| |
| """Corsa di conferma (preregistrata 17/10/2026, MAIN-RUN-CONFIRM.md): stesse 100 domande della corsa principale, con la |
| domanda resa non ambigua. Il banco intende "tool results for round N" come l'uscita dello strumento che PRECEDE il |
| messaggio "Round N: Guess ..." (verificato su 100/100 risposte ufficiali); la domanda lo dice ora esplicitamente. |
| Risposte ufficiali e scorer invariati. Crea alb-clar/ (dati con la nota, src condiviso) e le liste main-clar-*.txt.""" |
| import json |
| import os |
| from pathlib import Path |
|
|
| B = Path("/mnt/mneme-nvme/proxy-ideas/alb-bench-v2") |
| SRC = B / "alb/data/benchmark" |
| DST = B / "alb-clar/data/benchmark" |
| NOTE = (" Note: in this dialogue, \"the tool results for round N\" means the tool output that appears just BEFORE the" |
| " user message \"Round N: Guess ...\" (the result of the query made for the guess of round N), not the one" |
| " after it.") |
| (B / "alb-clar").mkdir(exist_ok=True) |
| if not (B / "alb-clar/src").exists(): |
| os.symlink("../alb/src", B / "alb-clar/src") |
| for L in ("64k", "128k"): |
| for t in ("count_frequency", "find_duplicates"): |
| rel = Path("kf-c/%s/tool_response/%s_tool.jsonl" % (L, t)) |
| (DST / rel).parent.mkdir(parents=True, exist_ok=True) |
| n = 0 |
| with open(SRC / rel) as fi, open(DST / rel, "w") as fo: |
| for line in fi: |
| r = json.loads(line) |
| assert "round" in r["question"], r["question"] |
| r["question"] = r["question"].rstrip() + NOTE |
| fo.write(json.dumps(r, ensure_ascii=False) + "\n") |
| n += 1 |
| print(rel, n) |
| src = (B / ("scripts/main-%s.txt" % L)).read_text().splitlines() |
| out = ["# CONFERMA (domanda non ambigua, MAIN-RUN-CONFIRM.md); dati in alb-clar/; stesse righe di main-%s.txt" % L] |
| out += [l for l in src if l and not l.startswith("#")] |
| (B / ("scripts/main-clar-%s.txt" % L)).write_text("\n".join(out) + "\n") |
| print("esempio:", json.loads(open(DST / "kf-c/64k/tool_response/count_frequency_tool.jsonl").readline())["question"]) |
|
|
