Increaseing the speed of execution

[Lahari-77]

Hi,
I am using nvidia-smi driver version 515
Cuda tool kit :11.7
Paddlepaddlegpu:2.4.2
I have a program which has 6 loops.Each time inside the loop paddleocr has to extract text.
Even after enabling gpu=True.Each loop i staking 2 seconds.Resulting in 12 seconds.Which is very bad according to the performance parameters.What can be done do cut down the 12 sec to atleast 2 3 seconds.

[tzktok]

find any solution @Lahari-77 ?

[tzktok]

I have initialize the model first itself..bt every time model loads taks time.. @Lahari-77

[GreatV]

Try using the Paddle dev version.

[tzktok]

Try using the Paddle dev version.

what's the different between stable nd dev version of paddle? @GreatV

[GreatV]

@tzktok https://www.paddlepaddle.org.cn/en/install/quick?docurl=/documentation/docs/en/develop/install/pip/linux-pip_en.html

[GreatV]

The new version of paddle has many more new features and fixes.

[Lahari-77]

May be u r loading and running at the same instance.Try loading the model while starting tjhe server and then try paddle.

[tzktok]

May be u r loading and running at the same instance.Try loading the model while starting tjhe server and then try paddle.

paddleocr start time in gpu is almost 6sec..how to reduce this time.. @Lahari-77 @GreatV