Hi. Thank you for your brilliant work, and I really like it, but I've met some problems during reproducing the results.
I noticed that the commands for Civil-Comments in the README are not consistent with those reported in the paper. For example, for the initial model, the learning rate is 1e-5 while it's 2e-5 in README. Batch size and Num Epoch are also different. Besides, it was said in another issue that _civil_commentsanalysis.py should be used for evaluation rather than the analysis.py as stated in README.
Could you please provide the commands with the exact same hyper-parameter settings as reported in the paper? I really appreciate your help! Thank you so much :)
Hi. Thank you for your brilliant work, and I really like it, but I've met some problems during reproducing the results.
I noticed that the commands for Civil-Comments in the README are not consistent with those reported in the paper. For example, for the initial model, the learning rate is 1e-5 while it's 2e-5 in README. Batch size and Num Epoch are also different. Besides, it was said in another issue that _civil_commentsanalysis.py should be used for evaluation rather than the analysis.py as stated in README.
Could you please provide the commands with the exact same hyper-parameter settings as reported in the paper? I really appreciate your help! Thank you so much :)