', train=True, transform=transform_train, download=False),1500)
testset=LimitDataset(torchvision.datasets.CIFAR10(root='./data', train=False, transform=transform_test, download=False),1500)
#trainset = torchvision.datasets.CIFAR10(root='./data', train=True, transform=transform_train, download=False)
#testset = torchvision.datasets.CIFAR10(root='./data', train=False, download=False, transform=transform_test)
print("Cell Execution Completed")


import torch.optim as optim

torch.manual_seed(42)

class Transfer(nn.Module):
    def __init__(self):
        super(Transfer, self).__init__()
        self.vgg16=vgg16.features.cuda()
        self.layer1=nn.Linear(512*7*7,200).cuda()
        self.layer2=nn.Linear(200,64).cuda()
        self.layer3=nn.Linear(64,32).cuda()
        self.layer4=nn.Linear(32,10).cuda()
        self.smax=nn.Softmax(dim=1).cuda()

    def forward(self,x):
        x=self.vgg16(x).cuda()
        x=self.layer1(x.view(-1,512*7*7)).cuda()
        x=self.layer2(x)
        x=self.layer3(x)
        x=self.layer4(x)
        x=self.smax(x).cuda()
        return(x)

model=Transfer().cuda()
optimizer = optim.SGD(model.parameters(), lr=0.01,momentum=0.6)
criterion = nn.CrossEntropyLoss().cuda()

# Config
epochs=10
batchSize=16

# Data Loader
import torch.utils.data as data
train_data_loader = data.DataLoader(trainset, batch_size=batchSize, shuffle=False,  num_workers=1)
test_data_loader  = data.DataLoader(testset, batch_size=batchSize, shuffle=False, num_workers=1)

# Model Training
model=Transfer().cuda()
optimizer = optim.Adam(model.parameters(), lr=0.0001)
for epoch in range(10000):
    totalLoss=0
    for step, (x, y) in enumerate(train_data_loader):
        if(step <= 100):
            model.zero_grad()
            output = model(x.cuda())
            loss = criterion(output, y.cuda())
            totalLoss=totalLoss + loss.item()
            loss.backward()
            optimizer.step()
    if(epoch%1==0):
        print("Epoch {0} Step {1} TotalLoss {2}".format(epoch,step,totalLoss))

print("Cell Execution Completed")

model=model.to('cpu')
torch.save(model,"torchNonQuantizedModel")
os.path.getsize("torchNonQuantizedModel")/1e6  # 79.003

model.qconfig = torch.quantization.default_qconfig
print(model.qconfig)
torch.quantization.prepare(model, inplace=True)

for curModule in model._modules['vgg16']:
    if(type(curModule)==torch.nn.modules.activation.ReLU):
        curModule.inplace=False

# Checking
for curModule in model._modules['vgg16']:
    if(type(curModule)==torch.nn.modules.activation.ReLU):
        print(curModule.inplace)

# Quantization of the model
newModel=torch.quantization.convert(model,inplace=False)
torch.save(newModel.state_dict(), "torchQuantizedModel")
os.path.getsize("torchQuantizedModel")/1e6 # 19.771906