code/slurm_logs/mega_expand_dsC_v3.sbatch

#!/bin/bash
#SBATCH --job-name=vl
#SBATCH --partition=gpu-large
#SBATCH --qos=batch-long
#SBATCH --gres=gpu:1
#SBATCH --cpus-per-task=4
#SBATCH --mem=80G
#SBATCH --time=10:00:00
#SBATCH --output=/weka/s225250685/mats-tist/slurm_logs/expand_dsC_v3_%j.out

# Expand Dataset C: run K=8 sampling with Qwen v1 SFT planner on 8316 UNCOVERED
# BIRD-TRAIN questions using GRIFFITH prompts. Keep correct trajectories.
# Goal: bring Dataset C from 4509 → ~10000+ pairs (closer to thanhdath's 7-9k).

set -u
cd /weka/s225250685/mats-tist

export HF_HOME=/weka/s225250685/Huggingface
export HF_HUB_CACHE=/weka/s225250685/Huggingface/hub
export DB_EXEC_API_DISABLE=1
export PYTHONNOUSERSITE=1
export NO_PROXY=localhost,127.0.0.1
export PYTHONPATH=/weka/s225250685/mats-tist
export TOKENIZERS_PARALLELISM=false

PY=/weka/s225250685/conda-envs/handbook/bin/python
VLLM=/weka/s225250685/conda-envs/handbook/bin/vllm

QWEN_V1=/weka/s225250685/mats-tist/alignment-handbook/output/planner-v1-qwen3b-griffith-sft
LOG=/weka/s225250685/mats-tist/slurm_logs/expand_dsC_v3_${SLURM_JOB_ID}.log
: > "$LOG"

nvidia-smi --query-gpu=name,memory.total --format=csv,noheader | tee -a "$LOG"
kill_vllm() { pkill -9 -f "vllm serve" 2>/dev/null || true; sleep 3; }
trap kill_vllm EXIT

echo "==== [A] serving Qwen v1 SFT planner ====" | tee -a "$LOG"
$VLLM serve "$QWEN_V1" --served-model-name planner --port 8100 \
    --dtype bfloat16 --gpu-memory-utilization 0.85 \
    --enforce-eager --max-model-len 8192 > "${LOG}.p" 2>&1 &

for i in {1..180}; do
    curl --noproxy '*' -fs http://localhost:8100/v1/models >/dev/null 2>&1 && break; sleep 5
done
echo "  planner READY" | tee -a "$LOG"

echo "==== [B] expanding Dataset C with griffith prompts + Qwen v1 K=8 ====" | tee -a "$LOG"

$PY - << 'PYEOF' 2>&1 | tee -a "$LOG"
import json, os, re, random, sqlite3, threading, requests
from datasets import load_dataset, load_from_disk, Dataset, DatasetDict

ROOT = "/weka/s225250685/mats-tist"; os.chdir(ROOT)
HF_CACHE = "/weka/s225250685/Huggingface/hub"

# Load existing Dataset C v2 — covered questions
existing = load_from_disk("data/hf_planner_sft_griffith_v2")
covered = set()
for split in ["train","test"]:
    for row in existing[split]:
        covered.add(row["question"].strip().lower())
print(f"Already covered: {len(covered)} questions", flush=True)

# Load griffith prompts (1106 covered + ~8200 uncovered)
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")

uncovered = []
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(): continue
    if bird_q.lower() in covered: continue
    uncovered.append({"user_msg": user_msg, "sid": sid,
                      "db_id": bird_train[sid].get("db_id",""),
                      "question": bird_q,
                      "gold_sql": bird_train[sid]["sql"],
                      "db_path": bird_train[sid].get("db_path","")})
print(f"Uncovered questions: {len(uncovered)}", flush=True)

def qwen_chat(p): return f"<|im_start|>user\n{p}<|im_end|>\n<|im_start|>assistant\n"
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(gold, pred):
    if gold is None or pred is None: return False
    def norm(rows):
        return sorted(tuple(str(v).strip().lower() if v is not None else "" for v in row) for row in rows)
    return norm(gold) == norm(pred)

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

new_rows = []
n_correct = 0; n_attempt = 0
random.seed(42); random.shuffle(uncovered)

for i, info in enumerate(uncovered):
    sid = info["sid"]; bird_q = info["question"]
    db_path = info["db_path"] or f"data/train_databases/{info['db_id']}/{info['db_id']}.sqlite"
    if not os.path.exists(db_path): continue

    planning_prompt = info["user_msg"].rstrip() + "\n\nPlanning:"
    chat_prompt = qwen_chat(planning_prompt)
    try:
        r = requests.post("http://localhost:8100/v1/completions", json={
            "model": "planner", "prompt": chat_prompt,
            "max_tokens": 1024, "temperature": 1.0, "top_p": 0.9,
            "n": 8, "seed": 100,
            "stop": ["<|im_end|>", "<|endoftext|>"],
        }, timeout=120)
        r.raise_for_status()
        outputs = [c["text"].strip() for c in r.json()["choices"]]
    except Exception as ex:
        continue
    n_attempt += 1

    gold_res, _ = safe_exec(db_path, info["gold_sql"])
    if gold_res is None: continue

    seen_cot = set()
    for cot in outputs:
        if not cot or "```" not in cot: continue
        if cot in seen_cot: continue
        pred_sql = extract_sql(cot)
        if not pred_sql: continue
        pred_res, err = safe_exec(db_path, pred_sql)
        if err or not results_match(gold_res, pred_res): continue
        seen_cot.add(cot)
        new_rows.append({"prompt": planning_prompt, "completion": cot,
                         "sample_id": sid, "db_id": info["db_id"], "question": bird_q})
        n_correct += 1

    if (i+1) % 500 == 0:
        print(f"  [{i+1}/{len(uncovered)}] new_pairs={n_correct} attempts={n_attempt}", flush=True)

print(f"\nFinal: {n_correct} new correct CoT pairs from {n_attempt} questions", flush=True)

# Merge with existing
all_rows = []
for split in ["train","test"]:
    for row in existing[split]:
        all_rows.append({k: row[k] for k in ["prompt","completion","sample_id","db_id","question"]})
all_rows.extend(new_rows)
print(f"Total Dataset C v3: {len(all_rows)} pairs (was {len(existing['train'])+len(existing['test'])})", flush=True)

random.shuffle(all_rows)
n_train = int(0.9 * len(all_rows))
DatasetDict({
    "train": Dataset.from_list(all_rows[:n_train]),
    "test":  Dataset.from_list(all_rows[n_train:]),
}).save_to_disk("data/hf_planner_sft_griffith_v3")
print(f"Saved → data/hf_planner_sft_griffith_v3 (train={n_train}, test={len(all_rows)-n_train})", flush=True)
PYEOF

echo "==== ALL_DONE ====" | tee -a "$LOG"