<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Reproductions — Frontier Checkpoint</title><description>A living, friendly logbook of reproduction efforts. We run the paper, attach the harness, seeds, and hardware, and report clearly what we found — so you can see exactly how a result holds together, build on solid ground, and reuse the work yourself.</description><link>https://frontiercheckpoint.com/</link><item><title>Reproducing the nanoGPT Speedrun: What Actually Moves the Loss Curve</title><link>https://frontiercheckpoint.com/reproductions/reproducing-nanogpt-speedrun/</link><guid isPermaLink="true">https://frontiercheckpoint.com/reproductions/reproducing-nanogpt-speedrun/</guid><description>[In the lab] On a small config the direction reproduces — with the architecture fixed, Muon reaches the 3.28 FineWeb target in markedly fewer optimizer steps than AdamW — but the headline wall-clock record is hardware-locked to 8×H100, and our reference-node run is still pending.</description><pubDate>Mon, 15 Jun 2026 00:00:00 GMT</pubDate><category>Reproductions</category><category>reproducibility</category><category>pretraining</category><category>optimization</category><category>distributed-training</category><author>editors@frontiercheckpoint.com</author></item></channel></rss>