pub struct ModelDims {
pub num_layers: u32,
pub max_blocks_per_layer: u32,
pub num_q_heads: u16,
pub num_kv_heads: u16,
pub head_dim: u16,
pub block_size: u16,
pub model_fp: Option<NonZeroU64>,
}Expand description
Per-call dimensions describing the model the orchestrator serves.
Fields§
§num_layers: u32§max_blocks_per_layer: u32§num_q_heads: u16§num_kv_heads: u16§head_dim: u16§block_size: u16§model_fp: Option<NonZeroU64>Config-derived model fingerprint (spark-model’s ModelFingerprint,
KV convention derive_kv: blob_bytes = 0) — the per-model identity
the KV paging namespace folds (kv_paging::ns::derive_kv_ns) so two
models sharing one paging peer can never collide. None when the
loader could not derive one (or in geometry-only tests/benches);
ATLAS_KV_PAGING=1 then fails fast at connect unless
ATLAS_KV_PAGING_NS is set explicitly. Unread on every other path.
Trait Implementations§
Auto Trait Implementations§
impl Freeze for ModelDims
impl RefUnwindSafe for ModelDims
impl Send for ModelDims
impl Sync for ModelDims
impl Unpin for ModelDims
impl UnwindSafe for ModelDims
Blanket Implementations§
Source§impl<T> BorrowMut<T> for Twhere
T: ?Sized,
impl<T> BorrowMut<T> for Twhere
T: ?Sized,
Source§fn borrow_mut(&mut self) -> &mut T
fn borrow_mut(&mut self) -> &mut T
Mutably borrows from an owned value. Read more