search

yfzhang114 / SliME

✨✨Beyond LLaVA-HD: Diving into High-Resolution Large Multimodal Models
Apache License 2.0
139 stars 7 forks source link

Question about the codes #12

Open TitleZ99 opened 2 weeks ago

TitleZ99 commented 2 weeks ago

1730991040157 Thanks a lot for publicizing such a high quality work, I'm a bit confused about performing the softmax_with_temperature operation twice in the code shown in the figure. Why is it necessary to do it twice and with different temperature coefficients? Looking forward to your reply.

yfzhang114 commented 2 weeks ago

The first line should be commented out, we will modify this code

TitleZ99 commented 2 weeks ago

Thanks for the prompt reply, very enlightening work, I'm still learning it.