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

[CI] [FlashInfer v0.6.7] Use offline quantized checkpoint for MXFP8 Gemm tests (#21625)

Browse Source
This commit is contained in:
Ziang Li
2026-03-29 22:47:46 -07:00
committed by GitHub
parent 5b19c9a05d
commit 1a4b383fac
1 changed files with 3 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/registered/quant/test_fp8_blockwise_gemm.py
+3 -4
View File
@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
register_cuda_ci(est_time=420, suite="stage-c-test-4-gpu-b200") register_cuda_ci(est_time=420, suite="stage-c-test-4-gpu-b200")
MODEL_PATH = "Qwen/Qwen3-4B-Instruct-2507-FP8" MODEL_PATH = "Qwen/Qwen3-4B-Instruct-2507-FP8"
BF16_MODEL_PATH = "Qwen/Qwen3-4B-Instruct-2507" MXFP8_MODEL_PATH = "zianglih/Qwen3-4B-Instruct-2507-MXFP8"
class FP8BlockwiseGemmBase: class FP8BlockwiseGemmBase:
@@ -67,12 +67,10 @@ class MXFP8GemmBase:
def setUpClass(cls): def setUpClass(cls):
if cls.backend is None: if cls.backend is None:
raise NotImplementedError("Subclass must set 'backend' attribute") raise NotImplementedError("Subclass must set 'backend' attribute")
cls.model = try_cached_model(BF16_MODEL_PATH) cls.model = try_cached_model(MXFP8_MODEL_PATH)
cls.base_url = DEFAULT_URL_FOR_TEST cls.base_url = DEFAULT_URL_FOR_TEST
other_args = [ other_args = [
"--trust-remote-code", "--trust-remote-code",
"--quantization",
"mxfp8",
"--fp8-gemm-backend", "--fp8-gemm-backend",
cls.backend, cls.backend,
] ]
@@ -122,6 +120,7 @@ class TestFP8BlockwiseGemmFlashinferDeepGemm(FP8BlockwiseGemmBase, unittest.Test
backend = "flashinfer_deepgemm" backend = "flashinfer_deepgemm"
@unittest.skip("Currently PCG capture takes too long to complete, disable until fixed")
@unittest.skipIf(get_device_sm() < 100, "Test requires CUDA SM 100 or higher") @unittest.skipIf(get_device_sm() < 100, "Test requires CUDA SM 100 or higher")
class TestMXFP8GemmTriton(MXFP8GemmBase, unittest.TestCase): class TestMXFP8GemmTriton(MXFP8GemmBase, unittest.TestCase):
backend = "triton" backend = "triton"