events
Scaling Marin's Open LLM Pretraining
Open Athena organizza una serata su come il progetto Marin ha portato su larga scala il preaddestramento di LLM aperti, dalle miscele di token all’addestramento di modelli sparsi su grandi cluster.

In inglese
- Starts
- 2026-11-18 02:00 UTC
- Ends
- 2026-11-18 04:00 UTC
- Organizer
- Personal
- Location
- San Francisco, United States
Open Athena hosts an evening at Laude Labs about scaling the Marin open language-model project. The team reports finishing pre-training of a 535B-parameter mixture-of-experts model on 18T tokens, on GPUs from the Huang Foundation, in what it calls the largest live-streamed pre-training run so far. Talks cover compute and data scaling, systems for turning raw tokens into training mixtures, efficient sparse-model training and infrastructure across heterogeneous clusters. Percy Liang gives welcome remarks, David Hall and Andy Konwinski hold a fireside chat, Will Held presents, and freeform conversations follow.
Event website