Interesting work, thanks for explaining it so lucidly! But this still involves loading three models (encoder-decoder ones mostly) in memory and running inference on them. So I don't see how this will solve the computational costs problem
Hi Pavan, firstly thanks for appreciating. So, the method works on the output dump from each of the summarization models involved. So one can very well generate outputs from them independently and not load all of them at once. The output files are now input to this function that does the ensembling, this is unlike runtime-ensemble, where loading all of the summarization models at once is a necessity.
I don’t see it’s mention in the paper. Maybe you can see in “papers with code”. Thanks for the appreciation. Yeah, I do some demo/library explainer videos as well, but yeah very limited as of now :)
![[Classic] Deep Residual Learning for Image Recognition (Paper Explained)](http://i.ytimg.com/vi/GWt6Fu05voI/mqdefault.jpg)
Interesting work, thanks for explaining it so lucidly! But this still involves loading three models (encoder-decoder ones mostly) in memory and running inference on them. So I don't see how this will solve the computational costs problem
Hi Pavan, firstly thanks for appreciating. So, the method works on the output dump from each of the summarization models involved. So one can very well generate outputs from them independently and not load all of them at once. The output files are now input to this function that does the ensembling, this is unlike runtime-ensemble, where loading all of the summarization models at once is a necessity.
Is the code available for the paper? Would love if you can also do demo of some of these papers. Thanks. Love your content!
I don’t see it’s mention in the paper. Maybe you can see in “papers with code”. Thanks for the appreciation. Yeah, I do some demo/library explainer videos as well, but yeah very limited as of now :)