diff --git a/csrc/moe/marlin_moe_wna16/marlin_template.h b/csrc/moe/marlin_moe_wna16/marlin_template.h index 1c255396099d5..8a913bb4a738c 100644 --- a/csrc/moe/marlin_moe_wna16/marlin_template.h +++ b/csrc/moe/marlin_moe_wna16/marlin_template.h @@ -1255,8 +1255,6 @@ __global__ void Marlin( if constexpr (has_zp && !is_zp_float) { if (is_new_zp) { if constexpr (group_blocks == -1) is_first_matmul_in_slice = false; - FragB frag_zp_0; - FragB frag_zp_1; int zp_quant_0, zp_quant_1; if constexpr (w_type.size_bits() == 4) { diff --git a/csrc/quantization/gptq_marlin/marlin_template.h b/csrc/quantization/gptq_marlin/marlin_template.h index e416d5a76a410..008663385707b 100644 --- a/csrc/quantization/gptq_marlin/marlin_template.h +++ b/csrc/quantization/gptq_marlin/marlin_template.h @@ -1113,8 +1113,6 @@ __global__ void Marlin( if constexpr (has_zp && !is_zp_float) { if (is_new_zp) { if constexpr (group_blocks == -1) is_first_matmul_in_slice = false; - FragB frag_zp_0; - FragB frag_zp_1; int zp_quant_0, zp_quant_1; if constexpr (w_type.size_bits() == 4) {