chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[WIP]enable mxfp8 on nvidia sm120 (#19112)

Browse Source 
Co-authored-by: Your Name <you@example.com>
This commit is contained in:
Henry
2026-03-01 22:06:43 -05:00
committed by GitHub
parent e3e71f275a
commit e5edf222cd
4 changed files with 23 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
python/sglang/test/test_block_fp8.py
View File

@@ -19,7 +19,7 @@ from sglang.srt.layers.quantization.fp8_utils import (
mxfp8_group_quantize,
triton_mxfp8_blockscaled_linear,
)
from sglang.srt.utils import is_sm100_supported
from sglang.srt.utils import is_sm100_supported, is_sm120_supported
from sglang.test.test_utils import CustomTestCase
_is_cuda = torch.cuda.is_available() and torch.version.cuda
@@ -452,8 +452,8 @@ class TestMXFP8DenseLinear(CustomTestCase):
def setUpClass(cls):
if not torch.cuda.is_available():
raise unittest.SkipTest("CUDA is not available")
if not is_sm100_supported():
raise unittest.SkipTest("MXFP8 requires Blackwell (SM100+)")
if not (is_sm100_supported() or is_sm120_supported()):
raise unittest.SkipTest("MXFP8 requires Blackwell (SM100/SM120)")
torch.set_default_device("cuda")
def _mxfp8_dense_linear(self, M, NK, dtype, seed):