WeightConverter(
    "mlp.experts.down_proj",
    "block_sparse_moe.experts.*.w2.weight",
    operations=[SplitModulelist(dim=0)],
)