Struct LLaVA 

pub struct LLaVA {
    pub clip_vision_tower: ClipVisionTower,
    pub image_newline: Tensor,
    pub mm_projector: MMProjector,
    pub llama: Llama,
    /* private fields */
}

Fields

clip_vision_tower: ClipVisionTower

image_newline: Tensor

mm_projector: MMProjector

llama: Llama

Implementations

impl LLaVA

pub fn load( vb: VarBuilder<'_>, config: &LLaVAConfig, clip_vision_config: Option<ClipVisionConfig>, ) -> Result<Self>

pub fn encode_images(&self, x: &Tensor) -> Result<Tensor>

pub fn prepare_inputs_labels_for_multimodal( &self, input_ids: &Tensor, images: &[Tensor], image_sizes: &[(u32, u32)], ) -> Result<Tensor>

pub fn forward( &self, input_embeds: &Tensor, position_id: usize, cache: &mut Cache, ) -> Result<Tensor>