Struct PagedAttentionConfig 

pub struct PagedAttentionConfig {
    pub attention: AttentionConfig,
    pub page_block_size: usize,
    pub max_blocks_per_seq: usize,
    pub max_num_seqs: usize,
}

Configuration for paged attention.

Fields

attention: AttentionConfig

Base attention config.

page_block_size: usize

Block size in the page table (tokens per block).

max_blocks_per_seq: usize

Maximum number of blocks per sequence.

max_num_seqs: usize

Maximum number of sequences.

Implementations

impl PagedAttentionConfig

pub fn new(page_block_size: usize) -> Self

Create config with the given block size.

pub fn with_attention(self, attention: AttentionConfig) -> Self

Set attention config.

pub fn with_max_blocks(self, max_blocks: usize) -> Self

Set max blocks per sequence.

pub fn max_context_len(&self) -> usize

Maximum context length supported.

pub fn validate(&self) -> ConfigResult<()>

Validate configuration values.

Trait Implementations

impl Clone for PagedAttentionConfig

fn clone(&self) -> PagedAttentionConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for PagedAttentionConfig

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

Formats the value using the given formatter.

impl Default for PagedAttentionConfig

fn default() -> Self

Returns the “default value” for a type.