Get the number of output features from the last layer and remove the last layer?

[srevandro]

Hello,

I am trying to add the LoRA layer to the EfficientNet-B0 till B7 algorithms. However, I am not succeed in getting the number of output features and remove the last fully connected layer. Can you help me?

I tried the following code, but there is no "._fc" attribute on Python pre-trained algorithm.

Define EfficientNet-B0 with LoRA

class EfficientNetB0LoRA(nn.Module):

def __init__(self, num_classes, lora_rank):
    super(EfficientNetB0LoRA, self).__init__()
    # Load pre-trained EfficientNet-B0 model
    self.efficientnet_b0 = EfficientNet.from_pretrained('efficientnet-b0')
    # Get number of input features for the LoRALayer
    num_features = self.efficientnet_b0._fc.in_features
    # Replace the classifier with an identity layer
    self.efficientnet_b0._fc = nn.Identity()
    # Add LoRA layer
    self.lora = LoRALayer(num_features, num_classes, lora_rank)

def forward(self, x):
    x = self.efficientnet_b0(x)
    x = self.lora(x)
    return x