submission 568211

# submission.py - 使用 AITER fused_moe (根据实际签名修正)

from task import input_t, output_t
import torch
import aiter
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

    hidden_pad = config["d_hidden_pad"] - config["d_hidden"]
    intermediate_pad = config["d_expert_pad"] - config["d_expert"]

    # 根据错误信息中的函数签名调用
    output = fused_moe(
        hidden_states=hidden_states,
        w1=gate_up_weight_shuffled,
        w2=down_weight_shuffled,
        topk_weight=topk_weights,       # 🔧 修正：不是 topk_weights
        topk_ids=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,
        block_size_M=-1,                # 新增参数
        num_local_tokens=None,          # 必须是 Tensor 或 None
        moe_sorting_dispatch_policy=False,
        dtype=None,
        hidden_pad=hidden_pad,
        intermediate_pad=intermediate_pad,
        bias1=None,
        bias2=None,
    )

    return output


ref_kernel = custom_kernel