About VQA mlm loss - Githubissues

salesforce / ALBEF

Code for ALBEF: a new vision-language pre-training method

BSD 3-Clause "New" or "Revised" License

1.46k stars 193 forks source link

About VQA mlm loss #90

Closed katrina433 closed 2 years ago

katrina433 commented 2 years ago

Hello!

I have a question about the mlm loss computation in model_vqa.py. It seems like alpha is never used in forward, so the answer_output is independent of logits_m. Is this the intended behavior?

LiJunnan1992 commented 2 years ago

Hi, thanks for pointing this out. The code has been updated now to use alpha when self.distill=True.