use cpu to offload moe weights to reduce the VRAM usage. · Issue #11772 · ollama/ollama

github.com

saved by: 0 people
related read on: Ollama: all aboard open models · Ollama Blog from the discussion