#!/bin/bash
#SBATCH --job-name=run_expts
#SBATCH --account=bfqt-delta-gpu
#SBATCH --partition=gpuH200x8
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=32
#SBATCH --gres=gpu:h200:8
#SBATCH --mem=400G
#SBATCH --time=48:00:00
#SBATCH --output=logs/run_expts_%j.out
#SBATCH --error=logs/run_expts_%j.err

# Run experiments with models loaded locally
# This job needs 8 GPUs:
# - 4 GPUs for 70B judge model
# - 2 GPUs for PersonalizedLLM (embedder, reranker, extractor, main LLM)
# - Reserve for headroom

set -e

cd /projects/bfqt/users/yurenh2/ml-projects/personalization-user-model

# Create logs and results directories
mkdir -p collaborativeagents/slurm/logs
mkdir -p collaborativeagents/results

echo "Starting experiments at $(date)"
echo "Job ID: $SLURM_JOB_ID"
echo "Node: $SLURMD_NODENAME"
echo "GPUs: $CUDA_VISIBLE_DEVICES"

# Activate environment
source /u/yurenh2/miniforge3/etc/profile.d/conda.sh
conda activate eval

# Check GPU availability
nvidia-smi

# Add project to path
export PYTHONPATH="${PWD}/src:${PWD}/collaborativeagents:${PYTHONPATH}"

# Run experiments
cd collaborativeagents/scripts

# Quick test first (2 profiles, 2 sessions)
echo "Running quick test..."
python run_experiments.py \
    --methods rag_vector \
    --datasets math-500 \
    --n-profiles 2 \
    --n-sessions 2 \
    --profile-path ../data/complex_profiles_v2/profiles_100.jsonl \
    --output-dir ../results/test_$(date +%Y%m%d_%H%M%S)

# Full run (uncomment when ready)
# echo "Running full experiments..."
# python run_experiments.py \
#     --methods vanilla,all_memory,rag,rag_vector \
#     --datasets math-500,gpqa,aime \
#     --n-profiles 100 \
#     --n-sessions 20 \
#     --profile-path ../data/complex_profiles_v2/profiles_100.jsonl \
#     --output-dir ../results/full_$(date +%Y%m%d_%H%M%S)

echo "Experiments completed at $(date)"