single_gpu_setup.yaml

# Single GPU Setup Configuration
# ThemisDB v1.4.0-alpha
# Use case: Development, small-scale training, inference

# Server Configuration
server:
  host: 0.0.0.0
  port: 8080
  workers: 4
  max_connections: 1000
  
  # Request handling
  request_timeout: 300  # seconds
  keepalive_timeout: 75
  
  # Resource limits
  max_request_size: 100MB
  max_response_size: 100MB

# GPU Configuration
gpu:
  enabled: true
  
  # Single GPU device
  devices:
    - id: 0
      name: "GPU-0"
      memory_limit: 0.9  # Use 90% of VRAM
      compute_capability: 8.6
      
  # GPU settings
  persistence_mode: true
  compute_mode: default
  power_limit: null  # Use default
  
  # Memory management
  memory_pool:
    enabled: true
    initial_size_mb: 2048
    max_split_size_mb: 512
  
  # VRAM optimization
  vram:
    secure_clear: true
    unified_memory: false
    oom_protection:
      enabled: true
      reserved_memory_mb: 1024

# LLM Configuration
llm:
  enabled: true
  backend: cuda
  
  # Model settings
  model_path: /models/llama-2-7b-chat.gguf
  model_type: llama
  
  # Context and batch
  context_length: 4096
  batch_size: 512
  n_gpu_layers: 35  # All layers on GPU
  
  # Performance
  threads: 8
  use_mmap: true
  use_mlock: false
  
  # Inference settings
  inference:
    temperature: 0.7
    top_p: 0.9
    top_k: 40
    repeat_penalty: 1.1
    
    # KV cache
    kv_cache:
      enabled: true
      max_tokens: 4096
      
# LoRA Configuration
lora:
  enabled: true
  adapter_path: /adapters
  max_adapters: 4
  gpu_enabled: true
  
  # Adapter management
  preload_adapters: []
  cache_adapters: true
  
  # LoRA settings
  rank: 16
  alpha: 32
  dropout: 0.05

# Training Configuration
training:
  enabled: true
  
  # Batch configuration
  batch_size: 16
  micro_batch_size: 8
  gradient_accumulation_steps: 2
  
  # Precision
  precision: fp16
  mixed_precision:
    enabled: true
    loss_scale: dynamic
  
  # Optimization
  optimizer: adamw
  learning_rate: 3e-4
  weight_decay: 0.01
  
  # Learning rate schedule
  lr_schedule:
    type: cosine_with_warmup
    warmup_steps: 500
    min_lr: 3e-5
  
  # Gradient management
  gradient_clipping:
    enabled: true
    max_norm: 1.0
  
  gradient_checkpointing:
    enabled: true
    checkpoint_segments: 2
  
  # Checkpoints
  checkpoint:
    enabled: true
    save_interval: 500
    max_keep: 3
    output_dir: /data/checkpoints
    compression: true
    async_save: true
    
  # Validation
  validation:
    enabled: true
    interval: 100
    split: 0.1

# Inference Configuration
inference:
  enabled: true
  
  # Batching
  batch_size: 1
  continuous_batching:
    enabled: false
    max_wait_time_ms: 50
  
  # Performance
  precision: fp16
  
  # Generation settings
  max_tokens: 512
  stream: true

# Storage Configuration
storage:
  # Data directory
  data_dir: /data/themisdb
  
  # RocksDB settings
  rocksdb:
    max_open_files: 1000
    max_background_jobs: 4
    write_buffer_size: 64MB
    max_write_buffer_number: 3
    
    # Compression
    compression: lz4
    
    # Cache
    block_cache_size: 512MB
    
  # Encryption (optional)
  encryption:
    enabled: false
    algorithm: AES-256-GCM

# Network Configuration
network:
  # HTTP/REST API
  http:
    enabled: true
    port: 8080
  
  # Binary protocol
  binary:
    enabled: true
    port: 18765
  
  # TLS (optional - enable for production)
  tls:
    enabled: false
    cert_file: /etc/themisdb/certs/server.crt
    key_file: /etc/themisdb/certs/server.key

# Monitoring Configuration
metrics:
  enabled: true
  port: 4318
  path: /metrics
  
  # Collection intervals
  gpu_metrics_interval: 1s
  training_metrics_interval: 1s
  inference_metrics_interval: 100ms
  
  # Exporters
  exporters:
    - type: prometheus
      endpoint: http://localhost:4318/metrics

# Logging Configuration
logging:
  level: info
  format: json
  output: /var/log/themisdb/app.log
  
  # Rotation
  rotation:
    max_size: 100MB
    max_age: 7
    max_backups: 5
    compress: true
  
  # Audit logging
  audit:
    enabled: true
    output: /var/log/themisdb/audit.log

# Security Configuration
security:
  # Authentication
  authentication:
    enabled: false  # Enable for production
    method: api_key
  
  # Authorization
  authorization:
    enabled: false  # Enable for production
  
  # GPU access control
  gpu:
    access_control:
      enabled: false
      mode: permissive

# Backup Configuration
backup:
  enabled: true
  
  # Schedule
  schedule:
    full_backup: "0 2 * * 0"  # Weekly on Sunday
    incremental: "0 2 * * 1-6"  # Daily
  
  # Retention
  retention:
    full: 2
    incremental: 7
  
  # Destination
  destination: /backup/themisdb

# Resource Limits
limits:
  # Memory
  max_memory_mb: 32768  # 32GB
  
  # Disk
  max_disk_usage_gb: 500
  
  # Connections
  max_concurrent_requests: 100
  max_queue_size: 1000

# Development Settings
development:
  debug_mode: false
  profiling: false
  hot_reload: false
  
  # Testing
  test_mode: false

# Example Usage:
# 1. Copy this file to /etc/themisdb/config.yaml
# 2. Adjust paths for your environment
# 3. Enable security features for production
# 4. Start service: sudo systemctl start themisdb