Gpuburnout-3b on GPUburnout | Jun Parkhttps://gpuburnout.com/tags/gpuburnout-3b/Recent content in Gpuburnout-3b on GPUburnout | Jun ParkGPUburnout | Jun Parkhttps://gpuburnout.com/images/og-default.pnghttps://gpuburnout.com/images/og-default.pngHugo -- 0.155.2en-usSun, 19 Apr 2026 00:00:00 +0000Nothing Happened for 75,000 Steps and It Was Glorioushttps://gpuburnout.com/posts/s5-ch3-nothing-happened/Sun, 19 Apr 2026 00:00:00 +0000https://gpuburnout.com/posts/s5-ch3-nothing-happened/Five days of training. One small architectural fix. Zero crises. The loss curve I had wanted for two seasons.My Code Agent Said It Was a Moose. I Said No. It Was a Moose.https://gpuburnout.com/posts/s5-ch2-moose/Sun, 12 Apr 2026 00:00:00 +0000https://gpuburnout.com/posts/s5-ch2-moose/Twelve hours of debugging, five wrong suspects, one network filesystem, and a lesson in trusting the diagnostic.I Have an A100. I Have 528 Shards of Data. I Cannot Combine Them.https://gpuburnout.com/posts/s5-ch1-528-shards/Tue, 07 Apr 2026 00:00:00 +0000https://gpuburnout.com/posts/s5-ch1-528-shards/Three days. Four GPUs. Three datacenters. Zero training tokens.