Marin 535B switches to cooldown mix: science data up, top-quality bucket at 28%
dlwh · x · 2026-10-24
For the final stretch of Marin's 535B pretraining, the team switched to a cooldown data mix: more natural-science research, with code and agent transcripts still prominent. Natural-science sampling weight rose from 6.9% to 11.3%, general software & web code fell from 4.5% to 3.1%, and the highest classifier-rated quality bucket increased from 22.9% to 28.0% of sampling weight. The recipe spans 40 topic clusters × 5 quality buckets.
More from Research
- Frequency drives conventionalization: higher-exposure communities conventionalize more words — abenitezburraco · 2026-10-24
- GeoWind2Plan: seconds-scale 3D urban wind prediction for energy-efficient drones — AnimaAnandkumar · 2026-10-24
- Flowing water and lava come to 3D Gaussian Splatting scenes — jonstephens85 · 2026-10-24
- Immorta Bio's AI-designed 9-amino-acid peptide SenoMune-0257 targets senescent cells — DavidmComfort · 2026-10-24
- France rebuilt its tax code as MLang, a formally verified open-source compiler — fkasummer · 2026-10-24
- Running RFD4-Proteina protein design model at thinking_high on consumer cloud GPUs — AllThingsApx · 2026-10-24