mdistiller
mdistiller copied to clipboard
Can you provide w-28-2, w-16-4, w-28-4 as teacher models?
I want to cite Table 11 data in my paper, but I found that the teacher link in the readme doesn't have the models w-28-2, w-16-4, w-28-4? Is it possible to upload them? (It's very very important for me.)
Can You also upload the models in Table12 if it is convenient for me to do so? Thank You!!!
By the way, can you provide the code for Figure-2?
Hi, I would really like to know how the experiments are set up in Tabel-1? Is it 0.9 * NCKD + 0.1 * CE and 0.9 * TCKD + 0.1 * CE? and the temperature is 4 ?
Hi @JinYu1998 , we release our trained teacher weights of the three models you mentioned in https://github.com/sunshangquan/logit-standardization-KD/releases/tag/extra_teacher_weights . Hope it could help you.