Trouble achieving results in paper

[0k9d0h1]

Hello, I attempted to replicate the results of the DAS baseline using your code, but encountered discrepancies with the results presented in the paper. I posttrained the model with a batch size of 16 due to resource limitations, while the README file suggests a batch size of 62. Below are the F1 scores and accuracies obtained:

Acc:
0.8679 0.0000 0.0000 0.0000 0.0000 0.0000
0.8616 0.7387 0.0000 0.0000 0.0000 0.0000
0.8616 0.7387 0.6889 0.0000 0.0000 0.0000
0.8616 0.7387 0.6889 0.8246 0.0000 0.0000
0.8616 0.7387 0.6889 0.8246 0.6880 0.0000
0.8616 0.7387 0.6889 0.8246 0.6880 0.8051

F1:
0.7925 0.0000 0.0000 0.0000 0.0000 0.0000
0.7858 0.7002 0.0000 0.0000 0.0000 0.0000
0.7858 0.7002 0.5529 0.0000 0.0000 0.0000
0.7858 0.7002 0.5529 0.7909 0.0000 0.0000
0.7858 0.7002 0.5529 0.7909 0.6880 0.0000
0.7858 0.7002 0.5529 0.7909 0.6880 0.6751

Could the discrepancy be related to the difference in batch size, or am I missing something else? Your insights would be appreciated. Thank you.