File size: 3,978 Bytes
778d47d | 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 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 | """
Rebuild Dataset C using ALL correct rollout trajectories (no per-question cap).
Previous build capped at 2 correct CoT per question → 2086 pairs.
This version uses ALL correct trajectories → ~6134 pairs (matches thanhdath's 7-9k).
prompt = griffith user_msg (rich NL schema + evidence + question) + "Planning:"
completion = full CoT from rollout (Goal -> Condition -> Tables -> Final SQL)
Match key: rollout question_lower → griffith bird_train[sample_id] question_lower
"""
import json, os, re, random
from datasets import load_dataset, Dataset, DatasetDict
ROOT = "/weka/s225250685/mats-tist"
os.chdir(ROOT)
HF_CACHE = "/weka/s225250685/Huggingface/hub"
OUT = "data/hf_planner_sft_griffith_v2"
ROLLOUT_FILES = [
"data/rollouts/scaleup_bird_train_2stage_K4.jsonl",
"data/rollouts/scaleup_bird_train_3stage_K4.jsonl",
"data/rollouts/bird_train_3stage_K4.jsonl",
"data/rollouts/iter2_bird_train_3stage_K8.jsonl",
]
# Load BIRD train and griffith prompts
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=HF_CACHE).filter(lambda x: x["model_name"]=="deepseek-reasoner")
# Build lookup: question_lower → griffith user_msg
griffith_lookup = {}
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
griffith_q = q_m.group(1).strip()
bird_q = bird_train[sid]["question"].strip()
if griffith_q.lower() == bird_q.lower():
griffith_lookup[bird_q.lower()] = {
"user_msg": user_msg,
"sample_id": sid,
"db_id": bird_train[sid].get("db_id",""),
"question": bird_q,
}
print(f"Griffith lookup: {len(griffith_lookup)} BIRD-TRAIN questions", flush=True)
# Collect ALL correct trajectories from rollout files (no cap)
rows = []
seen_cot = set() # dedup by (question, exact CoT text)
n_dup = 0
for path in ROLLOUT_FILES:
if not os.path.exists(path):
print(f" skip (missing): {path}", flush=True)
continue
print(f"Reading {path}...", flush=True)
with open(path) as f:
for line in f:
ex = json.loads(line)
q_key = ex["question"].strip().lower()
info = griffith_lookup.get(q_key)
if not info: continue
for t in ex.get("trajectories", []):
if not t.get("is_planner_correct"): continue
cot = t.get("planner_output","").strip()
if not cot: continue
# Must contain a SQL fenced block (otherwise CoT is malformed)
if "```" not in cot: continue
dedup_key = (q_key, cot)
if dedup_key in seen_cot:
n_dup += 1; continue
seen_cot.add(dedup_key)
rows.append({
"prompt": info["user_msg"].rstrip() + "\n\nPlanning:",
"completion": cot,
"sample_id": info["sample_id"],
"db_id": info["db_id"],
"question": info["question"],
})
print(f"\nTotal unique correct CoT pairs: {len(rows)}", flush=True)
print(f"Duplicates skipped: {n_dup}", flush=True)
unique_q = len(set(r["question"] for r in rows))
print(f"Unique questions covered: {unique_q}", flush=True)
print(f"Avg pairs per question: {len(rows)/unique_q:.2f}", flush=True)
# 90/10 split
random.seed(42)
random.shuffle(rows)
n_train = int(0.9 * len(rows))
DatasetDict({
"train": Dataset.from_list(rows[:n_train]),
"test": Dataset.from_list(rows[n_train:]),
}).save_to_disk(OUT)
print(f"\nSaved → {OUT} (train={n_train}, test={len(rows)-n_train})", flush=True)
|