# "Attention Is All You Need" model implementation using dl4j

**URL:** <https://community.konduit.ai/t/attention-is-all-you-need-model-implementation-using-dl4j/2563>\
**Category:** DL4J\
**Created:** [May 18, 2023, 7:06am UTC](https://community.konduit.ai/t/attention-is-all-you-need-model-implementation-using-dl4j/2563 "2023-05-18T07:06:23Z")\
**Posts on this page:** 1\
**Showing post:** 60

<div class="post-metadata">

**Author:** ![adonnini](https://avatars.discourse-cdn.com/v4/letter/a/5f8ce5/32.png) [@adonnini](https://community.konduit.ai/u/adonnini)\
**Post date:** [June 9, 2023, 5:40am UTC](https://community.konduit.ai/t/attention-is-all-you-need-model-implementation-using-dl4j/2563/60 "2023-06-09T05:40:25Z")

</div>

@agibsonccc would the suggestion you made in

> [@Reduce rank and set values according to 3rd dimension](https://community.konduit.ai/t/reduce-rank-and-set-values-according-to-3rd-dimension/2029):
>
> [[[1,0,0,0],[0,1,0,0],[0,0,1,0]], [[1,0,0,0],[0,1,0,0],[0,0,1,0]] ] How to get below result from above SDVariable? Reduce 3rd dimension, and the value = 0.1 when 1 exists in the first or 2rd position, otherwise value = 1. [[0.1,0.1,1], [0.1,0.1,1] ]

> [@Reduce rank and set values according to 3rd dimension](https://community.konduit.ai/t/reduce-rank-and-set-values-according-to-3rd-dimension/2029/10):
>
> is this for weighted loss? If not why not make that a placeholder?

apply in my case? May I ask what you meant by “make that a placeholder”? I am not sure I understand.

Thanks

---

_[View the full topic](https://community.konduit.ai/t/attention-is-all-you-need-model-implementation-using-dl4j/2563)._