A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 03:30 UTC
remove unused variables in marlin_template.h (#20236)
周周周 committed
Jul 2, 2025 at 08:51 UTC
9290de5667c92450768c79a50b748d358bc825c8
2 files changed
-4
csrc/moe/marlin_moe_wna16/marlin_template.h
-2
@@ -1255,8 +1255,6 @@ __global__ void Marlin(
1255
if constexpr (has_zp && !is_zp_float) {
1256
if (is_new_zp) {
1257
if constexpr (group_blocks == -1) is_first_matmul_in_slice = false;
1258
- FragB frag_zp_0;
1259
- FragB frag_zp_1;
1258
int zp_quant_0, zp_quant_1;
1259
1260
if constexpr (w_type.size_bits() == 4) {
csrc/quantization/gptq_marlin/marlin_template.h
-2
@@ -1113,8 +1113,6 @@ __global__ void Marlin(
1113
if constexpr (has_zp && !is_zp_float) {
1114
if (is_new_zp) {
1115
if constexpr (group_blocks == -1) is_first_matmul_in_slice = false;
1116
- FragB frag_zp_0;
1117
- FragB frag_zp_1;
1116
int zp_quant_0, zp_quant_1;
1117
1118
if constexpr (w_type.size_bits() == 4) {