Skip to content

Fast, Modern, and Low Precision PyTorch Optimizers

active 2023-10-272026-05-16 (UTC)

Complete coverage26,622 / 26,622 hourly files (100%) · 2 absent upstream2023-08-152026-08-28 (UTC)
Events
284
Pushes
119
Pull requests
5
Issues
7
Stars
100
Forks
4

Activity over time

Daily event counts in the loaded window

Line chart, 933 days from 2023-10-27 to 2026-05-16. Pushes: 119 total, peak 7 in a day. Pull requests: 5 total, peak 2 in a day. Issues: 7 total, peak 1 in a day. Comments: 10 total, peak 2 in a day. Stars: 100 total, peak 6 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

ContributorContributionsPushesPRsComments
warner-benjamin11810656
github-actions[bot]131300
mensfeld3001
sdbds3001
vince62s3002
zaptrem1000

Recent activity

Latest issues, pull requests and releases

  • Issue#7warner-benjamin2025-12-26 22:41
    Casting Existing FP32/FP16 model weights to BF16 + Kahan Summation
  • Issue comment#12warner-benjamin2025-12-26 22:39
    Keyerror when resuming training with optimizer adamw_stable
  • Releasewarner-benjamin2025-12-26 22:35
    v0.3.3
  • Pull request#13warner-benjamin2025-12-26 20:07
  • Issue comment#11warner-benjamin2025-12-08 00:54
    Is this LLM-only?
  • Issue comment#10warner-benjamin2025-09-22 05:53
    AdamW Triton implementation slower than compiled foreach
  • Releasewarner-benjamin2025-07-25 21:08
    v0.3.1
  • Releasewarner-benjamin2025-07-15 20:36
    v0.3.0: Fast Triton Kernels
  • Issue comment#8vince62s2025-03-10 17:23
    pure fp16 ?
  • Issue comment#8vince62s2025-03-10 09:47
    pure fp16 ?
  • Issue comment#8warner-benjamin2025-03-09 19:18
    pure fp16 ?
  • Issue#8vince62s2025-03-08 13:38
    pure fp16 ?
  • Issue#7zaptrem2024-07-30 01:56
    Casting Existing FP32/FP16 model weights to BF16 + Kahan Summation
  • Issue comment#5sdbds2024-07-02 05:08
    When using accelerator with Gradient Release hook, Increasing VRAM consumption
  • Issue comment#5warner-benjamin2024-07-01 02:07
    When using accelerator with Gradient Release hook, Increasing VRAM consumption
  • Issue#5sdbds2024-06-26 12:43
    When using accelerator with Gradient Release, Increasing VRAM consumption
  • Issue comment#4warner-benjamin2024-06-12 14:53
    Document error
  • Issue#4sdbds2024-06-12 07:37
    Document error
  • Releasewarner-benjamin2024-06-06 00:25
    v0.2.1: param_groups_weight_decay
  • Releasewarner-benjamin2024-06-05 22:59
    v0.2.1: param_groups_weight_decay
  • Releasewarner-benjamin2024-06-05 22:54
    V0.2.1
  • Releasewarner-benjamin2024-03-11 04:13
    v0.2.0
  • Pull request#3warner-benjamin2024-03-11 04:09
  • Pull request#3warner-benjamin2024-03-11 04:06
  • Pull request#2warner-benjamin2024-02-21 04:22

Totals cover only the window loaded into ClickHouse and count events, not GitHub's lifetime totals — 100 stars here means stars gained during the window, not the repo's star count.