Training speedrun record smashed: 30M model in 22.9s, 21.8x faster
RobertTLange · x · 2026-10-29
PR #52 on qlabs-eng/noise cuts the 30M-param Track 2 speedrun to 3.4976 val loss in 22.89s on one 8xH100 node — 21.8x faster than the prior record (499.6s) and 5.4x faster than the PR it builds on, via identity phase, fused Gluon chain kernels, and overlapped bf16 collectives.
More from Infra
- Map tracks data centers building their own gas plants, four near Dallas ozone zone — AINowInstitute · 2026-10-29
- GitHub commits hit 7.38B in September, 5x a year ago as agents flood in — rohanpaul_ai · 2026-10-29
- Why Macs lose to DGX Spark at batched inference: bandwidth wins solo, FP4 compute wins crowds — QuixiAI · 2026-10-29
- ComfyUI reproducibility: the workflow JSON alone isn't enough — Im_bored_rn69 · 2026-10-29
- Akamai launches Akamai Inference and Agent Sandbox for enterprise AI deployment — lucaolivari · 2026-10-29
- Cloudflare Q3 DDoS report: 8,200 attacks/hour as short bursts surge 208%, mega attacks drop 69% — Cloudflare Blog · 2026-10-29