Shane Caldwell 路 Mar 29, 2026GENERATIVE AI AND FOUNDATION MODELSProofJudge: Can we align vibe-proving with human taste?Towards measuring alignment with human taste in autoformalization with judge agents.GOOGLE
Shane Caldwell 路 Jan 05, 2026MICROPROCESSORS AND CHIPSETSIntro to GPUs For the ResearcherGetting comfortable with the hardware on a quest for more MFU.
Shane Caldwell 路 Dec 12, 2025HIGH PERFORMANCE COMPUTINGAll Reduce Across the Atlantic: Bandwidth in Decentralized TrainingThe practical realities of devestatingly high communication cost in training.FRANCEUNITED STATESPRIME INTELLECT
Shane Caldwell 路 Dec 01, 2025ML OPS AND MODEL ENGINEERINGTwenty Billion Tokens of What, Exactly?Looking at the data and letting it look back at us.MICROSOFT
Shane Caldwell 路 Nov 23, 2025ML OPS AND MODEL ENGINEERINGPretraining at home: 20B tokens from 222 hours to 12Optimizing training a Llama 3.2 1B model so we can pretrain in a day without going broke.