""" v8 — Build enriched-prompt pointwise data from BIRD-TRAIN paper-format rollouts. Enriched prompt contains: - Rich schema (table/column descriptions, sample values, FKs) - Question + evidence - Candidate SQL - Execution result (rows preview) - Validator critique (fb_select / fb_condition / fb_join / fb_order) - Planner reasoning trace (planner_output, first ~400 chars) - Structural hints: has LIMIT?, GROUP BY?, JOINs count, DISTINCT, aggregate functions Output: HF DatasetDict at data/sft_selector_v8_pointwise_enriched/{train,test} """ import argparse, json, os, re, sys, random from concurrent.futures import ThreadPoolExecutor, as_completed os.environ.setdefault("PYTHONNOUSERSITE", "1") os.environ.setdefault("DB_EXEC_API_DISABLE", "1") ROOT = "/weka/s225250685/mats-tist" os.chdir(ROOT); sys.path.insert(0, ROOT) from validator_data.validator import _execute_sql from datasets import Dataset, DatasetDict from scripts.rich_schema import render_rich_schema ENRICHED_PROMPT = ( "You are a SQL correctness judge for the BIRD benchmark. Use ALL the " "context below to decide if the candidate SQL is correct.\n\n" "Database Schema:\n{schema}\n\n" "Question: {question}\n" "External knowledge: {evidence}\n\n" "Candidate SQL:\n{sql}\n\n" "Execution result of the candidate:\n{exec_result}\n\n" "Structural features of the candidate:\n{struct}\n\n" "Validator critique of the planner draft:\n" " - select: {fb_select}\n" " - condition: {fb_condition}\n" " - join: {fb_join}\n" " - order: {fb_order}\n\n" "Planner reasoning (excerpt):\n{planner_excerpt}\n\n" "Does this SQL correctly answer the question? Answer YES or NO." ) MAX_SCHEMA_CHARS = 2500 # reduced because we added other context def safe_truncate(s, n): s = str(s) if s is not None else "" return s if len(s) <= n else s[:n] + "..." def exec_str_for(db_path, sql, timeout=8): if not sql or not sql.strip(): return "Error: empty SQL" try: r, err = _execute_sql("./" + db_path if not db_path.startswith("./") else db_path, sql, timeout=timeout) except Exception as e: return f"Error: {str(e)[:160]}" if err: return f"Error: {str(r)[:160]}" rows = str(r)[:260] return f"OK. Rows preview: {rows}" if rows.strip() and rows.strip() != "[]" else "OK. (no rows returned)" def struct_features(sql): sl = sql.lower() feats = [] if " distinct" in sl or "distinct " in sl: feats.append("uses DISTINCT") if " limit " in sl or sl.endswith("limit"): feats.append("uses LIMIT") if " group by " in sl: feats.append("uses GROUP BY") if " order by " in sl: feats.append("uses ORDER BY") if " having " in sl: feats.append("uses HAVING") n_joins = sl.count(" join ") if n_joins > 0: feats.append(f"{n_joins} JOIN(s)") aggs = [] for a in ("count(", "sum(", "avg(", "max(", "min("): if a in sl: aggs.append(a.rstrip("(")) if aggs: feats.append("aggregates: " + ", ".join(aggs)) if " is null" in sl: feats.append("uses IS NULL") if "strftime" in sl or " date(" in sl or " datetime(" in sl: feats.append("uses date functions") if "cast(" in sl: feats.append("uses CAST") return "; ".join(feats) if feats else "(plain SELECT)" def render(sample, t, schema_text): sql_fixed = (t.get("fixed_sql") or "").strip() sql = sql_fixed or (t.get("planner_sql") or "").strip() if not sql: return None is_correct = bool(t.get("is_fixed_correct") if sql_fixed else t.get("is_planner_correct")) ex = exec_str_for(sample["db_path"], sql) label = "YES" if is_correct else "NO" planner_out = (t.get("planner_output") or "").strip() # Extract Goal / Final SQL line if present planner_excerpt = safe_truncate(re.sub(r"\s+", " ", planner_out), 400) prompt = ENRICHED_PROMPT.format( schema=schema_text, question=sample.get("question", ""), evidence=sample.get("evidence", "") or "None", sql=safe_truncate(sql, 700), exec_result=safe_truncate(ex, 260), struct=struct_features(sql), fb_select=safe_truncate(t.get("fb_select") or "None", 180), fb_condition=safe_truncate(t.get("fb_condition") or "None", 180), fb_join=safe_truncate(t.get("fb_join") or "None", 180), fb_order=safe_truncate(t.get("fb_order") or "None", 180), planner_excerpt=planner_excerpt or "None", ) return { "prompt": prompt, "completion": label, "messages": [{"role": "user", "content": prompt}, {"role": "assistant", "content": label}], "question": sample.get("question", ""), "db_id": sample.get("db_id", ""), "is_yes": int(label == "YES"), "sql": sql, } def main(): ap = argparse.ArgumentParser() ap.add_argument("--input", default="eval_results/paper_SFT_VF_passAt8_bird_TRAIN.jsonl") ap.add_argument("--out", default="data/sft_selector_v8_pointwise_enriched") args = ap.parse_args() rng = random.Random(42) schema_cache = {} n_rows = 0 # Phase 1: collect jobs jobs = [] with open(args.input) as f: for line in f: line = line.strip() if not line: continue s = json.loads(line) n_rows += 1 seen = set() for t in s.get("trajectories", []): sql_fixed = (t.get("fixed_sql") or "").strip() sql = sql_fixed or (t.get("planner_sql") or "").strip() if not sql: continue norm = re.sub(r"\s+", " ", sql.lower()) if norm in seen: continue seen.add(norm) jobs.append((s, t)) print(f"questions: {n_rows}, jobs: {len(jobs)}", flush=True) for s, _ in jobs: key = s["db_id"] if key not in schema_cache: schema_cache[key] = safe_truncate(render_rich_schema(s, split="train"), MAX_SCHEMA_CHARS) records = [] n_yes = n_no = 0 n_done = 0 def _job(it): s, t = it return render(s, t, schema_cache[s["db_id"]]) with ThreadPoolExecutor(max_workers=32) as exe: futs = [exe.submit(_job, it) for it in jobs] for fut in as_completed(futs): try: r = fut.result() except Exception: continue n_done += 1 if r is None: continue records.append(r) if r["is_yes"]: n_yes += 1 else: n_no += 1 if n_done % 2000 == 0: print(f" rendered {n_done}/{len(jobs)} records={len(records)} (Y={n_yes}, N={n_no})", flush=True) print(f"\nTotal: {len(records)} (Y={n_yes}, N={n_no})", flush=True) # Inject gold candidate SQL as additional YES record per question print("Injecting gold candidates...", flush=True) # group by question -> sample by_q = {} for r in records: by_q.setdefault((r["question"], r["db_id"]), []).append(r) # Use raw_rows pass gold_added = 0 with open(args.input) as f: for line in f: line = line.strip() if not line: continue s = json.loads(line) existing = by_q.get((s.get("question",""), s.get("db_id",""))) if not existing: continue gold_norm = re.sub(r"\s+", " ", (s.get("sql") or "").strip().lower()) if not gold_norm: continue already = any(re.sub(r"\s+", " ", r["sql"].lower()) == gold_norm for r in existing) if already: continue ex = exec_str_for(s["db_path"], s["sql"]) if ex.startswith("Error"): continue # Build a synthetic trajectory entry with empty fb_* t_synth = { "planner_sql": s["sql"], "fixed_sql": "", "is_planner_correct": True, "is_fixed_correct": False, "planner_exec_ok": True, "fb_select": "None", "fb_condition": "None", "fb_join": "None", "fb_order": "None", "planner_output": "(gold reference)", } schema_text = schema_cache[s["db_id"]] rec = render(s, t_synth, schema_text) if rec: records.append(rec) n_yes += 1 gold_added += 1 print(f"gold injected: {gold_added}", flush=True) # Balance: NO ~= 1.2x YES yes_r = [r for r in records if r["is_yes"]] no_r = [r for r in records if not r["is_yes"]] rng.shuffle(no_r) keep_no = no_r[: min(len(no_r), int(1.2 * len(yes_r)))] final = yes_r + keep_no rng.shuffle(final) print(f"balanced: {len(final)} (Y={len(yes_r)}, N={len(keep_no)})", flush=True) by_q = {} for r in final: by_q.setdefault(r["question"], []).append(r) qs = list(by_q.keys()); rng.shuffle(qs) n_test_q = max(40, len(qs) // 25) test_qs = set(qs[:n_test_q]) train, test = [], [] for q, recs in by_q.items(): (test if q in test_qs else train).extend(recs) rng.shuffle(train); rng.shuffle(test) # Drop the 'sql' helper field before saving (only used for dedup logic above) for r in train + test: r.pop("sql", None) print(f"train: {len(train)} test: {len(test)}") DatasetDict({ "train": Dataset.from_list(train), "test": Dataset.from_list(test), }).save_to_disk(args.out) print(f"SAVED: {args.out}") if __name__ == "__main__": main()