casmlab/NPHardEval

Potential Bugs about postprocess in BSP?

Closed this issue · 1 comments

Thanks for your great work!

In

output, reasoning = parse_xml_to_dict(llm_string)

, llm_string has been processed to dict via parse_xml_to_dict and stored in "output". However in
solution, reasoning = parse_xml_to_dict(solution)
, bsp_check function parse "output" again and assert xml elements in it. This leads to the return value of bsp_check always being False thus the result of bsp task will always be 0. Remove the code in
solution, reasoning = parse_xml_to_dict(solution)
will fix it if I understand correctly.

Meanwhile, DATA_PATH in all run_xxx.py need to be fixed. For instance in

DATA_PATH = '../../Data/GCP_Decision/'
, DATA_PATH should be '../../Data/Zeroshot/GCP_Decision/' if I understand correctly.

Thanks for figuring this out. Please try the new version we released today - it should fix these bugs. If there are more bugs spotted, please let me know.