Struct PrefillInput 

pub struct PrefillInput {
    pub input_ids: TensorRef,
    pub attention_mask: Option<TensorRef>,
    pub position_ids: Option<TensorRef>,
    pub kv_cache: Option<Arc<dyn KvCacheHandle>>,
}

Input for prefill phase (processing the initial prompt)

Fields

input_ids: TensorRef

Input token IDs [batch_size, sequence_length]

attention_mask: Option<TensorRef>

Attention mask [batch_size, sequence_length] (optional)

position_ids: Option<TensorRef>

Position IDs [batch_size, sequence_length] (optional, for RoPE)

kv_cache: Option<Arc<dyn KvCacheHandle>>

Pre-allocated KV cache handle (optional, for paged attention)

Implementations

impl PrefillInput

pub fn new(input_ids: TensorRef) -> Self

Create new prefill input

pub fn with_kv_cache(self, kv_cache: Arc<dyn KvCacheHandle>) -> Self

Create prefill input with a pre-allocated KV cache handle.

pub fn with_attention_mask(self, mask: TensorRef) -> Self

Add attention mask

pub fn with_position_ids(self, positions: TensorRef) -> Self

Add position IDs

pub fn batch_size(&self) -> usize

Get batch size

pub fn sequence_length(&self) -> usize

Get sequence length

Trait Implementations

impl Clone for PrefillInput

fn clone(&self) -> PrefillInput

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for PrefillInput

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.