Stanford launches Marin 535B training with full open process
soumitrashukla9 · x · 2026-08-22
Stanford CRFM announced the start of training for Marin 535B-A23B. The plan involves 80% pretraining and 20% midtraining on 18.75T tokens, utilizing 11 x GB200 NVL72 clusters for approximately 3 months (2.7e24 FLOPs).
To ensure stability for their largest run yet, the team trained a 4-rung scaling ladder (1.6B to 27.7B) to debug issues and forecast performance. The project upholds principles defined in their Mistral framework from 5 years ago: simplicity, community, flexibility, transparency, reproducibility, pedagogy, and scalability.
More from Infra
- Large data centers employ only ~50-150 permanent on-site staff, Microsoft says — brezina · 2026-10-16
- Prediction: Opus 5.5-level models on laptops by 2027, backed by Qwen 27B data — cephaloform · 2026-10-16
- Oracle down 12% in a week, nearly 50% off June high as OpenAI exposure spooks market — alvelda · 2026-10-16
- US data center moratoriums top 400 — an AI bull explains why they're actually good for AI — kevinsxu · 2026-10-16
- Microsoft Research unveils Magnetar, the fastest zkVM at 10-39x speedups and post-quantum security — jedisct1 · 2026-10-16
- Bonsai 2 2B hits 120 tok/s on iPhone 18 Pro, outpaces Qwen 3.5 2B at half the memory — cephaloform · 2026-10-16