<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>BestServe: Serving Strategies with Optimal Goodput in Collocation and Disaggregation Architectures · HackerPapers discussions</title><link>https://hackerpaper.kaipfister.de/p/2506.05871/</link><description>Collected HN comments and submissions linking this paper.</description><language>en</language><lastBuildDate>Mon, 02 Nov 2026 11:16:22 GMT</lastBuildDate><atom:link href="https://hackerpaper.kaipfister.de/p/2506.05871/feed.xml" rel="self" type="application/rss+xml" /><item><title>BestServe: Serving Strategies with Optimal Goodput in Collocation and Disaggregation Architectures — mentioned by MrCoffee7</title><link>https://news.ycombinator.com/item?id=49369323</link><guid isPermaLink="false">urn:hackerpapers:mention:2506.05871:49369323</guid><pubDate>Thu, 20 Aug 2026 01:15:27 GMT</pubDate><description>&lt;p&gt;The roofline chart on page 63 has some problems. The X and Y axis are not labeled with units such as &amp;quot;FLOP/byte.&amp;quot; Also, proper roofline charts don&amp;#x27;t have the dashed line that the chart on page 63 has - they have a only flat line on the right part of the graph and that is why they are called a &amp;quot;roofline chart&amp;quot; - see the examples in the paper: https://arxiv.org/pdf/2506.05871 . I&amp;#x27;m a bit confused also about why that roofline chart does not show the effects of different levels of quantization on the roofline max thruput, which would successively climb the performance axis to higher levels as you…&lt;/p&gt;&lt;p&gt;&lt;a href="https://hackerpaper.kaipfister.de/p/2506.05871/"&gt;Paper and discussion history&lt;/a&gt;. A mention is not an endorsement.&lt;/p&gt;</description><category>cs.LG</category></item></channel></rss>
