submission 616140

#!POPCORN leaderboard amd-moe-mxfp4
#!POPCORN gpu MI355X
"""
MXFP4 MoE — Zhao Dylan
v7: 极简 — 直接 fused_moe(topk=9, 含shared expert)
无分离, 无额外kernel, 最小Python开销
"""
import torch
from task import input_t, output_t
from aiter import ActivationType, QuantType
from aiter.fused_moe import fused_moe


def custom_kernel(data: input_t) -> output_t:
    (
        hidden_states, gate_up_weight, down_weight,
        gate_up_weight_scale, down_weight_scale,
        gate_up_weight_shuffled, down_weight_shuffled,
        gate_up_weight_scale_shuffled, down_weight_scale_shuffled,
        topk_weights, topk_ids, config,
    ) = data

    return fused_moe(
        hidden_states, gate_up_weight_shuffled, down_weight_shuffled,
        topk_weights, topk_ids,
        expert_mask=None, activation=ActivationType.Silu,
        quant_type=QuantType.per_1x32, doweight_stage1=False,
        w1_scale=gate_up_weight_scale_shuffled,
        w2_scale=down_weight_scale_shuffled,
        a1_scale=None, a2_scale=None,
        hidden_pad=config["d_hidden_pad"] - config["d_hidden"],
        intermediate_pad=config["d_expert_pad"] - config["d_expert"],
    )