JThomas-CoE/College-of-Experts-AI

A framework decomposing large MoE models into disk-resident specialist networks for consumer hardware. Run SOTA-class ensembles locally on limited VRAM by dynamically loading experts. Proof of concept showed +27% benchmark gain over isolated models.

3 best points · 1 comments · 1 Hacker News post · 6 GitHub stars · View on GitHub

Hacker News discussions

• Slicing an 80B MoE LLM into 40B domain specialists (3 points, 1 comments, by JThomas-CoE, 2026-03-11)