← Back to PeopleLKResearcherLuo KairongLinkedInWebsiteStudent at Tsinghua University2Mentions2Articles1OutletsArticlesMost recent firstMentioned InReferenced in coverageNew transformer trick cuts training compute by up to 18% by looping middle layers twicearXivpositiveSep 4, 2026“Researcher who contributed to SMELT paper on looped transformers cutting training compute.”Open-source 2B model matches Qwen on consumer GPUs for under $7KarXivpositiveSep 2, 2026“Co-author of Puro-2B open-source model trained cost-efficiently on consumer GPUs.”