Data structure representing a single expert’s weight parameters and metadata.
Used during MoE fusion to track and organize expert-specific information across the pattern matching and transformation process.
Public Members
-
std::shared_ptr<Node> gate_proj_weight
Gate projection weight matrix.
-
std::shared_ptr<Node> up_proj_weight
Up projection weight matrix.
-
std::shared_ptr<Node> down_proj_weight
Down projection weight matrix.
-
size_t expert_id
Expert identifier (0-based index)
-
std::shared_ptr<Node> permute_node
Associated permute/transpose node for MoE layer grouping.