罗崚骁 (Luo, Lingxiao)

> > > > 但是点击图标不就是运行对应脚本吗，为什么直接运行脚本可以启动，点击图标却不行？ > > 可能脚本用的 wine5，图表用得wine6？ 有办法针对这个可能验证一下吗？

@ZikangZhou Hi, although this is an old thread, I want to share my thoughts here since there's still an open issue pointing here, I hope this doesn't cause any inconvenience....

> Hi, I`m curious about how the `g_loss` is derived without using a sigmoid or softplus function. After searching several hours, I cannot find any reference for using logits directly...

> @TomoshibiAkira @crowsonkb Thanks for sharing your insight, I am working on it recently and will update if concrete conclusion is reached. @fnzhan Congratulations on [your article](https://arxiv.org/abs/2303.06424) being accepted by...

Hello, will Hypothesis become compatible with PDF.js of higher versions in the future? It seems non-trivial to use a lower version of PDF.js in Firefox (https://github.com/mozilla/pdf.js/issues/10500).

@danthe3rd It seems that there's still no [conda build](https://anaconda.org/xformers/xformers/files?version=0.0.20) for Python 3.11. Could you please take a look?

Hi @danthe3rd , may I ask what if I do want padding sequences? I'm working on vision tasks, the lengths of sequence do not vary a lot. Currently, I'm creating...

@danthe3rd My use case is as follows: Suppose there are three images in a batch with shapes $(h_1, w_1)$, $(h_2, w_2)$, and $(h_3, w_3)$. After applying the patch embedding of...

Thank you for your suggestion. Yes, I've checked the `BlockDiagonalMask`. But my concern is: will it have much more computation cost compared to padding when the lengths of the sequence...

Wow that's cool, thanks for clarification!