Jekyll2026-10-18T08:05:11+00:00https://kairong.me/feed.xmlKairong Luo / 罗开荣Kairong Luo is a Ph.D. student at Tsinghua University studying data, optimization, and efficient language-model pretraining.Kairong Luoluokr2002@outlook.comPuro-2B: How Can a Poor Lab Pretrain a Model from Scratch?2026-10-14T00:00:00+00:002026-10-14T00:00:00+00:00https://kairong.me/blog/puro-2bThe Puro-2B story, from hardware and FP8 training to data curation and curriculum model averaging. Includes the original figures, interactive experiments, and both Chinese and English text.