#!/bin/bash
#SBATCH --job-name=test_real
#SBATCH --account=bfqt-delta-gpu
#SBATCH --partition=gpuH200x8-interactive
#SBATCH --gres=gpu:4
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=32
#SBATCH --mem=256G
#SBATCH --time=00:30:00
#SBATCH --output=/projects/bfqt/users/yurenh2/ml-projects/personalization-user-model/test_real-%j.out
#SBATCH --error=/projects/bfqt/users/yurenh2/ml-projects/personalization-user-model/test_real-%j.err

set -e
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="${PWD}:${PWD}/scripts:${PWD}/../src:${PYTHONPATH}"

MODEL_8B="/projects/bfqt/users/yurenh2/ml-projects/personalization-user-model/models/llama-3.1-8b-instruct"
PROFILE_PATH="/projects/bfqt/users/yurenh2/ml-projects/personalization-user-model/collaborativeagents/data/complex_profiles_v2/profiles_200.jsonl"

echo "=== Real Speed Test (5 profiles, 5 sessions) ==="
date
nvidia-smi --query-gpu=index,name,memory.total --format=csv

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

# Test 1: contextual (vLLM-based)
echo ""
echo "=== TEST 1: contextual (2 vLLM servers) ==="
CUDA_VISIBLE_DEVICES=0,1 python -m vllm.entrypoints.openai.api_server \
    --model $MODEL_8B --port 8004 --tensor-parallel-size 2 \
    --gpu-memory-utilization 0.90 --max-model-len 8192 \
    --disable-log-requests --dtype bfloat16 &
CUDA_VISIBLE_DEVICES=2,3 python -m vllm.entrypoints.openai.api_server \
    --model $MODEL_8B --port 8003 --tensor-parallel-size 2 \
    --gpu-memory-utilization 0.90 --max-model-len 8192 \
    --disable-log-requests --dtype bfloat16 &

for i in $(seq 1 120); do
    u=$(curl -s http://localhost:8004/health > /dev/null 2>&1 && echo 1 || echo 0)
    a=$(curl -s http://localhost:8003/health > /dev/null 2>&1 && echo 1 || echo 0)
    [ "$u" = "1" ] && [ "$a" = "1" ] && echo "Ready after $((i*2))s" && break
    sleep 2
done

time python scripts/run_experiments.py \
    --methods contextual \
    --datasets math-hard \
    --n-profiles 5 --n-sessions 5 --max-turns 10 \
    --use-vllm --parallel-profiles 5 \
    --output-dir ../results/speed_test \
    --profile-path "$PROFILE_PATH"

pkill -f "vllm.entrypoints" 2>/dev/null || true
sleep 5

# Test 2: all_memory (vLLM + transformers)
echo ""
echo "=== TEST 2: all_memory (vLLM user + transformers adapter) ==="
CUDA_VISIBLE_DEVICES=0,1 python -m vllm.entrypoints.openai.api_server \
    --model $MODEL_8B --port 8004 --tensor-parallel-size 2 \
    --gpu-memory-utilization 0.90 --max-model-len 8192 \
    --disable-log-requests --dtype bfloat16 &

for i in $(seq 1 120); do
    curl -s http://localhost:8004/health > /dev/null 2>&1 && echo "Ready after $((i*2))s" && break
    sleep 2
done

CUDA_VISIBLE_DEVICES=2,3 time python scripts/run_experiments.py \
    --methods all_memory \
    --datasets math-hard \
    --n-profiles 5 --n-sessions 5 --max-turns 10 \
    --use-vllm --parallel-profiles 5 \
    --output-dir ../results/speed_test \
    --profile-path "$PROFILE_PATH"

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

echo ""
echo "=== DONE ==="
date