Rare find

Slowrun. A competition to train the best AI language model when you have unlimited time but fixed training data.

github.com/qlabs-eng/slowrun

Vaya's read on this project

Problem, audience, market, and the verdict — sign in to see it.

Updates

July 2026
  • Add citation
  • Update leaderboard
  • Add document-level shuffling to two hour track (#96)
June 2026
  • Update leaderboard
  • Merge recurrence and fp8 kernel, 3.307, 14.1m (#94)
  • Update leaderboard
  • tiny: fp8 MTP + full-layer XSA → 3.316 val loss (#93)
  • Update leaderboard
  • Tiny-track RLM: Recursive LLM (2x) (#92)
  • Update leaderboard
  • Add learnable XSA (#91)
May 2026
  • Update leaderboard
  • Snapshot ensemble + chain distillation + gradient based model selecti…
  • Fixing FA3 import in main, tiny and 2 hour track. Adding inline comme…
  • Update README.md (#88)
  • Update leaderboard
  • Two hour track updated window schedule val loss 3.144 (#86)
April 2026
  • Hybrid Attention Square-GDN (#85)
  • Update leaderboard
  • Anti-phase WD sawtooth + sqrt warmdown for two hour track (#84)