Agustin Dal Lago

Results 2 comments of Agustin Dal Lago

5,1,5 doesn't fail cause the 5*weight gets internally rounded down to zero... matching "by luck" the original implementation output. The internal tests do a weird padding and refactor the input...

Hi @shoyer, thanks for your answer! I apologize for my description, it may have not been as clear as it sounded in my mind. From my experiments it seems to...