Why not set a fix seed? and the reproduce results is unstable. #10
Description
Hi, thanks for your wonderful job and sharing source codes. I have cloned your codes and tried to reproduce the results of gpt-j-6b for tabular data. However, I ran mutiple times and found that it returned mutiple different results. It's resonable because the seed is not fixed, but the results diverge significantly across different seed, which means the model is unstable). Could you please tell me why this is happening? How you run the results is shown in the paper?