pub enum KdaShard {
Replicated,
HeadRows,
ChannelRows,
ChannelCols,
}Expand description
How one KDA tensor maps onto TP ranks.
Variants§
Replicated
Every rank holds the whole tensor.
HeadRows
Leading dim is heads — slice by head range.
ChannelRows
Leading dim is heads * head_dim — slice by channel range.
ChannelCols
Trailing (input) dim is heads * head_dim — row-parallel GEMM, slice the
input dim, then all-reduce the output.
Trait Implementations§
impl Copy for KdaShard
impl Eq for KdaShard
impl StructuralPartialEq for KdaShard
Auto Trait Implementations§
impl Freeze for KdaShard
impl RefUnwindSafe for KdaShard
impl Send for KdaShard
impl Sync for KdaShard
impl Unpin for KdaShard
impl UnwindSafe for KdaShard
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
Source§impl<T> CloneToUninit for Twhere
T: Clone,
impl<T> CloneToUninit for Twhere
T: Clone,
§impl<Q, K> Equivalent<K> for Q
impl<Q, K> Equivalent<K> for Q
§fn equivalent(&self, key: &K) -> bool
fn equivalent(&self, key: &K) -> bool
Compare self to
key and return true if they are equal.