# Rerank 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
RERANK_DEFAULT_BACKEND=vllm

# Enable backends (true/false)
RERANK_ENABLE_VLLM=true
RERANK_ENABLE_LLAMACPP=false

# External URL for OpenAPI spec (REQUIRED)
RERANK_EXTERNAL_URL=http://localhost:14200

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

# Server binding
# RERANK_HOST=0.0.0.0
# RERANK_PORT=14200  # Prod: 14200, Dev: 54200

# =============================================================================
# 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
# RERANK_API_TOKENS=token1,token2,token3

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

# vLLM server URL (internal Docker network)
# RERANK_VLLM_BASE_URL=http://localhost:54201

# vLLM model to load (Hugging Face model ID for cross-encoder)
RERANK_VLLM_MODEL=BAAI/bge-reranker-v2-m3

# vLLM server port
# RERANK_VLLM_PORT=14201

# GPU assignment for vLLM
# RERANK_VLLM_GPU=0

# GPU memory utilization
# RERANK_VLLM_GPU_UTIL=0.50

# Maximum model sequence length
# RERANK_VLLM_MAX_LEN=8192

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

# llama.cpp server URL (internal Docker network)
# RERANK_LLAMACPP_BASE_URL=http://localhost:54210

# 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)
RERANK_LLAMACPP_MODEL=bge-reranker-v2-m3-q4_k_m.gguf

# llama.cpp server port
# RERANK_LLAMACPP_PORT=14210

# llama.cpp context size
# RERANK_LLAMACPP_CTX=8192

# llama.cpp performance settings
# RERANK_LLAMACPP_THREADS=4
# RERANK_LLAMACPP_PARALLEL=4

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

# Request timeout for rerank calls (seconds)
# RERANK_REQUEST_TIMEOUT=120.0

# Connection timeout (seconds)
# RERANK_CONNECT_TIMEOUT=10.0

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

# Requests per second limit
# RERANK_RATE_LIMIT_RPS=20.0

# Maximum burst size for rate limiting
# RERANK_RATE_LIMIT_BURST=50

# Maximum concurrent rerank requests
# RERANK_MAX_CONCURRENT_RERANKS=20

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

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

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

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

# HuggingFace cache directory
# HF_CACHE_DIR=/cai2_ds_storage/hf_cache

# HuggingFace token
# HF_TOKEN=