#!/usr/bin/env bash
# Regenerates output/dmr.txt and output/ollama.txt. Needs both servers up (scripts/start-dmr.sh,
# scripts/start-ollama.sh, or your own Docker Desktop Model Runner on :12434). Each engine is
# restarted first so its memory figure is a fresh container with only this one model loaded.
# Run it on an otherwise idle machine; the figures are only comparable within one run.
set -euo pipefail
cd "$(dirname "$0")/.."
ARGS="${BENCH_ARGS:---bench.repeats=3 --bench.max-tokens=48 --bench.tool-runs=10}"
docker restart ollama >/dev/null; sleep 5
scripts/bench.sh ollama http://localhost:11434/v1 qwen2.5:0.5b ollama $ARGS
docker restart dmr >/dev/null; sleep 5
scripts/bench.sh dmr http://localhost:12434/engines/v1 huggingface.co/qwen/qwen2.5-0.5b-instruct-gguf:Q4_K_M dmr $ARGS