microsoft

/

deberta-v3-base

like 253

Follow

Microsoft 5,075

Comparing inference time between deberta-v3-base and say bert-base-cased, I am getting that deberta is significantly slower. This is of course on the same set and same machine. (~50% increase).

model_name_or_path_bert-base-cased/dev_prediction_time.txt, Total prediction time = 73.53624510765076
model_name_or_path_microsoft/deberta-v3-base/dev_prediction_time.txt, Total prediction time = 111.61959910392761

Is that expected, or am I doing something wrong?