Hey @Reacher,

Congrats and nice solution overview.

We had the same idea for the Reverse self-distillation part. But since pseudo-labeling had failed on our end, we thought this would bring to much noise. Great to see it worked for you.

Great approach ..But it somehow didnt work for me ..j used 0.5 * original + 0.5*oof soft label ..the CV was good ..LB didnt agree ..

Same for test pseudolabel ..jt failed ..