Skip to content

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

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Fridge003 merged 3 commits into from
Mar 30, 2026
Merged

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

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
f7b2d07
Fix test
zianglih Mar 29, 2026
14b6d46
Drop disable Triton test
zianglih Mar 29, 2026
872114d
Disable Triton test again due to long PCG capture
zianglih Mar 29, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 3 additions & 4 deletions test/registered/quant/test_fp8_blockwise_gemm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,7 +15,7 @@
register_cuda_ci(est_time=420, suite="stage-c-test-4-gpu-b200")

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:
Expand Down Expand Up @@ -67,12 +67,10 @@ class MXFP8GemmBase:
def setUpClass(cls):
if cls.backend is None:
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
other_args = [
"--trust-remote-code",
"--quantization",
"mxfp8",
"--fp8-gemm-backend",
cls.backend,
]
Expand Down Expand Up @@ -122,6 +120,7 @@ class TestFP8BlockwiseGemmFlashinferDeepGemm(FP8BlockwiseGemmBase, unittest.Test
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")
class TestMXFP8GemmTriton(MXFP8GemmBase, unittest.TestCase):
backend = "triton"
Expand Down
Loading