Humaneval Results of LLADA Much Lower Than Reported in Paper

[Lyn-Lucy]

I tested LLADA on the HumanEval benchmark baseline following the instructions in the repository. However, the performance I obtained is significantly lower than the results reported in the paper.

Reported in paper: 49.4

My results:

Could you clarify if there are additional settings, evaluation scripts, or specific checkpoints required to reproduce the reported results?

Thanks in advance!