Trait Loader

source

pub trait Loader: Send + Sync {
    // Required methods
    fn load_model_from_hf(
        &self,
        revision: Option<String>,
        token_source: TokenSource,
        dtype: &dyn TryIntoDType,
        device: &Device,
        silent: bool,
        mapper: DeviceMapMetadata,
        in_situ_quant: Option<IsqType>,
        paged_attn_config: Option<PagedAttentionConfig>,
    ) -> Result<Arc<Mutex<dyn Pipeline + Sync + Send>>, Error>;
    fn load_model_from_path(
        &self,
        paths: &Box<dyn ModelPaths>,
        dtype: &dyn TryIntoDType,
        device: &Device,
        silent: bool,
        mapper: DeviceMapMetadata,
        in_situ_quant: Option<IsqType>,
        paged_attn_config: Option<PagedAttentionConfig>,
    ) -> Result<Arc<Mutex<dyn Pipeline + Sync + Send>>, Error>;
    fn get_id(&self) -> String;
    fn get_kind(&self) -> ModelKind;
}

Expand description

The Loader trait abstracts the loading process. The primary entrypoint is the load_model method.

§Example

use mistralrs_core::{Loader, TokenSource, DeviceMapMetadata, ModelDType};
use candle_core::Device;

let loader: Box<dyn Loader> = todo!();
let pipeline = loader.load_model_from_hf(
    None,
    TokenSource::CacheToken,
    &ModelDType::Auto,
    &Device::cuda_if_available(0).unwrap(),
    false,
    DeviceMapMetadata::dummy(),
    None,
    None,
).unwrap();

Required Methods§

source

fn load_model_from_hf( &self, revision: Option<String>, token_source: TokenSource, dtype: &dyn TryIntoDType, device: &Device, silent: bool, mapper: DeviceMapMetadata, in_situ_quant: Option<IsqType>, paged_attn_config: Option<PagedAttentionConfig>, ) -> Result<Arc<Mutex<dyn Pipeline + Sync + Send>>, Error>

If revision is None, then it defaults to main. If dtype is None, then it defaults to the model default (usually BF16). If model is not found on HF, will attempt to resolve locally.

source