Sum the selected expert outputs with their routing weights

Read the equation

Each chosen expert evaluates the same input representation h. Its output vector is multiplied by the associated routing weight, and the selected vectors are added. The implementation groups tokens by expert for efficient evaluation and restores their original order before summing; those data-movement steps preserve this intended mixture formula.

Terms & definitions

How it works

Read the definition from its right-hand expression to the quantity on the left. Its surrounding source records the intended input, state, or parameter context. Where the source leaves an operation unspecified, this page preserves that gap.

Assumptions & limits

Source recovery establishes that the expression appears in this record. Its mathematical validity, implementation behavior, and interpretation must be assessed separately. See the linked source and research notes for the checks available for this version.

Original context

Then in DeepseekV3MoE.forward (lines 522-532):

      identity = h                            # the PRE-MoE residual
      y = moe_infer(h, topk_idx, topk_weight) # sum_i w_i * Expert_i(h)
      y = y + shared_experts(identity)        # shared expert added afterwards

  The shared expert consumes the pre-MoE hidden state, not the routed output.

y = moe_infer(h, topk_idx, topk_weight) # sum_i w_i * Expert_i(h)

Related builds