pub struct Fp8KvWriteTarget {
pub kernel: KernelHandle,
pub k_pool: DevicePtr,
pub v_pool: DevicePtr,
pub block_size: u32,
pub cache_stride: u64,
pub key_stride: u32,
pub value_stride: u32,
pub slot: DevicePtr,
}Expand description
Everything the replay needs to re-run reshape_and_cache_fp8 for one layer.
Fields§
§kernel: KernelHandlereshape_and_cache_fp8 kernel handle (the layer’s reshape_cache_k).
k_pool: DevicePtrBase pointer of this layer’s K pool.
v_pool: DevicePtrBase pointer of this layer’s V pool.
block_size: u32Tokens per KV block.
cache_stride: u64Pool stride in ELEMENTS (block_size * num_kv_heads * head_dim).
key_stride: u32Source K row stride in elements, as passed to the live write.
value_stride: u32Source V row stride in elements, as passed to the live write.
slot: DevicePtrThis batch’s slot_mapping (int64 per token), staged alongside the
BF16 K/V so the replay lands on exactly the entries the window wrote.
Trait Implementations§
Source§impl Clone for Fp8KvWriteTarget
impl Clone for Fp8KvWriteTarget
Source§fn clone(&self) -> Fp8KvWriteTarget
fn clone(&self) -> Fp8KvWriteTarget
Returns a duplicate of the value. Read more
1.0.0 · Source§fn clone_from(&mut self, source: &Self)
fn clone_from(&mut self, source: &Self)
Performs copy-assignment from
source. Read moreSource§impl Debug for Fp8KvWriteTarget
impl Debug for Fp8KvWriteTarget
impl Copy for Fp8KvWriteTarget
Auto Trait Implementations§
impl Freeze for Fp8KvWriteTarget
impl RefUnwindSafe for Fp8KvWriteTarget
impl Send for Fp8KvWriteTarget
impl Sync for Fp8KvWriteTarget
impl Unpin for Fp8KvWriteTarget
impl UnwindSafe for Fp8KvWriteTarget
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