local-llm-speed-benchmarks

简介

Local LLM speed benchmarks (community union) Real measured local-LLM inference performance on consumer hardware: (accelerator, model, quant, context) -> prompt tok/s, decode tok/s, TTFT. Collected to train FitCheck's honest speed predictor (methodology: LLM-Pilot, leave-one-accelerator-out validation). Sources & attribution LocalScore (https://www.localscore.ai) — Mozilla Builders project by cjpais. The bulk of this dataset. Scraped politely from public pages;… See the full description on the dataset page: https://huggingface.co/datasets/cn0303/local-llm-speed-benchmarks.