from pydantic_settings import BaseSettings class Settings(BaseSettings): REPO_ID: str = "Qwen/Qwen2.5-Coder-7B-Instruct-GGUF" FILENAME: str = "qwen2.5-coder-7b-instruct-q5_k_m.gguf" CONTEXT_SIZE: int = 8192 DEFAULT_MAX_TOKENS: int = 4096 DEFAULT_TEMP: float = 0.4 N_THREADS: int = 2 N_GPU_LAYERS: int = 0 settings = Settings()