Hi there! I've been trying to reproduce the base model performance reported in the paper using your code and setup, but my results are consistently lower than the stated metrics. I followed all the steps in the README, checked hyperparameters, and verified the environment setup, but can't seem to match the baseline.
Could there be any missing details in the documentation—like specific version constraints, unmentioned preprocessing steps, or hidden configs—that I might have overlooked? Would really appreciate any pointers to help troubleshoot! Thanks!
Hi there! I've been trying to reproduce the base model performance reported in the paper using your code and setup, but my results are consistently lower than the stated metrics. I followed all the steps in the README, checked hyperparameters, and verified the environment setup, but can't seem to match the baseline.
Could there be any missing details in the documentation—like specific version constraints, unmentioned preprocessing steps, or hidden configs—that I might have overlooked? Would really appreciate any pointers to help troubleshoot! Thanks!