ExpertVariant enum
Feed-forward body inside an Expert.
- ExpertVariant.mlp — the classical two-layer MLP
x -> w2(act(w1(x))). - ExpertVariant.swiGlu — the SwiGLU gated body used by both
DeepSeek-V3 (
inference/model.py) and Mixtral (transformers/models/mixtral/modeling_mixtral.py):x -> w2(silu(w1(x)) * w3(x)). Adds one extraw3projection of shape[hiddenDim, dim]. Uses SiLU regardless of ExpertActivation (the gate itself is the nonlinearity).
Values
- mlp → const ExpertVariant
- swiGlu → const ExpertVariant
Properties
- hashCode → int
-
The hash code for this object.
no setterinherited
- index → int
-
A numeric identifier for the enumerated value.
no setterinherited
- name → String
-
Available on Enum, provided by the EnumName extension
The name of the enum value.no setter - runtimeType → Type
-
A representation of the runtime type of the object.
no setterinherited
Methods
-
noSuchMethod(
Invocation invocation) → dynamic -
Invoked when a nonexistent method or property is accessed.
inherited
-
toString(
) → String -
A string representation of this object.
inherited
Operators
-
operator ==(
Object other) → bool -
The equality operator.
inherited
Constants
-
values
→ const List<
ExpertVariant> - A constant List of the values in this enum, in order of their declaration.