"""Reranker service configuration (simple Python config).""" import os class RerankerConfig(object): # Server HOST = os.getenv("RERANKER_HOST", "0.0.0.0") PORT = int(os.getenv("RERANKER_PORT", 6007)) # Model MODEL_NAME = "Qwen/Qwen3-Reranker-0.6B" DEVICE = None # None -> auto (cuda if available) USE_FP16 = True BATCH_SIZE = 64 MAX_LENGTH = 512 CACHE_DIR = "./model_cache" ENABLE_WARMUP = True # Request limits MAX_DOCS = 1000 # Output NORMALIZE = True CONFIG = RerankerConfig()