diff --git a/custom_ops/gpu_ops/speculate_decoding/speculate_get_padding_offset.cu b/custom_ops/gpu_ops/speculate_decoding/speculate_get_padding_offset.cu index e37dacbf3..07f33ee2d 100644 --- a/custom_ops/gpu_ops/speculate_decoding/speculate_get_padding_offset.cu +++ b/custom_ops/gpu_ops/speculate_decoding/speculate_get_padding_offset.cu @@ -139,6 +139,7 @@ std::vector SpeculateGetPaddingOffsetInferDtype( PD_BUILD_STATIC_OP(speculate_get_padding_offset) .Inputs({"input_ids", "draft_tokens", + "cum_offsets" "token_num", "seq_len", "seq_lens_encoder"})