diff --git a/src/llamafactory/model/model_utils/moe.py b/src/llamafactory/model/model_utils/moe.py index 58039e2a..cd5558d3 100644 --- a/src/llamafactory/model/model_utils/moe.py +++ b/src/llamafactory/model/model_utils/moe.py @@ -61,7 +61,7 @@ def add_z3_leaf_module(model: "PreTrainedModel") -> None: _set_z3_leaf_modules(model, [MixtralSparseMoeBlock]) - if model_type == "qwen2moe": + if model_type == "qwen2_moe": from transformers.models.qwen2_moe.modeling_qwen2_moe import Qwen2MoeSparseMoeBlock _set_z3_leaf_modules(model, [Qwen2MoeSparseMoeBlock])