golem-flask-backend / llava_phi3_config.txt
mememechez's picture
Deploy final cleaned source code
ca28016
LLaVA-Phi3:3.8B Configuration for QWEN2Golem
==============================================
MODEL CONFIGURATION:
- Model: llava-phi3:3.8b
- Parameters: 3.8 billion
- Type: Multimodal (text + images)
- Provider: Ollama
HARDWARE REQUIREMENTS:
- RAM: Minimum 4GB free (7GB+ total recommended)
- GPU: NVIDIA GPU with CUDA support (optional, will use CPU if no GPU)
- Storage: ~4GB for model download
OPTIMIZED SETTINGS:
- GOLEM_AETHER_MAX_PATTERNS: 100,000 (reduced from 250k for stability)
- GOLEM_AETHER_SAMPLE_RATIO: 0.3 (30% sampling to save memory)
- GOLEM_MIN_FREE_GB: 4.0 GB (ensures system stability)
- CUDA: Enabled for GPU acceleration
- OMP_NUM_THREADS: 4 (parallel processing)
- PYTORCH_CUDA_ALLOC_CONF: max_split_size_mb:128 (GPU memory optimization)
USAGE:
1. Run: cd /home/chezy/Desktop/cursor/robust_zpe && ./run_all.sh
2. Or: cd /home/chezy/Desktop/cursor/robust_zpe/QWEN2Golem && ./start_llava_phi3.sh
SERVICES STARTED:
- Ollama Server (port 11434)
- Flask Golem Server (port 5000)
- Dataset Server (port 9001)
- QuantumWeaver Backend (port 9006)
- Frontend (port 9002)
TROUBLESHOOTING:
- If you get memory errors, close other applications
- If GPU not detected, model will run on CPU (slower)
- If download fails, check internet connection
- If port conflicts, kill existing processes first