# Embeddings Configuration
# Copy this file to .env and fill in ALL required values
# The application will fail to start if required variables are missing

# =============================================================================
# REQUIRED: Backend Selection (no defaults - must be set explicitly)
# =============================================================================

# Default backend: vllm or llamacpp
EMB_DEFAULT_BACKEND=vllm

# Enable backends (true/false)
EMB_ENABLE_VLLM=true
EMB_ENABLE_LLAMACPP=false

# External URL for OpenAPI spec (REQUIRED)
EMB_EXTERNAL_URL=http://localhost:14100

# =============================================================================
# OPTIONAL: API Server Settings
# =============================================================================

# Server binding
# EMB_HOST=0.0.0.0
# EMB_PORT=14100  # Prod: 14100, Dev: 54100

# =============================================================================
# OPTIONAL: API Authentication
# =============================================================================

# API tokens for Bearer authentication (comma-separated)
# If not set, authentication is disabled and all endpoints are public
# If set, requests to protected endpoints require: Authorization: Bearer <token>
# Health endpoints (/health, /ready) are always public
# EMB_API_TOKENS=token1,token2,token3

# =============================================================================
# OPTIONAL: vLLM Backend Configuration
# =============================================================================

# vLLM server URL (internal Docker network)
# EMB_VLLM_BASE_URL=http://localhost:54101

# vLLM model to load (Hugging Face model ID)
EMB_VLLM_MODEL=BAAI/bge-m3

# vLLM server port
# EMB_VLLM_PORT=14101

# GPU assignment for vLLM
# EMB_VLLM_GPU=0

# GPU memory utilization
# EMB_VLLM_GPU_UTIL=0.50

# Maximum model sequence length
# EMB_VLLM_MAX_LEN=8192

# =============================================================================
# OPTIONAL: llama.cpp Backend Configuration
# =============================================================================

# llama.cpp server URL (internal Docker network)
# EMB_LLAMACPP_BASE_URL=http://localhost:54110

# Directory containing model files (absolute path recommended)
MODELS_DIR=/path/to/models

# llama.cpp model file (GGUF format, filename only - must be in MODELS_DIR)
EMB_LLAMACPP_MODEL=bge-m3-q4_k_m.gguf

# llama.cpp server port
# EMB_LLAMACPP_PORT=14110

# llama.cpp context size
# EMB_LLAMACPP_CTX=8192

# llama.cpp performance settings
# EMB_LLAMACPP_THREADS=4
# EMB_LLAMACPP_PARALLEL=4

# =============================================================================
# OPTIONAL: Timeout Settings
# =============================================================================

# Request timeout for embedding calls (seconds)
# EMB_REQUEST_TIMEOUT=120.0

# Connection timeout (seconds)
# EMB_CONNECT_TIMEOUT=10.0

# =============================================================================
# OPTIONAL: Rate Limiting and Concurrency
# =============================================================================

# Requests per second limit
# EMB_RATE_LIMIT_RPS=20.0

# Maximum burst size for rate limiting
# EMB_RATE_LIMIT_BURST=40

# Maximum concurrent embedding requests
# EMB_MAX_CONCURRENT_REQUESTS=20

# =============================================================================
# OPTIONAL: Logging
# =============================================================================

# Log level: DEBUG, INFO, WARNING, ERROR, CRITICAL
# EMB_LOG_LEVEL=INFO

# Enable JSON logging format (true/false)
# EMB_LOG_JSON=false

# =============================================================================
# OPTIONAL: Shared Settings (for Docker)
# =============================================================================

# HuggingFace cache directory
# HF_CACHE_DIR=/cai2_ds_storage/hf_cache

# HuggingFace token
# HF_TOKEN=