How to calculate flops and params

[Zzhy2000]

Hello, I would like to ask how the flops and params of your model are calculated. Looking forward to your reply.

[mcpaulgeorge]

from fvcore.nn import FlopCountAnalysis
model = Walmafa(inp_channels=3,out_channels=3,dim = 16,num_blocks = [2,3,4],heads = [1,2,4,8],ffn_expansion_factor = 2.66,bias = False,LayerNorm_type = 'WithBias',attention=True,skip = False)
model = model.cuda("cuda:2")
input_example = torch.randn(1, 3, 256, 256).cuda("cuda:2")
flops = FlopCountAnalysis(model,(input_example))
print(f'GMac:{flops.total()/(102410241024)}')

In terms of params, you can find the calculated code in train.py