Struct FerrotensorHub 

pub struct FerrotensorHub { /* private fields */ }

Main entry point for loading models from HuggingFace Hub.

Currently supports loading from local directories. Download models using huggingface-cli download or the Python huggingface_hub library.

Example

let hub = FerrotensorHub::new();
let config = hub.load_config("meta-llama/Llama-3.2-1B")?;
println!("Model has {} layers", config.num_hidden_layers);

Implementations

impl FerrotensorHub

pub fn new() -> Self

Create a new FerrotensorHub instance with default cache directory.

Default cache: ~/.cache/huggingface/hub/models--{org}--{model}

pub fn with_cache_dir(cache_dir: PathBuf) -> Self

Create a FerrotensorHub with a custom cache directory.

pub fn get_model_path(&self, model_id: &str) -> PathBuf

Get the path to a cached model.

Compatible with HuggingFace cache format.

pub fn find_latest_snapshot(&self, model_id: &str) -> HubResult<PathBuf>

Find the latest snapshot of a model.

pub fn load_config(&self, model_id: &str) -> HubResult<ModelConfig>

Load model configuration.

pub fn is_cached(&self, model_id: &str) -> bool

Check if a model is cached locally.

pub fn list_cached(&self) -> Vec<String>

List all cached models.

pub fn get_weight_files(&self, model_id: &str) -> HubResult<Vec<PathBuf>>

Get safetensors file paths for a model.

Trait Implementations

impl Default for FerrotensorHub

fn default() -> Self

Returns the “default value” for a type.