Class Args

cake-core/src/lib.rs:150–295

#[derive(Clone, Parser, Default, Debug)]
#[command(author, version, about, long_about = None)]
pub struct Args {
    /// GPU device index.
    #[arg(long, default_value_t = 0)]
    pub device: usize,
    /// Mode (set by subcommand, not directly by user).
    #[arg(skip)]
    pub mode: Mode,
    /// Worker name.
    #[arg(long)]
    pub name: Option<String>,
    /// Binding address and port for workers.
    #[arg(long, default_value = "0.0.0.0:10128")]
    pub address: String,
    /// Enable OpenAI compatible chat completion API.
    #[arg(long)]
    pub api: Option<String>,
    /// Path to model directory, or HuggingFace repo ID (e.g., evilsocket/Qwen2.5-Coder-1.5B-Instruct).
    #[arg(long, default_value = "./cake-data/Meta-Llama-3-8B/")]
    pub model: String,
    /// Topology file.
    #[arg(long)]
    pub topology: Option<String>,
    /// The initial prompt (set programmatically from the positional argument in `cake run`).
    #[arg(skip = String::from("The sky is blue because "))]
    pub prompt: String,
    /// The system prompt.
    #[arg(long, default_value = "You are a helpful AI assistant.")]
    pub system_prompt: String,
    /// The seed to use when generating random samples.
    #[arg(long, default_value_t = 299792458)]
    pub seed: u64,
    /// The length of the sample to generate (in tokens).
    #[arg(short = 'n', long, default_value_t = 2048)]
    pub sample_len: usize,
    /// The temperature used to generate samples.
    #[arg(long, default_value_t = 1.0)]
    pub temperature: f64,
    /// Nucleus sampling probability cutoff.
    #[arg(long)]
    pub top_p: Option<f64>,
    /// Only sample among the top K samples.
    #[arg(long)]
    pub top_k: Option<usize>,
    /// Penalty to be applied for repeating tokens, 1. means no penalty.
    #[arg(long, default_value_t = 1.1)]
    pub repeat_penalty: f32,
    /// The context size to consider for the repeat penalty.
    #[arg(long, default_value_t = 128)]
    pub repeat_last_n: usize,
    /// Use different dtype than f16
    #[arg(long)]
    pub dtype: Option<String>,

    /// Cluster key for zero-config mDNS discovery and PSK authentication.
    /// When set on both master and workers, enables automatic discovery,
    /// layer assignment, and model data push without topology files.
    #[arg(long, env = "CAKE_CLUSTER_KEY")]
    pub cluster_key: Option<String>,