pub struct ShortConvWeights {
pub in_proj: QTensor,
pub conv: Vec<f32>,
pub out_proj: QTensor,
}Expand description
Weights of one LFM2 short-convolution mixer
(model.layers.{i}.short_conv.*, renamed from the vendor conv.* at
convert time). No recurrent condensate — the only state is the causal
conv ring (the last kernel−1 gated inputs per channel).
Fields§
§in_proj: QTensor[3·hidden, hidden] — fused (B, C, x) projection.
conv: Vec<f32>[hidden · kernel] depthwise conv taps, flattened [channel][tap]
(the source [hidden, 1, kernel] with the singleton group axis
dropped). Tap kernel−1 multiplies the current position.
out_proj: QTensor[hidden, hidden] — output projection.
Auto Trait Implementations§
impl Freeze for ShortConvWeights
impl RefUnwindSafe for ShortConvWeights
impl Send for ShortConvWeights
impl Sync for ShortConvWeights
impl Unpin for ShortConvWeights
impl UnsafeUnpin for ShortConvWeights
impl UnwindSafe for ShortConvWeights
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