bananaml / potassium

An HTTP serving framework by Banana
Apache License 2.0
97 stars 9 forks source link

Inference is taking too much time. #34

Open sanjay-nit opened 1 year ago

sanjay-nit commented 1 year ago

I tested the same piece of code on google colab on A100 GPU and there it is taking only 0.5 sec to 1 sec. But when I try the exactly same code on banana's GPU it is taking around 6 to 7 seconds. I kept all packages use the same version as colab. But still taking that much time.

Is there anything else I should consider specific to banana? I'm expecting a quick response. Thanks!! Also I increased minimum-replicas but it didn't help. I can share the code if needed.