Where can I get the model architecture and the weights loaded for it?

[MotorCityCobra]

I found some weights and I think I found the code for it but things don't look right in the layers of the saved weights. This is without talking about the code in PyTorch that goes with it. Something in the weights in the VAE between the encoder and decoder don't seem to match.

It's ([8, 512, 3, 3]) conv out and then it's ([512, 4, 3, 3]) conv in. These are both Conv2D according to the code. What's going on here?

What kind of network can function with Conv2d shapes like this? 0 t 0.weight torch.Size([8, 512, 3, 3]) 1 t 0.bias torch.Size([8]) 2 t 2.weight torch.Size([512, 4, 3, 3]) 3 t 2.bias torch.Size([512])

From the weights. 97 encoder.mid.block_2.conv1.bias torch.Size([512]) 98 encoder.mid.block_2.norm2.weight torch.Size([512]) 99 encoder.mid.block_2.norm2.bias torch.Size([512]) 100 encoder.mid.block_2.conv2.weight torch.Size([512, 512, 3, 3]) 101 encoder.mid.block_2.conv2.bias torch.Size([512]) 102 encoder.norm_out.weight torch.Size([512]) 103 encoder.norm_out.bias torch.Size([512]) 104 encoder.conv_out.weight torch.Size([8, 512, 3, 3]) 105 encoder.conv_out.bias torch.Size([8]) 106 decoder.conv_in.weight torch.Size([512, 4, 3, 3]) 107 decoder.conv_in.bias torch.Size([512]) 108 decoder.mid.block_1.norm1.weight torch.Size([512]) 109 decoder.mid.block_1.norm1.bias torch.Size([512]) 110 decoder.mid.block_1.conv1.weight torch.Size([512, 512, 3, 3]) 111 decoder.mid.block_1.conv1.bias torch.Size([512]) 112 decoder.mid.block_1.norm2.weight torch.Size([512]) 113 decoder.mid.block_1.norm2.bias torch.Size([512]) 114 decoder.mid.block_1.conv2.weight torch.Size([512, 512, 3, 3]) 115 decoder.mid.block_1.conv2.bias torch.Size([512])

Maybe I can just be directed to a good file of the saved weights and a good file of the PyTorch code of the network. Thank you.