code/scripts/build_orpo_v3_fast.py

"""
v3 preference dataset builder for ORPO validator training.

TWO-STAGE LABELING (combines INDEP verdict signal + COLLAB content signal):
  chosen iff (Conclude verdict matches planner correctness) AND (fixer-with-critique → correct SQL)
  rejected otherwise

This:
- INDEP-style rewards: chosen has correct verdict (whatever planner is, the chosen critique's
  Conclude: token matches it).
- COLLAB-style rewards: chosen critique also makes the fixer produce the right SQL.
- Penalize: critiques with wrong verdict (misleading) AND critiques whose content can't get
  the fixer to succeed even when verdict is right.

YIELD MAX: 9428 BIRD-train questions × K critiques × ALL chosen × ALL rejected pairs
  (no [:2] truncation). Realistic ~45-75K pairs on K=8.

Chunking: --start_idx / --end_idx for parallel SLURM jobs. ThreadPoolExecutor for client-side
concurrency over questions; vLLM batches incoming requests.
"""
import argparse
import json
import os
import re
import random
import sqlite3
import threading
from concurrent.futures import ThreadPoolExecutor, as_completed

os.environ.setdefault("PYTHONNOUSERSITE", "1")
os.environ["NO_PROXY"] = "localhost,127.0.0.1"

import requests
from datasets import load_dataset, Dataset, DatasetDict

_db_lock = threading.Lock()


def safe_exec(db_path, sql, timeout=5):
    r = [None]; e = [None]
    def _run():
        try:
            c = sqlite3.connect(db_path); c.text_factory = lambda b: b.decode(errors="ignore")
            r[0] = c.execute(sql).fetchmany(100); c.close()
        except Exception as ex:
            e[0] = str(ex)
    t = threading.Thread(target=_run, daemon=True); t.start(); t.join(timeout)
    return (None, "TIMEOUT") if t.is_alive() else (r[0], e[0])


def results_match(g, p):
    if g is None or p is None: return False
    def n(rs): return sorted(tuple(str(v).strip().lower() if v is not None else "" for v in r) for r in rs)
    return n(g) == n(p)


def extract_sql(text):
    m = re.search(r"```(?:sql)?\s*(.*?)\s*```", text, re.DOTALL)
    if m:
        s = m.group(1).strip()
        return s[3:].strip() if s.upper().startswith("SQL") else s
    return ""


def qwen_chat(p):
    return f"<|im_start|>user\n{p}<|im_end|>\n<|im_start|>assistant\n"


def llama3_chat(p):
    return (f"<|begin_of_text|><|start_header_id|>user<|end_header_id|>\n\n"
            f"{p}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n")


def vllm_complete(host, model, prompt, n, temperature, top_p, max_tokens, seed, stop=None):
    try:
        r = requests.post(f"{host}/v1/completions", json={
            "model": model, "prompt": prompt,
            "n": n, "temperature": temperature, "top_p": top_p,
            "max_tokens": max_tokens, "seed": seed,
            "stop": stop or ["<|eot_id|>", "<|im_end|>"],
        }, timeout=300)
        r.raise_for_status()
        return [c["text"].strip() for c in r.json()["choices"]]
    except Exception:
        return []


FIXER_INSTR = (
    "You are an expert SQL judge and fixer. You will see a candidate SQL, its execution result, "
    "and a validator's critique.\n\n"
    "Your task:\n"
    "1. Decide if the candidate SQL correctly answers the question. Consider the validator's "
    "critique as a hint, but verify with your own SQL expertise.\n"
    "2. If the candidate SQL is correct, output it UNCHANGED.\n"
    "3. If the candidate SQL has a real issue, output a corrected SQL.\n"
    "4. Prefer keeping the candidate unchanged when in doubt.\n\n"
    "Output ONLY the final SQL inside ```sql ... ``` markers."
)


def build_fixer_prompt(schema_str, question, evidence, planner_sql, exec_response, wrapped_critique):
    body = (
        f"\n\nDatabase Schema:\n{schema_str.rstrip()}\n\n"
        f"Question: {question}\n"
        f"External knowledge: {evidence or 'None'}\n\n"
        f"Candidate SQL:\n{planner_sql}\n\n"
        f"Execution response:\n{exec_response}\n\n"
        f"Validator critique:\n{wrapped_critique}\n\nFinal SQL:"
    )
    return FIXER_INSTR + body


def parse_verdict(text):
    """Returns 'correct', 'incorrect', or 'unknown'."""
    if not text: return 'unknown'
    if 'Conclude: correct' in text: return 'correct'
    if 'Conclude: incorrect' in text: return 'incorrect'
    return 'unknown'


def process_one(args, q_lower, info, bird_train, side, idx):
    bt = bird_train[info["sid"]]
    db_path = bt.get("db_path") or f"data/train_databases/{bt['db_id']}/{bt['db_id']}.sqlite"
    if not os.path.exists(db_path):
        return ("skip_no_db", [], 0, 0)

    question = bt["question"]
    evidence = bt.get("evidence", "") or ""

    user_msg = info["user_msg"]
    if "Database Schema:" in user_msg:
        schema_str = user_msg.split("Database Schema:", 1)[1].split("Question:", 1)[0].rstrip()
    else:
        schema_str = user_msg

    planning_prompt = user_msg.rstrip() + "\n\nPlanning:"
    plans = vllm_complete(
        args.planner_host, "planner", qwen_chat(planning_prompt),
        n=1, temperature=0.0, top_p=1.0, max_tokens=1024, seed=args.seed + idx,
    )
    if not plans:
        return ("no_planner", [], 0, 0)
    m = re.search(r"Final SQL query:\s*```(?:sql)?\s*(.+?)```", plans[0], re.DOTALL | re.IGNORECASE)
    planner_sql = m.group(1).strip() if m else extract_sql(plans[0])
    if not planner_sql:
        return ("no_planner", [], 0, 0)

    with _db_lock:
        gold_res, _ = safe_exec(db_path, bt["sql"])
        pred_res, perr = safe_exec(db_path, planner_sql)
    if gold_res is None:
        return ("no_gold", [], 0, 0)
    planner_correct = (not perr) and results_match(gold_res, pred_res)
    exec_response = (f"Error: {perr[:200]}" if perr
                     else f"OK. Result rows (preview): {str(pred_res)[:300]}")

    # Generate K critiques (paper format, seeded with clause token)
    clause_token = "SELECT." if side == "sel" else "CONDITION."
    schema_in_val_prompt = (info["user_msg"]
                            .split("Database Schema:", 1)[1].split("Question:", 1)[0]).rstrip() \
        if "Database Schema:" in info["user_msg"] else info["user_msg"]
    val_prompt = (
        f"Generate feedbacks to fix the following SQL query:\n"
        f"Database Schema:{schema_in_val_prompt}\n\n"
        f"Question: {question}\n"
        f"External knowledge: {evidence}\n\n"
        f"SQL query: {planner_sql}\n\n"
        f"Execution response:\n{exec_response}\n\n"
        f"Feedback:"
    )
    seeded_prompt = val_prompt + "\n" + clause_token + "\n"
    critiques = vllm_complete(
        args.validator_host, "validator", llama3_chat(seeded_prompt),
        n=args.K, temperature=args.temperature, top_p=0.9,
        max_tokens=384, seed=args.seed + idx,
    )
    if not critiques:
        return ("no_val", [], 0, 0)
    critiques = [f"{clause_token}\n{c.lstrip()}" for c in critiques]

    chosen, rejected = [], []
    for crit in critiques:
        verdict = parse_verdict(crit)
        if verdict == 'unknown':
            # Critiques without a clear Conclude token are unusable for verdict learning; drop.
            continue
        verdict_matches = (
            (planner_correct and verdict == 'correct') or
            (not planner_correct and verdict == 'incorrect')
        )

        wrapped_crit = (
            f"<{'select' if side == 'sel' else 'condition'}>\n{crit}\n"
            f"</{'select' if side == 'sel' else 'condition'}>"
        )
        fix_prompt = build_fixer_prompt(schema_str, question, evidence, planner_sql, exec_response, wrapped_crit)
        fix_outs = vllm_complete(
            args.fixer_host, "fixer_big", qwen_chat(fix_prompt),
            n=1, temperature=0.0, top_p=1.0, max_tokens=512,
            seed=args.seed + idx,
        )
        if not fix_outs:
            rejected.append(crit)
            continue
        fix_sql = extract_sql(fix_outs[0])
        if not fix_sql:
            rejected.append(crit)
            continue
        with _db_lock:
            fix_res, fix_err = safe_exec(db_path, fix_sql)
        fix_correct = (not fix_err) and results_match(gold_res, fix_res)

        # TWO-STAGE LABELING
        if verdict_matches and fix_correct:
            chosen.append(crit)
        else:
            rejected.append(crit)

    # ALL chosen × ALL rejected (no [:2] truncation)
    pairs = []
    for c in chosen:
        for r in rejected:
            pairs.append({"prompt": val_prompt, "chosen": c, "rejected": r})

    status = "planner_correct" if planner_correct else "planner_wrong"
    return (status, pairs, len(chosen), len(rejected))


def main():
    p = argparse.ArgumentParser()
    p.add_argument("--planner_host",   default="http://localhost:8100")
    p.add_argument("--validator_host", default="http://localhost:8101")
    p.add_argument("--fixer_host",     default="http://localhost:8102")
    p.add_argument("--side",           required=True, choices=["sel", "cond"])
    p.add_argument("--K", type=int, default=8)
    p.add_argument("--temperature", type=float, default=1.0)
    p.add_argument("--start_idx", type=int, default=0, help="start index in shuffled griffith list")
    p.add_argument("--end_idx",   type=int, default=-1, help="end index (exclusive); -1 means all")
    p.add_argument("--threads", type=int, default=32)
    p.add_argument("--seed", type=int, default=42)
    p.add_argument("--out", required=True)
    args = p.parse_args()

    print("Loading BIRD-train + griffith prompts...", flush=True)
    with open("data/sft_bird_with_evidence_train_text2sql.json") as f:
        bird_train = json.load(f)
    ds_g = load_dataset("griffith-bigdata/sft_text2sql", split="train_sft",
                        cache_dir="/weka/s225250685/Huggingface/hub"
                       ).filter(lambda x: x["model_name"] == "deepseek-reasoner")
    griffith = {}
    for row in ds_g:
        sid = int(row["sample_id"])
        if not (0 <= sid < len(bird_train)): continue
        user_msg = row["messages"][1]["content"]
        q_m = re.search(r"Question:\s*(.+?)(?:\n|$)", user_msg)
        if not q_m: continue
        q = q_m.group(1).strip()
        if q.lower() == bird_train[sid]["question"].strip().lower():
            griffith[q.lower()] = {"user_msg": user_msg, "sid": sid}
    print(f"  griffith: {len(griffith)} questions", flush=True)

    random.seed(args.seed)
    items = list(griffith.items()); random.shuffle(items)
    end = args.end_idx if args.end_idx > 0 else len(items)
    chunk = items[args.start_idx:end]
    print(f"  chunk: items[{args.start_idx}:{end}] = {len(chunk)} questions",
          f"K={args.K} side={args.side} threads={args.threads}", flush=True)

    rows_all = []
    counters = {"planner_correct": 0, "planner_wrong": 0,
                "no_planner": 0, "skip_no_db": 0, "no_gold": 0, "no_val": 0}
    total_chosen = 0
    total_rejected = 0

    with ThreadPoolExecutor(max_workers=args.threads) as ex:
        futures = []
        for idx, (q_lower, info) in enumerate(chunk):
            futures.append(ex.submit(process_one, args, q_lower, info, bird_train, args.side, args.start_idx + idx))
        done = 0
        for fut in as_completed(futures):
            try:
                status, pairs, n_c, n_r = fut.result()
                total_chosen += n_c
                total_rejected += n_r
            except Exception as e:
                print(f"  worker exception: {e}", flush=True)
                continue
            counters[status] = counters.get(status, 0) + 1
            rows_all.extend(pairs)
            done += 1
            if done % 100 == 0:
                print(f"  [{done}/{len(chunk)}] pairs={len(rows_all)} "
                      f"chosen_traj={total_chosen} rejected_traj={total_rejected} "
                      f"ok={counters['planner_correct']} wrong={counters['planner_wrong']} "
                      f"no_planner={counters['no_planner']} no_gold={counters['no_gold']} no_val={counters['no_val']}",
                      flush=True)

    print(f"\nGenerated {len(rows_all)} (chosen, rejected) pairs", flush=True)
    print(f"  counters: {counters}", flush=True)
    print(f"  total chosen={total_chosen}, rejected={total_rejected}", flush=True)
    if not rows_all:
        print("ERROR: no rows generated"); return

    random.seed(42); random.shuffle(rows_all)
    n_train = int(0.95 * len(rows_all))
    DatasetDict({
        "train_dpo": Dataset.from_list(rows_all[:n_train]),
        "test_dpo":  Dataset.from_list(rows_all[n_train:]),
    }).save_to_disk(args.out)
    print(f"Saved → {args.out}  train={n_train}  test={len(rows_all) - n_train}", flush=True)


if __name__ == "__main__":
    main()