empire-eval / scripts /run_baselines.sh
Lilambd's picture
Publish Lilambd Empire Eval
7f9a652 verified
Raw History Blame Contribute Delete
1.34 kB
#!/usr/bin/env bash
# Run the no-network baselines on data/dev and save results/baseline_<name>/score.json.
# usage: bash scripts/run_baselines.sh (from the package root, or anywhere)
set -euo pipefail
cd "$(dirname "$0")/.."
PY=${PYTHON:-python3}
for name in oracle null random-guess; do
dir="results/baseline_${name}"
mkdir -p "$dir"
echo "== $name" >&2
rm -f "$dir/SCORED_WITH_GUARD.txt"
"$PY" scripts/run_eval.py --qa data/dev --provider "$name" --no-resume --out "$dir/predictions.jsonl"
if ! "$PY" scripts/score.py --qa data/dev --pred "$dir/predictions.jsonl" --out "$dir/score.json" > "$dir/score.txt" 2> "$dir/score.err"; then
echo "!! score.py crashed on $name (see $dir/score.err); rescoring with scripts/score_guarded.py" >&2
"$PY" scripts/score_guarded.py --qa data/dev --pred "$dir/predictions.jsonl" --out "$dir/score.json" > "$dir/score.txt"
cp "$dir/score.err" "$dir/SCORED_WITH_GUARD.txt"
fi
"$PY" - "$dir/score.json" <<'EOF'
import json, sys
s = json.load(open(sys.argv[1]))['summary']
print(f"{sys.argv[1]}: accuracy={s['accuracy']} answerable={s['answerable_accuracy']} "
f"abstention={s['abstention_accuracy']} hallucination={s['hallucination_rate']} "
f"false_abstention={s['false_abstention']} n={s['questions']} predicted={s['predicted']}")
EOF
done