Skip to content

Pull requests: microsoft/dion

New pull request New

8 Open 71 Closed

Pull requests list

Bump gram-newton-schulz to 0.1.5 (cutlass-dsl 4.5.2)

#95 opened Jun 26, 2026 by JohnLangford Contributor

Loading...

Fix AdamW LM head fallback scaling

#94 opened Jun 22, 2026 by jibby1729

Loading...

7

Add Soft-Muon softening knob for finite-Schatten-p updates

#92 opened Jun 17, 2026 by JohnLangford Contributor

Loading...

Add split_sizes param-group option + megabatch cpu overhead optimizations (2x faster cpu time)

#90 opened Jun 12, 2026 by alint77 Contributor

Loading...

13

Add Aurora optimizer for non-square matrices

#80 opened May 10, 2026 by JohnLangford Contributor

Loading...

4 of 5 tasks

8

NorMuon: opt-in nan_guard_fallback skips step on non-finite NS output

#79 opened May 8, 2026 by JohnLangford Contributor

Loading...

4 tasks done

megabatch: env-gated NaN capture wrapper around newton_schulz_func

#78 opened May 8, 2026 by JohnLangford Contributor

Loading...

5 tasks done

Dion2 with shard-independent update route added

#25 opened Jan 16, 2026 by kwangjunahn Contributor

Loading...

1

ProTip! Follow long discussions with comments:>50.