#!/bin/bash
#SBATCH --job-name=ctx_test
#SBATCH --account=bfqt-delta-gpu
#SBATCH --partition=gpuH200x8-interactive
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=16
#SBATCH --gres=gpu:4
#SBATCH --mem=100G
#SBATCH --time=00:20:00
#SBATCH --output=/projects/bfqt/users/yurenh2/ml-projects/personalization-user-model/ctx_test-%j.out
#SBATCH --error=/projects/bfqt/users/yurenh2/ml-projects/personalization-user-model/ctx_test-%j.err

# Small-scale contextual test: 1 profile, 15 sessions
# Testing fix: token estimation ratio changed from 4:1 to 2.5:1

cd /projects/bfqt/users/yurenh2/ml-projects/personalization-user-model/collaborativeagents
source /u/yurenh2/miniforge3/etc/profile.d/conda.sh
conda activate eval
export HF_HOME=/projects/bfqt/users/yurenh2/hf_cache/huggingface
export PYTHONPATH="/projects/bfqt/users/yurenh2/ml-projects/personalization-user-model/src:$PYTHONPATH"

PROFILE_PATH="/projects/bfqt/users/yurenh2/ml-projects/personalization-user-model/collaborativeagents/data/complex_profiles_v2/profiles_200.jsonl"
AGENT_MODEL="/projects/bfqt/users/yurenh2/ml-projects/personalization-user-model/models/llama-3.1-8b-instruct"
USER_MODEL="meta-llama/Llama-3.1-70B-Instruct"

echo "=== Contextual Test (Token Fix) ==="
echo "Fix: token estimation 4:1 -> 2.5:1"
echo "1 profile, 15 sessions"
date
nvidia-smi --query-gpu=index,name,memory.total --format=csv

# Start vLLM servers
# User simulator: GPUs 0,1
CUDA_VISIBLE_DEVICES=0,1 python -m vllm.entrypoints.openai.api_server \
    --model $USER_MODEL \
    --port 8004 --tensor-parallel-size 2 --gpu-memory-utilization 0.90 \
    --max-model-len 16384 --dtype bfloat16 --download-dir $HF_HOME &

# Agent: GPUs 2,3
CUDA_VISIBLE_DEVICES=2,3 python -m vllm.entrypoints.openai.api_server \
    --model $AGENT_MODEL \
    --port 8003 --tensor-parallel-size 2 --gpu-memory-utilization 0.90 \
    --max-model-len 16384 --dtype bfloat16 &

echo "Waiting for vLLM servers..."
for i in {1..200}; do
    if curl -s http://localhost:8004/health > /dev/null 2>&1; then
        echo "User simulator ready after $((i*5))s"
        break
    fi
    sleep 5
done
for i in {1..60}; do
    if curl -s http://localhost:8003/health > /dev/null 2>&1; then
        echo "Agent ready after $((i*5))s"
        break
    fi
    sleep 5
done
sleep 5

OUTPUT_DIR="../results/contextual_test_$(date +%Y%m%d_%H%M%S)"

echo ""
echo "============================================"
echo "Testing: contextual (with token fix)"
echo "============================================"
date

python scripts/run_experiments.py --methods contextual \
    --datasets math-hard --n-profiles 1 --n-sessions 15 --max-turns 15 \
    --use-vllm --no-batch-processing --parallel-profiles 1 \
    --output-dir $OUTPUT_DIR --profile-path $PROFILE_PATH

echo ""
echo "=== Done ==="
date

pkill -f "vllm.entrypoints" 2>/dev/null || true