https://hackerpaper.kaipfister.de/p/2410.14649/Collected HN comments and submissions linking this paper.enMon, 02 Nov 2026 11:16:22 GMThttps://news.ycombinator.com/item?id=47524161urn:hackerpapers:mention:2410.14649:47524161Wed, 25 Mar 2026 22:32:03 GMT<p>This is a very well established idea. It&#x27;s called dynamic quantization. Vary the quantization bit-width (or skip quantization altogether) on a layer by layer basis, using a calibration dataset. EvoPress is the first time that comes to my mind, when I think of dynamic quantization. https://arxiv.org/abs/2410.14649</p><p><a href="https://hackerpaper.kaipfister.de/p/2410.14649/">Paper and discussion history</a>. A mention is not an endorsement.</p>cs.LG