Why the inference time isn't changed?

[mmjwxbc]

# the latest version
self.prot_model = FAEsmForMaskedLM.from_pretrained("./esm_weights/esm2_35M").to(torch.float16) if pretrained else EsmModel(EsmConfig.from_pretrained('esm_weights/esm2_35M'))

# the old version
self.prot_model = ESMModel.from_pretrained("./esm_weights/esm2_35M").to(torch.float16) if pretrained else EsmModel(EsmConfig.from_pretrained('esm_weights/esm2_35M'))

the memory usage is also equal, and the inference time isn't changed

[pengzhangzhi]

Hi,
I'm not sure, but I'd like to know if that's bc ur 35m is too small so the diff is minimal. I suggest u run the test we provided and see the difference.