pub enum DsaShard {
Replicated,
HeadRows,
HeadCols,
}Expand description
How one DSA tensor maps onto TP ranks.
Variants§
Replicated
Every rank holds the whole tensor: the indexer, the latent KV projection, and every low-rank down-projection / norm.
HeadRows
Leading dim is heads * per_head — slice by this rank’s head range.
HeadCols
Trailing (input) dim is heads * v_head_dim — row-parallel GEMM, slice the
input dim, then all-reduce.
Trait Implementations§
impl Copy for DsaShard
impl Eq for DsaShard
impl StructuralPartialEq for DsaShard
Auto Trait Implementations§
impl Freeze for DsaShard
impl RefUnwindSafe for DsaShard
impl Send for DsaShard
impl Sync for DsaShard
impl Unpin for DsaShard
impl UnwindSafe for DsaShard
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.