#!/bin/bash
#SBATCH --job-name=reflection_grpo
#SBATCH --account=bfqt-delta-gpu
#SBATCH --partition=gpuA100x4
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=16
#SBATCH --gres=gpu:nvidia_a100:4
#SBATCH --mem=200G
#SBATCH --time=48:00:00
#SBATCH --output=logs/reflection_grpo_%j.out
#SBATCH --error=logs/reflection_grpo_%j.err

set -e
cd /projects/bfqt/users/yurenh2/ml-projects/personalization-user-model
mkdir -p collaborativeagents/slurm/logs collaborativeagents/results

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

export HF_HOME=/projects/bfqt/users/yurenh2/hf_cache/huggingface
export PYTHONPATH="${PWD}/src:${PWD}/collaborativeagents:${PYTHONPATH}"
export NCCL_P2P_DISABLE=1

cd collaborativeagents/scripts
python run_experiments.py \
    --methods reflection_grpo \
    --datasets mmlu,aime,math-hard,humaneval \
    --n-profiles 30 \
    --n-sessions 20 \
    --profile-path ../data/complex_profiles_v2/profiles_100.jsonl \
    --output-dir ../results/reflection_grpo_$(date +%Y%m%d_%H%M%S)