Request for Public Test Code and Multi-GPU Evaluation Scripts #3
Description
Hi Neil,
I’m Marlin Yang’s visiting student, currently working on extending your paper and implementation.
I have successfully trained the model following the provided code, but I am having trouble reproducing the results reported in the paper. My custom test script does not fully match the original setup — the accuracy is consistently about 3% lower than reported.
It seems that the repository does not include the official test/evaluation code, and this makes it difficult to verify correctness or run multi-GPU evaluation.
Could you please consider publishing the test code (especially the evaluation script used in your experiments) or provide some guidance on the correct data format and evaluation settings?
This would greatly help me and others who are trying to reproduce and extend your work.
Thank you very much for your time and for maintaining this excellent project!
Best regards,
Marlin Yang’s Visiting Student (ZL OY)