Dion optimizer algorithm
active 2025-08-02 → 2026-07-27 (UTC)
Activity over time
Daily event counts in the loaded window
Line chart, 360 days from 2025-08-02 to 2026-07-27. Pushes: 69 total, peak 10 in a day. Pull requests: 29 total, peak 6 in a day. Issues: 7 total, peak 1 in a day. Comments: 76 total, peak 12 in a day. Stars: 264 total, peak 94 in a day.
- Pushes
- Pull requests
- Issues
- Comments
- Stars
Stars, PRs, issues and forks are under-captured in the later part of this window. GH Archive progressively stopped capturing non-push events during 2026 — −95% or worse by the end of the window. Every series here except Pushes fades for that reason, so a decline above reflects the archive, not this repository. Pushes stay reliable throughout, so read them, and the contributor counts derived from them, as the real signal. Data health has the measurements.
Top contributors
Pushes, PRs, issues, reviews and comments — stars and forks excluded, so this is contribution rather than popularity
| Contributor | Contributions | Pushes | PRs | Comments |
|---|---|---|---|---|
| JohnLangford | 82 | 49 | 13 | 19 |
| byronxu99 | 49 | 2 | 2 | 22 |
| kwangjunahn | 32 | 17 | 1 | 7 |
| alint77 | 7 | 0 | 2 | 5 |
| NoahAmsel | 7 | 0 | 6 | 0 |
| thib-s | 6 | 0 | 1 | 3 |
| sippycoder | 4 | 0 | 0 | 3 |
| eric-tc-wong | 4 | 0 | 0 | 1 |
| microsoft-github-policy-service[bot] | 3 | 0 | 0 | 3 |
| zichongli5 | 3 | 0 | 1 | 1 |
| samsja | 3 | 0 | 2 | 1 |
| lalo | 2 | 0 | 0 | 1 |
| not-lain | 2 | 0 | 0 | 1 |
| ad8e | 2 | 0 | 0 | 1 |
| anhnguyen-namitech | 2 | 0 | 0 | 1 |
| segyges | 2 | 0 | 0 | 2 |
| shermansiu | 2 | 0 | 0 | 2 |
| semran9 | 1 | 0 | 0 | 0 |
| gbouras13 | 1 | 0 | 0 | 1 |
| gagmag | 1 | 1 | 0 | 0 |
Recent activity
Latest issues, pull requests and releases
- Issue comment#105JohnLangford2026-07-16 15:16Batch the Dion2/NorDion2 megabatch per-matrix scatter loops (bit-exact host reduction)
- Issue comment#97JohnLangford2026-07-02 01:48Add selection_scope to Dion2/NorDion2 (fix FSDP2 padding regression)
- Issue comment#90alint772026-06-16 05:59Add split_sizes param-group option for per-block Newton-Schulz on fused weights
- Pull request#86JohnLangford2026-06-06 20:34
- Pull request#85JohnLangford2026-06-06 20:25
- Issue comment#82ayfeng232026-05-29 03:03Add NorDion2 optimizer combining NorMuon and Dion2 paths
- Issue comment#81alint772026-05-14 21:20Add Triton kernel for dion2 post-orthogonalize
- Issue comment#71JohnLangford2026-05-05 01:04Add: MuonH (Hyperball optimizer) + add option for 'neuron' or 'short_axis' normalization for norMuon and MuonH
- Issue comment#71JohnLangford2026-05-04 01:53Add: MuonH (Hyperball optimizer) + add option for 'neuron' or 'short_axis' normalization for norMuon and MuonH
- Issue comment#71alint772026-05-02 23:07Add: MuonH (Hyperball optimizer) + add option for 'neuron' or 'short_axis' normalization for norMuon and MuonH
- Issue comment#71JohnLangford2026-05-02 23:07Add: MuonH (Hyperball optimizer) + add option for 'neuron' or 'short_axis' normalization for norMuon and MuonH
- Pull request#73NoahAmsel2026-05-02 19:59
- Issue comment#73JohnLangford2026-05-02 19:49Pin quack-kernels dependency and upgrade gram-newton-schulz to 0.1.4
- Pull request#73NoahAmsel2026-05-02 19:21
- Pull request#36JohnLangford2026-04-29 11:42
- Issue comment#68JohnLangford2026-04-28 20:33Work around torch.compile inductor bug for Dion2 3D tensor tests
- Issue comment#61gbouras132026-04-21 00:55normuon ieee error
- Issue comment#11JohnLangford2026-04-18 20:58Lightning precision in orthogonalize()
- Issue comment#18JohnLangford2026-04-18 20:57Dion2 fails to do a distributed check point
- Issue comment#22JohnLangford2026-04-18 20:56Dion2 Conv1D error
- Issue#61JohnLangford2026-04-18 20:55normuon ieee error
- Issue comment#61JohnLangford2026-04-18 20:55normuon ieee error
- Issue comment#61JohnLangford2026-04-18 20:46normuon ieee error
- Pull request#64JohnLangford2026-04-18 20:37
- Issue comment#63JohnLangford2026-04-17 23:29Fused scalar_opts adamw: replace foreach with torch._fused_adamw_ (10x fewer kernel launches)
Totals cover only the window loaded into ClickHouse and count events, not GitHub's lifetime totals — 264 stars here means stars gained during the window, not the repo's star count.