Meta tags:
description=Harold Benoit - Personal Website;

Headings (most frequently used words):

harold, benoit, news, fun, stuff,

Text of the page (most frequently used words):
the (9), 2025 (5), 2024 (4), and (3), #harold (3), benoit (3), models (3), opinions (2), pretraining (2), thinking (2), model (2), accepted (2), training (2), diffusion (2), june (2), microsoft (2), blog (2), notes (2), name (2), expressed, are, solely, own, not, express, views, employer, all, rights, reserved, new, version, original, currently, member, technical, staff, working, multimodal, have, where, write, down, almost, everything, learn, for, more, polished, thoughts, nowadays, tend, spend, time, about, compute, efficiency, hardware, aware, design, multimodality, iclr, diversification, methods, paper, january, finished, 1st, place, llm, hackathon, april, implemented, from, scratch, protein, structure, tmlr, controlled, data, generation, with, march, released, one, main, contributors, lfm2, july, joined, december, tech, report, out, worked, mai, 2026, news, fun, stuff, stillness, night, jobs, hum, quietly, across, gpus, mind, finds, rare, serenity, knowing, work, carries, without, you, email, first, last, hotmail, scholar, linkedin, github, papers, home,

Text of the page (random words):
harold benoit home blog notes papers harold benoit github x linkedin cv g scholar email my first name _ my last name hotmail ch in the stillness of the night as jobs hum quietly across gpus the mind finds a rare serenity knowing the work carries on without you i m currently a member of technical staff at microsoft ai working on pretraining multimodal models i have notes where i write down almost everything i learn and a blog for more polished thoughts nowadays i tend to spend time thinking about compute efficiency hardware aware model design and multimodality news fun stuff june 2026 mai thinking 1 tech report is out worked on the pretraining december 2025 joined microsoft ai july 2025 released the lfm2 models i am one of the main contributors march 2025 controlled training data generation with diffusion models accepted at tmlr 2025 june 2024 implemented from scratch a protein structure diffusion model april 2024 finished 1st place in the llm training hackathon january 2024 diversification methods paper accepted at iclr 2024 original new sd 1 5 version of me 2025 harold benoit all rights reserved opinions expressed are solely my own and do not express the views or opinions of my employer