pub struct DenseFfnWeights {
pub gate_proj: QuantizedWeight,
pub up_proj: QuantizedWeight,
pub down_proj: QuantizedWeight,
pub gate_proj_t: Option<QuantizedWeight>,
pub up_proj_t: Option<QuantizedWeight>,
pub down_proj_t: Option<QuantizedWeight>,
}Fields§
§gate_proj: QuantizedWeight§up_proj: QuantizedWeight§down_proj: QuantizedWeight§gate_proj_t: Option<QuantizedWeight>Transposed ([K/2, N]) copies for the fast w4a16_gemm_t_m128 prefill
kernel. None → prefill falls back to the slow M64xN64 base kernel.
The non-transposed copies above are kept for the decode gemv path.
up_proj_t: Option<QuantizedWeight>§down_proj_t: Option<QuantizedWeight>Auto Trait Implementations§
impl Freeze for DenseFfnWeights
impl RefUnwindSafe for DenseFfnWeights
impl Send for DenseFfnWeights
impl Sync for DenseFfnWeights
impl Unpin for DenseFfnWeights
impl UnwindSafe for DenseFfnWeights
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