pub struct SsmWeightsQwen35 {
pub in_proj_qkv: DenseWeight,
pub in_proj_z: DenseWeight,
pub in_proj_a: DenseWeight,
pub in_proj_b: DenseWeight,
pub conv1d: DenseWeight,
pub a_log: DenseWeight,
pub dt_bias: DenseWeight,
pub norm: DenseWeight,
pub out_proj: DenseWeight,
}Expand description
Qwen3.5 SSM weights with separate projections.
Fields§
§in_proj_qkv: DenseWeightQKV projection: [qkv_size, hidden_size] BF16 (Q+K+V, no Z).
in_proj_z: DenseWeightZ gate projection: [z_size, hidden_size] BF16.
in_proj_a: DenseWeightAlpha projection: [num_value_heads, hidden_size] BF16.
in_proj_b: DenseWeightBeta projection: [num_value_heads, hidden_size] BF16.
conv1d: DenseWeightConv1d weight: [d_inner, 1, d_conv] BF16.
a_log: DenseWeightA_log parameter: [num_v_heads] FP32.
dt_bias: DenseWeightdt_bias parameter: [num_v_heads] FP32.
norm: DenseWeightGate norm weight: [value_dim] BF16.
out_proj: DenseWeightOutput projection: [value_dim, hidden_size] BF16 (NOT NVFP4 — quantizer skipped these).
Auto Trait Implementations§
impl Freeze for SsmWeightsQwen35
impl RefUnwindSafe for SsmWeightsQwen35
impl Send for SsmWeightsQwen35
impl Sync for SsmWeightsQwen35
impl Unpin for SsmWeightsQwen35
impl UnwindSafe for SsmWeightsQwen35
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