A classic OOM problem

[zwj2023zwj]

Dear Author,

I am currently a student researching existing methods in the POI recommendation domain. I am highly interested in your team's proposed LLM4POI model, which I find highly innovative. I am attempting to reproduce your model but have encountered a challenging issue.

When following the commands provided in the markdown file to run traj_sim.py and construct key-query pairs, I encountered a classic OOM problem. However, my A100 GPUs have sufficient memory, and I have access to multiple cards. I suspect the issue might be due to GPU memory fragmentation.

Could you kindly provide some suggestions or solutions? I am genuinely puzzled and would greatly appreciate your help.

Thank you very much!