Abheesht

Results 106 comments of Abheesht

Hello! Have the pretrained models for TK been released? If not, could you please share them? Would be of great help! :)

@mattdangerw, I'd like to take this up. Thanks!

Hey, @innat! Are you still working on this? If not, I'd like to work on this. Thanks! :)

Thanks, @innat! @mattdangerw, I can take this up (if we want this layer) :)

Awesome! Thanks, @chenmoneygithub. Will add it to the doc :)

Hello, @mattdangerw! As discussed, I'd be glad to take this up. Thanks! :)

Linking the PR here: https://github.com/keras-team/keras-io/pull/890

Hello, @chenmoneygithub! I think the reason is as follows: XLNet has multiple factorisation orders since it permutes the input sequence. Suppose our input text is [1, 2, 3, 4], and...

This figure explains it well: ![image](https://user-images.githubusercontent.com/31360468/165980813-36e86e66-e202-4363-af38-68b6ac2d770c.png)

A more concrete explanation: ![image](https://user-images.githubusercontent.com/31360468/165983856-a686c596-bcc1-4d37-97d0-4be025afc9b7.png) https://www.borealisai.com/en/blog/understanding-xlnet/