File size: 2,836 Bytes
510314e | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 | """Caricamento dei risultati della corsa principale (solo lettura dei file copiati da PC4070TI)."""
import json
from pathlib import Path
ROOT = Path(__file__).resolve().parents[1] # out/main-run-confirm
RUNS = ROOT / "runs"
PREFIX = "main-clar-" # corsa di conferma
MAIN_RUNS = ROOT.parent / "main-run" / "runs" # corsa principale (confronto)
LENGTHS = ["64k", "128k"]
POLS = ["A", "D", "B", "T"]
DEV_QID = "38_i13_j22_a_0031"
DATA = ROOT.parents[1] / "alb-bench-v2" / "alb" / "data" / "benchmark" / "kf-c"
def qtype_short(qt):
return "CF" if qt.startswith("Count Frequency") else ("FD" if qt.startswith("Find Duplicates") else qt)
def load_results(runs=None, prefix=None):
"""Una riga per (lunghezza, politica, qid), dai result.json (non da results.jsonl)."""
runs = runs or RUNS
prefix = prefix or PREFIX
rows = []
for L in LENGTHS:
for pol in POLS:
for d in sorted((runs / (prefix + L) / pol / "kf-c" / L).iterdir()):
r = json.loads((d / "result.json").read_text())
rows.append({
"L": L, "pol": pol, "qid": r["qid"], "dir": str(d),
"sample_id": int(r["qid"].split("_")[0]),
"qt": qtype_short(r["question_type"]),
"status": r["status"],
"off": float(r["score_official"] or 0), "norm": float(r["score_format_norm"] or 0),
"full": int(r["fully_correct"] or 0),
"empty": bool(r["answer_empty"]), "trunc": bool(r["answer_truncated"]),
"resid": bool(r["answer_residual_calls"]), "invalid": bool(r["answer_invalid_format"]),
"error": bool(r["answer_error"]),
"wall_s": r["wall_s"], "window": r["window"], "gold": r["gold"],
"pred": r.get("pred_answer"), "raw": r.get("raw_response"),
"finish": r.get("finish_reason"),
"engine": r.get("engine") or {}, "final_call": r.get("final_call") or {},
"margin_viol": (r.get("engine") or {}).get("margin_violations"),
})
return rows
def load_engine_calls(d):
return [json.loads(l) for l in open(Path(d) / "engine_calls.jsonl")]
def load_resources(d):
out = []
for l in open(Path(d) / "resources.jsonl"):
try:
out.append(json.loads(l))
except ValueError:
pass
return out
def samples(L, qt):
f = DATA / L / "tool_response" / ("count_frequency_tool.jsonl" if qt == "CF" else "find_duplicates_tool.jsonl")
out = {}
for l in open(f):
x = json.loads(l)
out[x["id"]] = x
return out
def load_main():
"""Corsa principale (domanda ambigua), per il confronto affiancato."""
return load_results(MAIN_RUNS, "main-")
|