Ollama says my model does 13,826 tokens/sec. It does 43.

That number is not a typo, and my GPU has not improved. Both figures came out of the same daemon,...