Varun Singh
Bio
Varun Singh is currently pre-training lead at Arcee AI where he works on the end to end pre-training of large language models, with a strong interest in architecture and optimization. He has led the pre-training of Arcee's Trinity series of models, ranging from a 6B mixture-of-experts to a 400B mixture-of-experts model.
Session (1)
Varun Singh went from software engineer to leading pre-training on a 400B-parameter sparse MoE model (Trinity Large) at Arcee AI in roughly one year, and is first author of the Trinity Large technical report — making him a rare practitioner with hands-on frontier open-source pre-training experience at startup scale. Attending his session offers direct insight into the practical engineering and data decisions that let Arcee train a model rivaling closed-source frontier systems on a fraction of the budget.
Recent talks (1)
Recent writing (3)
- Arcee Trinity Large Technical Report · paper · Feb 2026
- Deep Dive: AFM-4.5B, the First Arcee Foundation Model · blog · Jun 2025
- Announcing Arcee Foundation Models · blog · Jun 2025
Podcasts & interviews (2)
- Arcee AI goes all-in on open models built in the U.S. · Interconnects (Nathan Lambert) · Jan 2026
- I Talked with Arcee AI for 100 Minutes and Everyone Needs to Know This · Himanshu's Substack · Apr 2026
Public activity researched automatically · as of Jun 2026