GenZ-LLM-Analyzer icon indicating copy to clipboard operation
GenZ-LLM-Analyzer copied to clipboard

[bug] incorrect unit in funcion get_offload_system when calculating new_offchip_BW

Open troore opened this issue 4 months ago • 0 comments

Hi @abhibambhaniya,

I found one suspicious bug in function get_offload_system, in file GenZ/LLM_inference/utils.py. The unit of total_memory_req passed to this function seems to be MB, while total_device_memory is clearly GB as commented by the first line of this function.

It was found when I created a case to show how offload would affect performance, and the offloaded memory size always dominates. I made a quick fixing by simply dividing total_memory_req by 1024, as did on total_device_memory.

Could you help double check?

Thanks, troore

troore avatar Oct 09 '24 12:10 troore