Replication of results on the OPT-1.3B model for the GLUE dataset

[Xingke-yang]

Hi, thanks for the great code. I recently tried to use your open glue_qst script to replicate the test results from the paper on the OPT-1.3B model on GLUE data. I didn't change your program too much, just set the model to OPT-1.3B, but I can't seem to achieve the accuracy results in the paper, the difference is about 10%. Are there any other changes that need to be made to the current code? Thanks a lot! @YouAreSpecialToMe

[YouAreSpecialToMe]

Hi, sorry for that. We conducted experiments based on this script. Can you maybe try some other hyperparameters since we conducted further experiments and probably modified the script after submitting the paper? Thanks!

[wzy1026wzy]

Hi, sorry for that. We conducted experiments based on this script. Can you maybe try some other hyperparameters since we conducted further experiments and probably modified the script after submitting the paper? Thanks!

Hi, I also encountered similar issues when trying to replicate the experimental results. I achieved relatively close results on the SST-2 task, but the performance on other tasks was not satisfactory. Could you please provide the hyperparameters used in the experiments? Your assistance would be greatly appreciated. Thank you!