quelllm.fr - reproducible pilot of 2026-09-12 (NVIDIA GeForce RTX 5070 Ti Laptop GPU, Ollama 0.30.6)

Contents: protocol (PROTOCOL.md), script (benchmark.py), hardware and initial state (metadata.json),
full responses and GPU placement (raw.json), model identities (*-show.json), code and document tasks
(*-code.json, *-documents.json), results (summary.json) and every repetition (measurements.csv).

measurements.csv is derived from raw.json: one row per generation, tokens_per_second = eval_count / eval_duration.
Warmup rows (warmup=true) are excluded from the median, minimum and maximum in summary.json.

Packaged on 2026-10-08. Same content as the files published under https://quelllm.fr/data/experiments/2026-09-12/
except: the machine host name in metadata.json is replaced by "redacted-host", and the output folder
in benchmark.py is a relative path (experiment/).

License: CC BY 4.0 for our synthetic inputs, measurements, protocol and summaries.
Model outputs are included as experimental evidence; model licenses remain applicable.
