Dion2: A new simple method to shrink matrix in Muon
118
Microsoft Research355 тыс
Следующее
8 часов – 2841:03Популярные
74 дня – 1 9770:52
Опубликовано 3 марта 2026, 17:59
Dion2 reduces the cost of Muon’s orthonormalization step by orthonormalizing only a small, selected submatrix at each iteration. This lightweight approach preserves Muon’s strong performance while significantly improving scalability of optimizer at scale.
GitHub Repo: github.com/microsoft/dion
This session aired on March 3, 2026, at Microsoft Research Forum, Season 2 Episode 3.
Register for the series to learn about future episodes: events.microsoft.com/flow/ms/r...
Explore all previous episodes: aka.ms/researchforumYTplaylist
GitHub Repo: github.com/microsoft/dion
This session aired on March 3, 2026, at Microsoft Research Forum, Season 2 Episode 3.
Register for the series to learn about future episodes: events.microsoft.com/flow/ms/r...
Explore all previous episodes: aka.ms/researchforumYTplaylist
Свежие видео
4 дня – 1670:35
4 дня – 5281:51
6 дней – 844 1593:12
6 дней – 1326:14
6 дней – 3 2300:37
7 дней – 33 3961:28Случайные видео
99 дней – 70 19623:25
315 дней – 69 0292:09
08.12.24 – 10 4540:20
07.07.23 – 21 9931:00
14.02.07 – 228 7871:03
8 часов – 279:55
2 дня – 1981:04
4 дня – 034:54
7 дней – 2471:40
11 дней – 1720:27
14 дней – 534 8801:14
18 дней – 45 5410:30