Marin 535B training starts with full open process and scaling ladder

ysu_nlp · x · 2026-08-22

Marin 535B-A23B started training this week with the entire process open. The model will train on 11x GB200 NVL72 clusters for 3 months on 18.75T tokens (2.7e24 FLOPs). Before the hero run, a scaling ladder from 1.6B to 27.7B parameters was trained to debug issues and forecast performance.

Related event: Stanford's Marin Open-Sources 23T Tokens of Pretraining Data and Launches Fully Transparent 535B Model Training(6 posts)→

Original post →

More from Infra

• Meta spends hundreds of millions on Azure tokens — Beth_Kindig · 2026-08-23
• Over 50% of GitHub's Capacity Issues Caused by Inefficient CI Tasks — cramforce · 2026-08-23
• Run DeepSeek-V3 on $47 Hardware with Pruning and Quantization — StefanoGogioso · 2026-08-23
• Call for Top Architects to Design Beautiful Datacenter Exteriors — EddyVGG · 2026-08-23
• 25% chance orbital data centers launch by end of next year — Polymarket · 2026-08-23
• VC proposes network of giant data centers along US-Mexico border — Polymarket · 2026-08-23

Infra channel →