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

[Refactor] Set fp4-gemm-backend=auto on SM100 and rename fp4-gemm-backend with flashinfer_ prefix (#17309)

Browse Source
This commit is contained in:
b8zhong
2026-01-19 20:09:07 +08:00
committed by GitHub
parent 5c02217746
commit f374623fa9
7 changed files with 42 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/srt/test_nvfp4_gemm.py
+6 -6
View File
@@ -64,18 +64,18 @@ class TestFP4GemmAuto(FP4GemmBase, unittest.TestCase):
@unittest.skipIf(get_device_sm() < 100, "Test requires CUDA SM 100 or higher")
class TestFP4GemmCutlass(FP4GemmBase, unittest.TestCase):
backend = "cutlass"
class TestFP4GemmFlashinferCutlass(FP4GemmBase, unittest.TestCase):
backend = "flashinfer_cutlass"
@unittest.skipIf(get_device_sm() < 100, "Test requires CUDA SM 100 or higher")
class TestFP4GemmCudnn(FP4GemmBase, unittest.TestCase):
backend = "cudnn"
class TestFP4GemmFlashinferCudnn(FP4GemmBase, unittest.TestCase):
backend = "flashinfer_cudnn"
@unittest.skipIf(get_device_sm() < 100, "Test requires CUDA SM 100 or higher")
class TestFP4GemmTrtllm(FP4GemmBase, unittest.TestCase):
backend = "trtllm"
class TestFP4GemmFlashinferTrtllm(FP4GemmBase, unittest.TestCase):
backend = "flashinfer_trtllm"
if __name__ == "__main__":