Olmo 3 32B uses a 55x compute multiplier on code in its training mix
DZhang50 · x · 2026-11-14
A researcher on the Olmo team shared that at full Olmo 3 32B scale, the compute multiplier on code is 55x — meaning code data receives a vastly amplified share of training compute in the data mix. A concrete data point for the open model's training recipe.
More from Research
- Yamanaka lab preprint: partial reprogramming outcomes diverge sharply by cell lineage — MaxUnfried · 2026-11-14
- New paper speeds up Delaunay-based nearest neighbor search with grid queries and parallel construction — ssh4net · 2026-11-14
- Runway open-sources Praxis-1: 27B robot model trained on web video, tops RoboLab-120 at 46.2% — iamfakhrealam · 2026-11-14
- Paper on Predicting Memorization Before Fine-Tuning Accepted to EACL 2027 — niloofar_mire · 2026-11-14
- HarborRL update adds elastic scaling, SSH-free cluster deployment for agentic RL training — Dear_Treat3688 · 2026-11-14
- q6 vs q8 A/B testing at 300-400k context shows quantization loss amplified — Viktri1 · 2026-11-14