pub struct Ablation {
pub attnres_mix: f32,
pub force_expert: Option<usize>,
pub skip_layer: Option<usize>,
pub o_proj_tp: usize,
pub drop_o_proj_rank: Option<usize>,
}Expand description
Graph mutants. Mix=0 / force-expert-0 are C1; skip-layer is C2;
o_proj TP is C7.
Fields§
§attnres_mix: f32§force_expert: Option<usize>§skip_layer: Option<usize>Skip this layer index in forward_token (C2 known-bad).
o_proj_tp: usizeColumn-parallel o_proj world size. 1 = unsplit, 2 = in-process TP.
drop_o_proj_rank: Option<usize>Zero this TP rank’s o_proj shard (C7 known-bad). Requires o_proj_tp=2.
Implementations§
Trait Implementations§
impl Copy for Ablation
Auto Trait Implementations§
impl Freeze for Ablation
impl RefUnwindSafe for Ablation
impl Send for Ablation
impl Sync for Ablation
impl Unpin for Ablation
impl UnwindSafe for Ablation
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