search

KaLiMaLi555 / model_extraction_interiit

1 stars 0 forks source link

Convit #18

Open anime-sh opened 2 years ago

anime-sh commented 2 years ago

implementation of convit

anime-sh commented 2 years ago

this might need a lot of parameter tuning, in this code the embedding dimension for each head (9 heads in total) is 48, 48*9 is less than 600 (num classes for k600) so dekhlena thora..... However itne par hi cuda out of memory araha