why the params size of b0 model is 20M rather than 5.3M

[lightfate]

Hi, @lukemelas ~
I'm so excited about your code.
I have a question. You said that the Params size of the b0 model is 5.3M，but the size of pretrained model what you offered is 20.3M,
and then I test it through torchsummary, which shows that Params size (MB): 20.17, why?

---

#torchsummary code:
from efficientnet_pytorch import EfficientNet
modelBody= EfficientNet.from_pretrained('efficientnet-b2').to(device)
from torchsummary import summary
summary(modelBody, input_size=(3, 224, 224))

---

#result
Total params: 5,288,548
Trainable params: 5,288,548
Non-trainable params: 0
Input size (MB): 0.57
Forward/backward pass size (MB): 107.31
Params size (MB): 20.17
Estimated Total Size (MB): 128.06

---

[lukemelas]

5.3M refers to the number of parameters, whereas 20 refers to the size of the params in megabytes. It makes sense that it's around 4 times the number of params.

[lukemelas]

Closing this.

[EtheneXiang]

torchsummary

one para's type is float32 = 4 * int8 = 32bit
so 5,288,548 * 4 = 20.17MB
that's all