This message was deleted.
# ask-for-helps
Slackbot
10/27/2022, 3:44 PMThis message was deleted.
m
Mike Kuhlen
10/27/2022, 4:47 PMI have no idea what kind of payload you’re dealing with, but I’ll say that I’ve had good success with using parquet (rather than JSON) for passing in a large payload of tabular data (a DataFrame).
(DataFrame -> Parquet -> DataFrame) is much faster than (DataFrame -> JSON -> DataFrame).
🔥 1
➕ 1
s
Shiva Charan Velichala
10/27/2022, 7:20 PMgotcha, will give that a try
s
Sean
11/02/2022, 3:26 AM@Shiva Charan Velichala are you still converting JSON to Numpy for your input?
Sean
11/02/2022, 3:30 AMCould you please also share the absolute latency numbers, both inference latency and serialization latency?
j
Jiang
11/02/2022, 5:11 AMUsing json as an intermediate data format for large dataframes would theoretically lead to conversions that consume very many resources and are slow.
Jiang
11/02/2022, 5:12 AMSo we'd better avoid such an architecture.