diff --git a/fastdeploy/spec_decode/mtp_cuda.py b/fastdeploy/spec_decode/mtp_cuda.py index e84ab7bd538..ae4468b9e5c 100644 --- a/fastdeploy/spec_decode/mtp_cuda.py +++ b/fastdeploy/spec_decode/mtp_cuda.py @@ -303,6 +303,7 @@ def _propose(self, step_use_cudagraph: bool = False, is_dummy_run: bool = False, recover_batch_index_for_sampler_output( sampler_output, self.model_inputs.index_to_batch_id, + self.model_inputs.enable_pd_reorder, ) recover_model_output_map = recover_batch_index_for_output( self.model_inputs,