experiment/scripts/run_job_validate.pbs

#!/bin/bash
#PBS -N validate
#PBS -l select=1:ngpus=2
#PBS -l walltime=02:00:00
#PBS -q ic102
#PBS -P 71001002
#PBS -j oe

cd ${PBS_O_WORKDIR}

if [ -d "venv" ]; then
    source venv/bin/activate
elif [ -d "env" ]; then
    source env/bin/activate
fi

export PATH=$HOME/.local/bin:$PATH
export HF_HOME=${PBS_O_WORKDIR}/hf_cache
export PYTHONPATH="${PBS_O_WORKDIR}:${PYTHONPATH:-}"
export TORCHINDUCTOR_CACHE_DIR="${HOME}/scratch/.cache/torchinductor"
export TRITON_CACHE_DIR="${HOME}/scratch/.cache/triton"

mkdir -p logs/experiment
LOGFILE="logs/experiment/validate_${PBS_JOBID}.log"

echo "========================================" | tee -a "${LOGFILE}"
echo "Job: Validate (CLIPScore + LLM Judge)" | tee -a "${LOGFILE}"
echo "Job ID: ${PBS_JOBID}" | tee -a "${LOGFILE}"
echo "Node: $(hostname)" | tee -a "${LOGFILE}"
echo "GPUs: 2" | tee -a "${LOGFILE}"
echo "Started: $(date)" | tee -a "${LOGFILE}"
echo "========================================" | tee -a "${LOGFILE}"

echo "" | tee -a "${LOGFILE}"
echo "GPU Info:" | tee -a "${LOGFILE}"
nvidia-smi --query-gpu=index,name,memory.total --format=csv | tee -a "${LOGFILE}"
echo "CUDA_VISIBLE_DEVICES(raw): ${CUDA_VISIBLE_DEVICES:-<unset>}" | tee -a "${LOGFILE}"

BASE_MODEL="llava-hf/llava-1.5-7b-hf"
VAL_CSV="/home/users/ntu/cong045/scratch/testing/hallucination/CC3M-Dataset/data-qwenvl32b/bathroom_toilet_labels.csv"
VAL_IMAGE_DIR="CC3M-Dataset/cc3m_images/train"
OUTPUT_DIR="./step4_outputs"
NUM_PER_CATEGORY=50
MAX_NEW_TOKENS=300
MENTION_METHOD="both"
JUDGE_DEVICE="cuda:1"
JUDGE_MODEL="Qwen/Qwen3-VL-32B-Instruct"
INFERENCE_BACKEND="transformers"
VLLM_BATCH_SIZE=64
VLLM_TENSOR_PARALLEL_SIZE=1
VLLM_GPU_MEMORY_UTILIZATION=0.9
VLLM_MAX_MODEL_LEN=4096
MODEL_TYPE="lora"
MODEL_PATH="./step3_lora_v3_outputs/run_20260303_234354/step_800"

echo "" | tee -a "${LOGFILE}"
echo "Running validate (model_type=${MODEL_TYPE}, mention_method=${MENTION_METHOD})..." | tee -a "${LOGFILE}"
echo "Model path: ${MODEL_PATH}" | tee -a "${LOGFILE}"

export BASE_MODEL VAL_CSV VAL_IMAGE_DIR OUTPUT_DIR
export NUM_PER_CATEGORY MAX_NEW_TOKENS MENTION_METHOD
export JUDGE_DEVICE JUDGE_MODEL MODEL_TYPE MODEL_PATH
export INFERENCE_BACKEND VLLM_BATCH_SIZE VLLM_TENSOR_PARALLEL_SIZE
export VLLM_GPU_MEMORY_UTILIZATION VLLM_MAX_MODEL_LEN

bash experiment/scripts/run_validate.sh 2>&1 | tee -a "${LOGFILE}"

EXIT_CODE=${PIPESTATUS[0]}

echo "" | tee -a "${LOGFILE}"
echo "========================================" | tee -a "${LOGFILE}"
if [ $EXIT_CODE -eq 0 ]; then
    echo "Status: SUCCESS" | tee -a "${LOGFILE}"
else
    echo "Status: FAILED (exit code $EXIT_CODE)" | tee -a "${LOGFILE}"
fi
echo "Completed: $(date)" | tee -a "${LOGFILE}"
echo "========================================" | tee -a "${LOGFILE}"

exit $EXIT_CODE