File tree Expand file tree Collapse file tree 1 file changed +2
-2
lines changed Expand file tree Collapse file tree 1 file changed +2
-2
lines changed Original file line number Diff line number Diff line change @@ -243,6 +243,7 @@ set(VLLM_EXT_SRC
243
243
"csrc/sampler.cu"
244
244
"csrc/cuda_view.cu"
245
245
"csrc/quantization/gptq/q_gemm.cu"
246
+ "csrc/quantization/compressed_tensors/int8_quant_kernels.cu"
246
247
"csrc/quantization/fp8/common.cu"
247
248
"csrc/quantization/fused_kernels/fused_layernorm_dynamic_per_token_quant.cu"
248
249
"csrc/quantization/gguf/gguf_kernel.cu"
@@ -296,8 +297,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
296
297
"csrc/sparse/cutlass/sparse_scaled_mm_entry.cu"
297
298
"csrc/cutlass_extensions/common.cpp"
298
299
"csrc/attention/mla/cutlass_mla_entry.cu"
299
- "csrc/quantization/fp8/per_token_group_quant.cu"
300
- "csrc/quantization/compressed_tensors/int8_quant_kernels.cu" )
300
+ "csrc/quantization/fp8/per_token_group_quant.cu" )
301
301
302
302
set_gencode_flags_for_srcs(
303
303
SRCS "${VLLM_EXT_SRC} "
You can’t perform that action at this time.
0 commit comments