pub struct ProjBytes {
pub packed_bytes: u64,
pub scale_bytes: u64,
}Expand description
Byte geometry of one NVFP4 projection sub-buffer inside an expert record.
NVFP4 (W4A4, group_size 16) stores each projection as two device buffers:
packed— E2M1 nibbles, 2 values/byte,[K/2, N]in prefill-resident (transposed) layout, sopacked_bytes = N * K / 2.scale— per-group FP8-E4M3 block scales,[K/16, N], soscale_bytes = N * K / group_size.
The two per-projection scalars (weight_scale_2, input_scale) are carried
in the record header, which is why they are absent here.
Fields§
§packed_bytes: u64§scale_bytes: u64Implementations§
Trait Implementations§
impl Copy for ProjBytes
impl Eq for ProjBytes
impl StructuralPartialEq for ProjBytes
Auto Trait Implementations§
impl Freeze for ProjBytes
impl RefUnwindSafe for ProjBytes
impl Send for ProjBytes
impl Sync for ProjBytes
impl Unpin for ProjBytes
impl UnwindSafe for ProjBytes
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.