Inference Notes
Record durable observations from local llama.cpp optimization runs here.
Keep entries concise:
- target model and exact GGUF file
- baseline command
- changed variable
- benchmark result
- whether the change became the new fastest reproducible command