torch.compile (inductor)

import torch
import torch.nn as nn
import torch.nn.functional as F

class Model(nn.Module):
    def __init__(self, num_classes=1000):
        """
        EfficientNetB0 architecture implementation in PyTorch.

        :param num_classes: The number of output classes (default is 1000 for ImageNet).
        """
        super(Model, self).__init__()
        
        # Initial convolutional layer
        self.conv1 = nn.Conv2d(3, 32, kernel_size=3, stride=2, padding=1, bias=False)
        self.bn1 = nn.BatchNorm2d(32)
        
        # MBConv blocks
        self.blocks = nn.Sequential(
            # MBConv1 (32, 16, 1, 1)
            MBConv(32, 16, kernel_size=3, stride=1, expand_ratio=1),
            # MBConv6 (16, 24, 2, 6)
            MBConv(16, 24, kernel_size=3, stride=2, expand_ratio=6),
            # MBConv6 (24, 24, 1, 6)
            MBConv(24, 24, kernel_size=3, stride=1, expand_ratio=6),
            # MBConv6 (24, 40, 2, 6)
            MBConv(24, 40, kernel_size=5, stride=2, expand_ratio=6),
            # MBConv6 (40, 40, 1, 6)
            MBConv(40, 40, kernel_size=5, stride=1, expand_ratio=6),
            # MBConv6 (40, 80, 2, 6)
            MBConv(40, 80, kernel_size=3, stride=2, expand_ratio=6),
            # MBConv6 (80, 80, 1, 6)
            MBConv(80, 80, kernel_size=3, stride=1, expand_ratio=6),
            # MBConv6 (80, 112, 1, 6)
            MBConv(80, 112, kernel_size=5, stride=1, expand_ratio=6),
            # MBConv6 (112, 112, 1, 6)
            MBConv(112, 112, kernel_size=5, stride=1, expand_ratio=6),
            # MBConv6 (112, 192, 2, 6)
            MBConv(112, 192, kernel_size=5, stride=2, expand_ratio=6),
            # MBConv6 (192, 192, 1, 6)
            MBConv(192, 192, kernel_size=5, stride=1, expand_ratio=6),
            # MBConv6 (192, 192, 1, 6)
            MBConv(192, 192, kernel_size=5, stride=1, expand_ratio=6),
            # MBConv6 (192, 320, 1, 6)
            MBConv(192, 320, kernel_size=3, stride=1, expand_ratio=6)
        )
        
        # Final convolutional layer
        self.conv2 = nn.Conv2d(320, 1280, kernel_size=1, stride=1, padding=0, bias=False)
        self.bn2 = nn.BatchNorm2d(1280)
        
        # Fully connected layer
        self.fc = nn.Linear(1280, num_classes)
    
    def forward(self, x):
        """
        Forward pass of the EfficientNetB0 model.

        :param x: The input tensor, shape (batch_size, 3, 224, 224)
        :return: The output tensor, shape (batch_size, num_classes)
        """
        x = F.relu(self.bn1(self.conv1(x)))
        x = self.blocks(x)
        x = F.relu(self.bn2(self.conv2(x)))
        x = F.adaptive_avg_pool2d(x, (1, 1))
        x = x.view(x.size(0), -1)
        x = self.fc(x)
        return x

class MBConv(nn.Module):
    def __init__(self, in_channels, out_channels, kernel_size, stride, expand_ratio):
        """
        MBConv block implementation.

        :param in_channels: Number of input channels.
        :param out_channels: Number of output channels.
        :param kernel_size: Kernel size for the depthwise convolution.
        :param stride: Stride for the depthwise convolution.
        :param expand_ratio: Expansion ratio for the intermediate channels.
        """
        super(MBConv, self).__init__()
        
        self.use_residual = (stride == 1 and in_channels == out_channels)
        hidden_dim = in_channels * expand_ratio
        
        if expand_ratio != 1:
            self.expand_conv = nn.Sequential(
                nn.Conv2d(in_channels, hidden_dim, kernel_size=1, stride=1, padding=0, bias=False),
                nn.BatchNorm2d(hidden_dim),
                nn.ReLU6(inplace=True)
            )
        
        self.depthwise_conv = nn.Sequential(
            nn.Conv2d(hidden_dim, hidden_dim, kernel_size=kernel_size, stride=stride, padding=(kernel_size-1)//2, groups=hidden_dim, bias=False),
            nn.BatchNorm2d(hidden_dim),
            nn.ReLU6(inplace=True)
        )
        
        self.project_conv = nn.Sequential(
            nn.Conv2d(hidden_dim, out_channels, kernel_size=1, stride=1, padding=0, bias=False),
            nn.BatchNorm2d(out_channels)
        )
    
    def forward(self, x):
        """
        Forward pass of the MBConv block.

        :param x: The input tensor, shape (batch_size, in_channels, H, W)
        :return: The output tensor, shape (batch_size, out_channels, H', W')
        """
        identity = x
        
        if hasattr(self, 'expand_conv'):
            x = self.expand_conv(x)
        
        x = self.depthwise_conv(x)
        x = self.project_conv(x)
        
        if self.use_residual:
            x += identity
        
        return x

# Test code
batch_size = 10
num_classes = 1000

def get_inputs():
    return [torch.rand(batch_size, 3, 224, 224)]

def get_init_inputs():
    return [num_classes]