#!/bin/bash
#SBATCH --job-name=vl
#SBATCH --partition=gpu-large,gpu
#SBATCH --qos=batch-long
#SBATCH --gres=gpu:1
#SBATCH --cpus-per-task=8
#SBATCH --mem=60G
#SBATCH --time=01:30:00
#SBATCH --output=/weka/s225250685/mats-tist/slurm_logs/build_fixer_data_v7_%j.out

# v7: Critique-conditional fixer SFT data builder.
# Output completion = planner_sql if validator-OK, else gold_sql.

set -u; cd /weka/s225250685/mats-tist
set -a; source /weka/s225250685/mats-tist/.env; set +a
export HF_HOME=/weka/s225250685/Huggingface HF_HUB_CACHE=/weka/s225250685/Huggingface/hub
export DB_EXEC_API_DISABLE=1 PYTHONNOUSERSITE=1 NO_PROXY=localhost,127.0.0.1
export PYTHONPATH=/weka/s225250685/mats-tist TOKENIZERS_PARALLELISM=false

PY=/weka/s225250685/conda-envs/handbook/bin/python
VLLM=/weka/s225250685/conda-envs/handbook/bin/vllm
AH=/weka/s225250685/mats-tist/alignment-handbook

PLANNER=$AH/output/sft-planner-3B-griffith-v4
VAL_SEL=$AH/output/sft-validator-sel-llama1b-paper-v1
VAL_COND=$AH/output/sft-validator-cond-llama1b-paper-v1

LOG=/weka/s225250685/mats-tist/slurm_logs/build_fixer_data_v7_${SLURM_JOB_ID}.log
: > "$LOG"

kill_vllm() { pkill -9 -f "vllm serve" 2>/dev/null || true; sleep 5; }
trap kill_vllm EXIT
wait_url() { for i in {1..180}; do curl --noproxy '*' -fs "$1" >/dev/null 2>&1 && return 0; sleep 5; done; return 1; }

echo "[$(date)] Booting planner + val-sel + val-cond..." | tee -a "$LOG"

$VLLM serve "$PLANNER" --served-model-name planner --port 8200 \
    --dtype bfloat16 --gpu-memory-utilization 0.40 \
    --enforce-eager --max-model-len 8192 > "${LOG}.p" 2>&1 &
wait_url http://localhost:8200/v1/models && echo "[$(date)]  planner ready" | tee -a "$LOG"

$VLLM serve "$VAL_SEL" --served-model-name validator --port 8201 \
    --dtype bfloat16 --gpu-memory-utilization 0.15 \
    --enforce-eager --max-model-len 6144 > "${LOG}.vs" 2>&1 &
wait_url http://localhost:8201/v1/models && echo "[$(date)]  val-sel ready" | tee -a "$LOG"

$VLLM serve "$VAL_COND" --served-model-name validator --port 8204 \
    --dtype bfloat16 --gpu-memory-utilization 0.15 \
    --enforce-eager --max-model-len 6144 > "${LOG}.vc" 2>&1 &
wait_url http://localhost:8204/v1/models && echo "[$(date)]  val-cond ready" | tee -a "$LOG"

echo "[$(date)] === Building critique-conditional fixer SFT data (v7) ===" | tee -a "$LOG"
$PY scripts/build_fixer_critique_conditional_v7.py \
    --planner_host  http://localhost:8200 \
    --val_sel_host  http://localhost:8201 \
    --val_cond_host http://localhost:8204 \
    --K 8 --temperature 1.0 --max_questions -1 --threads 12 --seed 42 \
    --out data/hf_fixer_critique_conditional_v7 \
    2>&1 | tee -a "$LOG.build"

echo "[$(date)] === DONE ===" | tee -a "$LOG"
ls -la data/hf_fixer_critique_conditional_v7/ 2>&1 | tee -a "$LOG"
kill_vllm