Preprocessing with Chain-of-Thought (CoT)... is extremely time-consuming.

[drilistbox]

I changed the following in scripts/run_nuplan_preprocessing.sh:

1. Changed INCLUDE_COT=false to INCLUDE_COT=true
2. Changed CUDA_VISIBLE_DEVICES=0,1 to CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
   Then I executed bash scripts/run_nuplan_preprocessing.sh. The entire process took 10 hours.

And this is only for the mini dataset. If I switch to trainval, it would take thousands of hours.
Is this reasonable?