35B-param MoE model with 3B active runs local at crazy speed, 100% accurate on simple tasks
draginol · x · 2026-10-19
draginol tested Ornith's 35B-parameter model with only 3B active params locally via ollama (23.5GB dedicated GPU memory). His verdict: "crazy fast," and 100% accurate on simple tasks in his tests. He's now pushing it with tougher tasks in a multi-part thread ([1/N]).
Related event: Ornith 35B/3B MoE Runs Fast and Accurate Locally(2 posts)→
More from Infra
- Underwater AI data center planned in Alaska's Cook Inlet, home to 330 endangered belugas — anselm · 2026-10-19
- DeepInfra selling DeepSeek v4.1 at 49% off sparks doubts it can turn a profit — gharik · 2026-10-19
- OpenFaaS bought 4x DGX Sparks for air-gapped coding and red-teaming — natesiggard · 2026-10-19
- wllama brings llama.cpp to the browser via WebAssembly, 1.3k stars — mervenoyann · 2026-10-19
- CoreWeave RL Rollouts hot-loads weights 15x faster, lifts BrowseComp to 45.45% — _ScottCondron · 2026-10-19
- Close to sub-30s full-weight load of GLM 5.3 Flash NVFP4 — TheZachMueller · 2026-10-19