pub struct Nf4ExpertPayload<B>where
B: Backend,{
pub packed: <B as BackendTypes>::IntTensorPrimitive,
pub scales: <B as BackendTypes>::FloatTensorPrimitive,
pub codebook: <B as BackendTypes>::FloatTensorPrimitive,
pub options: Nf4GroupedOptions,
}Expand description
Actual source expert payload; frozen FP32 scales/book never become activation storage.
Fields§
§packed: <B as BackendTypes>::IntTensorPrimitiveOriginal high-nibble-first U8 bytes for flattened [E,N,K].
scales: <B as BackendTypes>::FloatTensorPrimitiveOriginal FP32 flat-block scales, including blocks crossing expert boundaries.
codebook: <B as BackendTypes>::FloatTensorPrimitiveOriginal sixteen FP32 codebook values.
options: Nf4GroupedOptionsExplicit source geometry, ownership and execution choices.
Trait Implementations§
Source§impl<B> Clone for Nf4ExpertPayload<B>
impl<B> Clone for Nf4ExpertPayload<B>
Source§fn clone(&self) -> Nf4ExpertPayload<B>
fn clone(&self) -> Nf4ExpertPayload<B>
Returns a duplicate of the value. Read more
1.0.0 (const: unstable) · Source§fn clone_from(&mut self, source: &Self)
fn clone_from(&mut self, source: &Self)
Performs copy-assignment from
source. Read moreAuto Trait Implementations§
impl<B> Freeze for Nf4ExpertPayload<B>where
<B as BackendTypes>::IntTensorPrimitive: Freeze,
<B as BackendTypes>::FloatTensorPrimitive: Freeze,
impl<B> RefUnwindSafe for Nf4ExpertPayload<B>where
<B as BackendTypes>::IntTensorPrimitive: RefUnwindSafe,
<B as BackendTypes>::FloatTensorPrimitive: RefUnwindSafe,
impl<B> Send for Nf4ExpertPayload<B>where
<B as BackendTypes>::IntTensorPrimitive: Send,
<B as BackendTypes>::FloatTensorPrimitive: Send,
impl<B> Sync for Nf4ExpertPayload<B>where
<B as BackendTypes>::IntTensorPrimitive: Sync,
<B as BackendTypes>::FloatTensorPrimitive: Sync,
impl<B> Unpin for Nf4ExpertPayload<B>where
<B as BackendTypes>::IntTensorPrimitive: Unpin,
<B as BackendTypes>::FloatTensorPrimitive: Unpin,
impl<B> UnsafeUnpin for Nf4ExpertPayload<B>where
<B as BackendTypes>::IntTensorPrimitive: UnsafeUnpin,
<B as BackendTypes>::FloatTensorPrimitive: UnsafeUnpin,
impl<B> UnwindSafe for Nf4ExpertPayload<B>where
<B as BackendTypes>::IntTensorPrimitive: UnwindSafe,
<B as BackendTypes>::FloatTensorPrimitive: UnwindSafe,
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