<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Stuart Rowlands</title><description>Tech things that interest Stuart Rowlands, founder of QuantCDN.</description><link>https://stuartrowlands.com/</link><item><title>Your model may already ship a speculative decoder. Almost nothing loads it.</title><link>https://stuartrowlands.com/posts/mtp-heads-nothing-loads/</link><guid isPermaLink="true">https://stuartrowlands.com/posts/mtp-heads-nothing-loads/</guid><description>Several popular open models carry a multi-token prediction head that most runtimes ignore. MTPLX loads them on Apple Silicon, and I have been widening the set of models it can take.</description><pubDate>Sat, 05 Sep 2026 00:00:00 GMT</pubDate></item><item><title>The sortable IDs that stopped sorting on 14 August 2026</title><link>https://stuartrowlands.com/posts/sortable-ids-stopped-sorting/</link><guid isPermaLink="true">https://stuartrowlands.com/posts/sortable-ids-stopped-sorting/</guid><description>opencode packs a 53-bit value into a 48-bit field, so its sortable IDs roll over every 795 days. The last rollover was in August. It showed up as three unrelated-looking UI bugs.</description><pubDate>Thu, 03 Sep 2026 00:00:00 GMT</pubDate></item><item><title>One guard, 110× off the verify step. Speculative decoding on MLA models in MLX.</title><link>https://stuartrowlands.com/posts/mla-absorbed-threshold/</link><guid isPermaLink="true">https://stuartrowlands.com/posts/mla-absorbed-threshold/</guid><description>Every MLA model in mlx-lm only used its fast attention path when there was exactly one query, which made speculative decoding slower than not speculating. Here is what the fix is and what it measured at.</description><pubDate>Tue, 01 Sep 2026 00:00:00 GMT</pubDate></item></channel></rss>