TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

Tensor evolution: A framework for fast tensor computations using recurrences

53 点作者 matt_d4 个月前

4 条评论

sakras4 个月前
&gt; Tensor Evolution (TeV)<p>Oh no, I&#x27;m never not going to read this as &quot;tera-electron Volts&quot;. I wish they picked a different acronym!
评论 #43098160 未加载
thomasahle4 个月前
This is neat! I&#x27;ve been thinking about how to best express recursion with diagrams for tensorgrad (<a href="https:&#x2F;&#x2F;github.com&#x2F;thomasahle&#x2F;tensorgrad">https:&#x2F;&#x2F;github.com&#x2F;thomasahle&#x2F;tensorgrad</a>) and this might just be it.<p>So much computation these days are done on tensors, mostly for easy parallelism reasons. But we can&#x27;t get completely rid of recursion. Great to have work like this making them more computationally compatible!
pertymcpert4 个月前
I don&#x27;t see a lot of use for this since it only extends element-wise operations and not things like multiplication which are the more interesting operations on tensors. Yeah you can model tensors in the compiler with this but no loops in the compiler are actually doing this. Unlike SCEV which is used for almost every loop.
评论 #43098513 未加载
almostgotcaught4 个月前
lol who is interested in this on hn? this has nothing to do with frameworks or training or inference or whatever. this is purely about &quot;when can i infer statically XYZ property about tensor constants in a program&quot;. mind you, <i>small tensor constants</i>, because this thing is effectively performing actual tensor arithmetic and you don&#x27;t want <i>slow tensor arithmetic</i> in your compiler.<p>also FYI they&#x27;re not magically calculating closed form solutions to recurrence relations here (that&#x27;s not possible) - they&#x27;re basically just iterating the loop arithmetic.
评论 #43100818 未加载
评论 #43095479 未加载
评论 #43098403 未加载
评论 #43095611 未加载