Smaller test/val loss but lower evaluation accuracy

When I finetune llama-7b on gsm-8k with different finetuning methods. I compared the test loss and evaluation accuracy of different methods and found that one of the method has smaller test/val loss but lower evaluation accuracy. Is it reasonable?

No files or tests are named. Start by reproducing the llama-7b fine-tuning comparison on gsm-8k across the reported methods, then compare the test/validation loss with evaluation accuracy. Done means establishing whether the discrepancy is expected or identifying a reproducible implementation problem.

Written by the indexing model from the issue text.