minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 13 Packages Projects Releases Wiki Activity

[Piecewise CUDA Graph] Support W4A8 (#13179)

Browse Source
This commit is contained in:
b8zhong
2025-11-16 11:53:50 +08:00
committed by GitHub
parent 12c789ebd1
commit f35f7f1245
1 changed files with 4 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/layers/quantization/fp8_kernel.py
+4
View File
@@ -45,6 +45,10 @@ _use_aiter = get_bool_env_var("SGLANG_USE_AITER") and _is_hip
if _is_cuda: if _is_cuda:
from sgl_kernel import sgl_per_tensor_quant_fp8, sgl_per_token_quant_fp8 from sgl_kernel import sgl_per_tensor_quant_fp8, sgl_per_token_quant_fp8
@torch.library.register_fake("sgl_kernel::sgl_per_tensor_quant_fp8")
def _sgl_per_tensor_quant_fp8(input, output_q, output_s, is_static):
return
# Temporary # Temporary
try: try:
from sgl_kernel import sgl_per_token_group_quant_8bit from sgl_kernel import sgl_per_token_group_quant_8bit