code/slurm_logs/eval_after_orpo_iter1.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=100G
#SBATCH --time=08:00:00
#SBATCH --output=/weka/s225250685/mats-tist/slurm_logs/eval_after_orpo_iter1_%j.out

# After ORPO iter1, eval 3 pipeline configs to compare collab vs independent:
#   COLLAB:      planner-orpo + val-sel-collab + val-cond-collab + fixer-orpo + selector(SFT)
#   INDEPENDENT: planner-orpo + val-sel-indep  + val-cond-indep  + fixer-orpo + selector(SFT)
#   pass@1 greedy on planner only (sanity)
# Show: oracle pass@8, selector EX for COLLAB and INDEPENDENT side-by-side

set -u; cd /weka/s225250685/mats-tist
set -a; source .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=$AH/output/orpo-planner-iter1
VAL_SEL_C=$AH/output/orpo-val-sel-collab-iter1
VAL_SEL_I=$AH/output/orpo-val-sel-indep-iter1
VAL_COND_C=$AH/output/orpo-val-cond-collab-iter1
VAL_COND_I=$AH/output/orpo-val-cond-indep-iter1
FIXER=$AH/output/orpo-fixer-iter1
SELECTOR=$AH/output/sft-selector-3B-griffith-v5

LOG=/weka/s225250685/mats-tist/slurm_logs/eval_after_orpo_iter1_${SLURM_JOB_ID}.log
: > "$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; }

# Wait for all 6 ORPO models
for i in $(seq 1 96); do
    ok=0
    for p in "$PLANNER" "$VAL_SEL_C" "$VAL_SEL_I" "$VAL_COND_C" "$VAL_COND_I" "$FIXER"; do
        [ -f "$p/config.json" ] && ok=$((ok+1))
    done
    echo "  [$i*5min] ORPO models ready: $ok/6" | tee -a "$LOG"
    [ "$ok" -eq 6 ] && break
    sleep 300
done

run_pipeline() {
    local LABEL=$1; local VSEL=$2; local VCOND=$3
    kill_vllm
    $VLLM serve "$PLANNER" --served-model-name planner --port 8100 \
        --dtype bfloat16 --gpu-memory-utilization 0.30 \
        --enforce-eager --max-model-len 8192 > "${LOG}.p" 2>&1 &
    wait_url http://localhost:8100/v1/models
    $VLLM serve "$VSEL" --served-model-name validator_sel --port 8101 \
        --dtype bfloat16 --gpu-memory-utilization 0.10 \
        --enforce-eager --max-model-len 6144 > "${LOG}.vs" 2>&1 &
    wait_url http://localhost:8101/v1/models
    $VLLM serve "$VCOND" --served-model-name validator_cond --port 8104 \
        --dtype bfloat16 --gpu-memory-utilization 0.10 \
        --enforce-eager --max-model-len 6144 > "${LOG}.vc" 2>&1 &
    wait_url http://localhost:8104/v1/models
    $VLLM serve "$FIXER" --served-model-name fixer --port 8102 \
        --dtype bfloat16 --gpu-memory-utilization 0.12 \
        --enforce-eager --max-model-len 6144 > "${LOG}.f" 2>&1 &
    wait_url http://localhost:8102/v1/models

    OUT=eval_results/orpo1_${LABEL}_passAt8_bird_dev.jsonl
    rm -f "$OUT"
    $PY scripts/run_pipeline_rollouts.py \
        --input_file data/sft_bird_with_evidence_dev_text2sql.json \
        --output_file "$OUT" \
        --planner_host http://localhost:8100 --planner_format qwen \
        --validator_host none \
        --validator_sel_host http://localhost:8101 \
        --validator_cond_host http://localhost:8104 \
        --fixer_host http://localhost:8102 \
        --fixer_gate_exec_ok \
        --griffith_prompts \
        --K 8 --K_val 1 --K_fix 1 \
        --temperature 1.0 --top_p 0.9 \
        --max_planner_tokens 1024 --max_validator_tokens 384 --max_fixer_tokens 512 \
        --max_questions -1 --n_threads 4 2>&1 | tee -a "$LOG"
    $PY scripts/compute_bestofn_metrics.py "$OUT" orpo1_${LABEL}_passAt8 2>&1 | tee -a "$LOG"
}

# COLLAB pipeline
echo "==== [COLLAB] planner-orpo + val-sel-collab + val-cond-collab + fixer-orpo ====" | tee -a "$LOG"
run_pipeline COLLAB "$VAL_SEL_C" "$VAL_COND_C"

# INDEPENDENT pipeline
echo "==== [INDEP] planner-orpo + val-sel-indep + val-cond-indep + fixer-orpo ====" | tee -a "$LOG"
run_pipeline INDEP "$VAL_SEL_I" "$VAL_COND_I"

# Selector EX on both
kill_vllm
$VLLM serve "$SELECTOR" --served-model-name selector --port 8103 \
    --dtype bfloat16 --gpu-memory-utilization 0.85 \
    --enforce-eager --max-model-len 4096 > "${LOG}.sel" 2>&1 &
wait_url http://localhost:8103/v1/models

for L in COLLAB INDEP; do
    $PY scripts/compute_bestofn_with_selector.py \
        eval_results/orpo1_${L}_passAt8_bird_dev.jsonl orpo1_${L}_selectorEX \
        --selector_host http://localhost:8103 --row_preview 2>&1 | tee -a "$LOG"
done

echo "==== ALL_DONE ====" | tee -a "$LOG"
echo "" | tee -a "$LOG"
echo "=== FINAL COMPARISON (ORPO iter1) ===" | tee -a "$LOG"
grep -E "orpo1_COLLAB_|orpo1_INDEP_|oracle|selector" "$LOG" | tail -20 | tee -a "$LOG"