lanyunshi/Multi-hopComplexKBQA

machine resource used in the experiment

Closed this issue · 2 comments

xwjim commented

Hi. My lab is short of gpus. I want to know what is the machine resource used in the experiment and the time consuming.

lanyunshi commented

Hi,

Our GPU config is as follows:

image

If you don't have a large memory in your GPU, you can slightly tune the parameter args.train_limit_number to fit your machine.

Regarding the training time, it costs several hours to train on WBQ and CQ datasets but it costs around 2 weeks on CWQ dataset the first time I trained on CWQ, because the code is querying the database all the time. After I collected a large KB cache, later on, it costs only 3-4 days to train on the CWQ dataset. Also the training time is different based on your database and machine configurations.

  • 👍1
xwjim commented

Hi,

Our GPU config is as follows:

image

If you don't have a large memory in your GPU, you can slightly tune the parameter args.train_limit_number to fit your machine.

Regarding the training time, it costs several hours to train on WBQ and CQ datasets but it costs around 2 weeks on CWQ dataset the first time I trained on CWQ, because the code is querying the database all the time. After I collected a large KB cache, later on, it costs only 3-4 days to train on the CWQ dataset. Also the training time is different based on your database and machine configurations.

Thank you