pub const DEFAULT_EMBEDDING_MODEL: &str = "katanemo/bge-large-en-v1.5-onnx"; pub const DEFAULT_INTENT_MODEL: &str = "katanemo/deberta-base-nli-onnx"; pub const DEFAULT_PROMPT_TARGET_THRESHOLD: f64 = 0.8; pub const DEFAULT_HALLUCINATED_THRESHOLD: f64 = 0.1; pub const RATELIMIT_SELECTOR_HEADER_KEY: &str = "x-arch-ratelimit-selector"; pub const SYSTEM_ROLE: &str = "system"; pub const USER_ROLE: &str = "user"; pub const GPT_35_TURBO: &str = "gpt-3.5-turbo"; pub const ARC_FC_CLUSTER: &str = "arch_fc"; pub const ARCH_FC_REQUEST_TIMEOUT_MS: u64 = 120000; // 2 minutes pub const MODEL_SERVER_NAME: &str = "model_server"; pub const ARCH_ROUTING_HEADER: &str = "x-arch-llm-provider"; pub const ARCH_MESSAGES_KEY: &str = "arch_messages"; pub const ARCH_PROVIDER_HINT_HEADER: &str = "x-arch-llm-provider-hint"; pub const CHAT_COMPLETIONS_PATH: &str = "v1/chat/completions"; pub const ARCH_STATE_HEADER: &str = "x-arch-state"; pub const ARCH_FC_MODEL_NAME: &str = "Arch-Function-1.5B";