STING Performance Quick Reference

# For Virtual Machines (Recommended)
echo "PERFORMANCE_PROFILE=vm_optimized" >> .env
docker compose restart llm-gateway

# For GPU Hardware  
echo "PERFORMANCE_PROFILE=gpu_accelerated" >> .env
docker compose restart llm-gateway

# Auto-detect (Default)
echo "PERFORMANCE_PROFILE=auto" >> .env
docker compose restart llm-gateway

# Run built-in performance test
./test_performance.sh

# Quick health check
curl http://localhost:8085/health

# Quick chat test
curl -X POST http://localhost:8081/chat/message \
  -H "Content-Type: application/json" \
  -d '{"message": "Hello", "user_id": "test"}'

# Monitor resource usage
docker stats sting-llm-gateway-1

# Check logs
docker compose logs llm-gateway --tail=20

# Check configuration
docker compose exec llm-gateway env | grep -E "(PERFORMANCE|TORCH|OMP)"

# Switch to VM optimized
PERFORMANCE_PROFILE=vm_optimized docker compose restart llm-gateway

# Enable quantization
QUANTIZATION=int8 docker compose restart llm-gateway

# Use aggressive quantization
QUANTIZATION=int4 docker compose restart llm-gateway

# Use smaller model
MODEL_NAME=phi3 docker compose restart llm-gateway

# Force all CPU cores
OMP_NUM_THREADS=auto docker compose restart llm-gateway

# Check current threading
docker compose exec llm-gateway python -c "import torch; print(f'Threads: {torch.get_num_threads()}')"

# View current profile
echo $PERFORMANCE_PROFILE

# Check environment in container
docker compose exec llm-gateway env | grep PERFORMANCE_PROFILE

# View applied settings in logs
docker compose logs llm-gateway | grep -i "performance profile"

# Remove custom settings
unset PERFORMANCE_PROFILE
unset QUANTIZATION
unset TORCH_DEVICE

# Restart with auto-detection
PERFORMANCE_PROFILE=auto docker compose restart llm-gateway

# If services won't start
docker compose down
docker compose up -d db vault kratos app frontend

# Start with minimal LLM
PERFORMANCE_PROFILE=vm_optimized \
QUANTIZATION=int4 \
docker compose up -d llm-gateway

export PERFORMANCE_PROFILE=vm_optimized
export QUANTIZATION=int8
export OMP_NUM_THREADS=auto
export MAX_TOKENS=512  # Shorter responses = faster

export PERFORMANCE_PROFILE=gpu_accelerated
export QUANTIZATION=none
export TORCH_PRECISION=fp16
export MAX_TOKENS=2048

export PERFORMANCE_PROFILE=vm_optimized
export QUANTIZATION=int8
export MAX_TOKENS=256  # Very fast responses for testing