{"name":"ModelFit Measured Local-LLM Benchmark Dataset","description":"Real tokens/sec measured on submitters' machines with the fixed reference model qwen2.5:1.5b-instruct-q4_K_M (fixed prompt, comparable across machines), aggregated to per-hardware medians. These are measurements from the wild, not estimates.","referenceModel":"qwen2.5:1.5b-instruct-q4_K_M","promptId":"modelfit-bench-v1","source":"https://modelfit.io/benchmark/","license":"CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)","attribution":"ModelFit: https://modelfit.io/","methodology":"Each submission is one `modelfit bench` run (ollama, fixed reference model, fixed prompt). Aggregates are medians per hardware key; n is the submission count. Variance between runs on the same machine can be significant; trust medians at n>=3.","updated":"2026-10-03T00:05:38.960Z","count":1,"aggregates":[{"key":"Apple M4","kind":"chip","model":"qwen2.5:1.5b-instruct-q4_K_M","n":1,"medianTokPerSec":71.3,"minTokPerSec":71.3,"maxTokPerSec":71.3,"lastAt":"2026-10-03T00:05:38.960Z"}]}