Local LLM / Local inference performance
Local LLM: Performance rollback criteria
Define acceptable quality, p95 latency and memory limits before tuning.
Environment and inputs: Ollama, llama.cpp or vLLM serving; tune only features supported by the selected release and hardware.
When to use it
Compare performance with a fixed workload and a documented quality gate. Prompt processing, generation, queueing and model loading contribute different delays; isolate the stage that dominates the user-visible experience before changing configuration.
Check the result
Revert a faster configuration when it fails any required acceptance limit.
Save and compare results
Save the test input, configuration and result. Fix the failing boundary, repeat the same test and compare it with the saved working configuration.
Scope and limits
These exercises have not been tested on your systems. Check feature support and configuration names against the installed release and linked documentation.