A LocalLLaMA community post asks for ways to compare offline language models locally across speed, efficiency, reasoning, tool use and coding, including whether alternative inference engines are worth testing.
The post asks what strategies, prompts and software can benchmark multiple local models across different capabilities and performance measures. Its author wants to compare models with similar speeds and assess whether slower, more complex models offer enough benefit to justify the trade-off.
The author also asks whether benchmarking can be run locally and whether engines beyond llama.cpp and strata are worth considering. The post does not report benchmark results or recommend a particular tool.