chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 11 Packages Projects Releases Wiki Activity

[Kernel Slimming] Migrate GPTQ-Marlin repack kernel to JIT (#18543)

Browse Source 
Co-authored-by: Xiaoyu Zhang <35585791+BBuf@users.noreply.github.com>
This commit is contained in:
Linyu Wu
2026-02-13 22:29:22 +08:00
committed by GitHub
parent 37273408eb
commit 0012d6a4eb
7 changed files with 615 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_wNa16.py
View File

@@ -43,7 +43,7 @@ from sglang.srt.utils import is_cuda
_is_cuda = is_cuda()
if _is_cuda:
from sgl_kernel import gptq_marlin_repack
from sglang.jit_kernel.gptq_marlin_repack import gptq_marlin_repack
ScalarType, scalar_types = get_scalar_types()

4
python/sglang/srt/layers/quantization/gptq.py
View File

@@ -61,7 +61,9 @@ if TYPE_CHECKING:
_is_cuda = is_cuda()
if _is_cuda:
from sgl_kernel import gptq_gemm, gptq_marlin_repack, gptq_shuffle
from sgl_kernel import gptq_gemm, gptq_shuffle
from sglang.jit_kernel.gptq_marlin_repack import gptq_marlin_repack
_is_npu = is_npu()

3
python/sglang/srt/layers/quantization/marlin_utils_fp8.py
View File

@@ -17,9 +17,8 @@ from sglang.srt.utils import is_cuda
_is_cuda = is_cuda()
if _is_cuda:
from sgl_kernel import gptq_marlin_repack
from sglang.jit_kernel.gptq_marlin import gptq_marlin_gemm
from sglang.jit_kernel.gptq_marlin_repack import gptq_marlin_repack
ScalarType, scalar_types = get_scalar_types()