alea/gen_hf4.py

"""gen_hf4: third wave — science/commonsense MCQ + math word problems."""
import argparse, json, random
from datasets import load_dataset

rng = random.Random(77)


def choice_row(state, instructions, options_texts, correct_idx, family,
               meta=None):
    n = len(options_texts)
    keys = ["opt_%d" % i for i in range(n)]
    seen, fo, fk, ni = {}, [], [], None
    for i, t in enumerate(options_texts):
        t = str(t).strip()
        if not t:
            t = "(empty %d)" % i
        if t in seen:
            if i == correct_idx:
                ni = seen[t]
            continue
        seen[t] = len(fo); fo.append(t); fk.append(keys[i])
        if i == correct_idx:
            ni = len(fo) - 1
    if ni is None or len(fo) < 2:
        return None
    tgt = [0.0] * len(fo); tgt[ni] = 1.0
    return {"state": str(state)[:20000],
            "questions": {"q0": {"type": "choice", "instructions": instructions,
                                 "criteria": dict(zip(fk, fo))}},
            "targets": {"q0": tgt}, "family": family, "meta": meta or {}}


def gen_sciq(n, rng):
    ds = load_dataset("allenai/sciq", split="train")
    out = []
    for r in ds:
        if len(out) >= n:
            break
        opts = [r["correct_answer"], r["distractor1"], r["distractor2"],
                r["distractor3"]]
        order = list(range(4)); rng.shuffle(order)
        shuffled = [opts[j] for j in order]
        ci = shuffled.index(r["correct_answer"])
        st = "Question: %s\n(Support: %s)" % (r["question"],
                                             (r.get("support") or "")[:800])
        row = choice_row(st, "Choose the correct answer.", shuffled, ci,
                         "hf_sciq", meta={"src": "sciq"})
        if row:
            out.append(row)
    return out


def gen_qasc(n, rng):
    ds = load_dataset("allenai/qasc", split="train")
    out = []
    for r in ds:
        if len(out) >= n:
            break
        ch = r["choices"]
        opts, ci = ch["text"], ch["label"].index(r["answerKey"])
        row = choice_row("Question: %s" % r["question"],
                         "Choose the correct answer.", opts, ci, "hf_qasc",
                         meta={"src": "qasc"})
        if row:
            out.append(row)
    return out


def gen_arc(n, rng):
    ds = load_dataset("allenai/ai2_arc", "ARC-Challenge", split="train")
    out = []
    for r in ds:
        if len(out) >= n:
            break
        ch = r["choices"]
        labels = ch["label"]
        opts = ch["text"]
        if r["answerKey"] not in labels:
            continue
        ci = labels.index(r["answerKey"])
        row = choice_row("Question: %s" % r["question"],
                         "Choose the correct answer.", opts, ci, "hf_arc",
                         meta={"src": "arc_c"})
        if row:
            out.append(row)
    return out


def gen_csqa(n, rng):
    ds = load_dataset("tau/commonsense_qa", split="train")
    out = []
    for r in ds:
        if len(out) >= n:
            break
        ch = r["choices"]
        labels = ch["label"]; opts = ch["text"]
        if r["answerKey"] not in labels:
            continue
        ci = labels.index(r["answerKey"])
        st = ("Context: %s\n\nQuestion: %s" %
              (r.get("question_concept", ""), r["question"]))
        row = choice_row(st, "Choose the most sensible answer.", opts, ci,
                         "hf_csqa", meta={"src": "csqa"})
        if row:
            out.append(row)
    return out


def gen_obqa(n, rng):
    ds = load_dataset("allenai/openbookqa", "main", split="train")
    out = []
    for r in ds:
        if len(out) >= n:
            break
        ch = r["choices"]; opts = ch["text"]
        if r["answerKey"] not in ch["label"]:
            continue
        ci = ch["label"].index(r["answerKey"])
        row = choice_row("Question: %s" % r["question_stem"],
                         "Choose the correct answer.", opts, ci, "hf_obqa",
                         meta={"src": "obqa"})
        if row:
            out.append(row)
    return out


def gen_orca(n, rng):
    """Orca math word problems -> choice over perturbed numeric answers."""
    ds = load_dataset("microsoft/orca-math-word-problems-200k", split="train")
    out = []
    idxs = list(range(len(ds))); rng.shuffle(idxs)
    import re
    for i in idxs:
        if len(out) >= n:
            break
        r = ds[i]
        ans_txt = r["answer"]
        m = re.findall(r"-?\d[\d,]*\.?\d*", ans_txt.split("\n")[-1])
        if not m:
            continue
        try:
            a = float(m[-1].replace(",", ""))
        except ValueError:
            continue
        cands = {a * f for f in (0.5, 0.8, 1.25, 1.5, 2.0)} |                 {a + d for d in (-2, -1, 1, 2, 10)}
        cands = [c for c in cands if abs(c - a) > max(0.4, 0.03 * abs(a))]
        rng.shuffle(cands)
        fmt = lambda x: ("%g" % x)
        opts = [fmt(x) for x in ([a] + cands[:3])]
        order = list(range(4)); rng.shuffle(order)
        shuffled = [opts[j] for j in order]
        ci = shuffled.index(fmt(a))
        st = "Problem:\n%s" % r["question"].strip()
        row = choice_row(st, "Solve the problem and choose the exact answer.",
                         shuffled, ci, "hf_orca", meta={"src": "orca_math"})
        if row:
            out.append(row)
    return out


GENS = {"sciq": gen_sciq, "qasc": gen_qasc, "arc": gen_arc,
        "csqa": gen_csqa, "obqa": gen_obqa, "orca": gen_orca}


def main():
    ap = argparse.ArgumentParser()
    ap.add_argument("--out", default="data/hf4.jsonl")
    ap.add_argument("--per", type=int, default=6000)
    args = ap.parse_args()
    rows = []
    for name, fn in GENS.items():
        try:
            got = fn(args.per, rng)
        except Exception as e:
            print(name, "FAILED", type(e).__name__, str(e)[:120]); continue
        print(name, len(got)); rows += got
    rng.shuffle(rows)
    with open(args.out, "w") as f:
        for r in rows:
            f.write(json.dumps(r, ensure_ascii=False) + "\n")
    print("TOTAL", len(rows), "->", args.out)


if __name__ == "__main__":
    main()