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

[CI] Fix stale GPU capability test patches (#37148)

Browse Source
This commit is contained in:
YAMY
2026-08-30 10:12:32 -07:00
committed by GitHub
parent 6a9366f036
commit 9a03bc2dc3
2 changed files with 25 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/registered/attention/unittests/dsv4/test_deepseek_v4.py
+4 -3
View File
@@ -444,8 +444,8 @@ class TestDSV4BreakableCudaGraphMetadataContract(CustomTestCase):
envs.SGLANG_ENABLE_PREFILL_WAR_READ_DONE.override(True), envs.SGLANG_ENABLE_PREFILL_WAR_READ_DONE.override(True),
envs.SGLANG_OPT_FLASHMLA_SPARSE_PREFILL.override(False), envs.SGLANG_OPT_FLASHMLA_SPARSE_PREFILL.override(False),
mock.patch( mock.patch(
"sglang.srt.layers.attention.deepseek_v4_backend._is_sm120", "sglang.srt.layers.attention.deepseek_v4_backend.get_platform",
False, return_value=SimpleNamespace(is_sm120=False),
), ),
): ):
backend.prepare_prefill_shared_read_snapshot( backend.prepare_prefill_shared_read_snapshot(
@@ -490,7 +490,8 @@ class TestDSV4BreakableCudaGraphMetadataContract(CustomTestCase):
envs.SGLANG_ENABLE_PREFILL_WAR_READ_DONE.override(True), envs.SGLANG_ENABLE_PREFILL_WAR_READ_DONE.override(True),
envs.SGLANG_OPT_FLASHMLA_SPARSE_PREFILL.override(True), envs.SGLANG_OPT_FLASHMLA_SPARSE_PREFILL.override(True),
mock.patch( mock.patch(
"sglang.srt.layers.attention.deepseek_v4_backend._is_sm120", False "sglang.srt.layers.attention.deepseek_v4_backend.get_platform",
return_value=SimpleNamespace(is_sm120=False),
), ),
self.assertRaisesRegex(RuntimeError, "snapshot failed"), self.assertRaisesRegex(RuntimeError, "snapshot failed"),
): ):
test/registered/quant/test_fp8_utils.py
+21 -12
View File
@@ -73,9 +73,9 @@ class TestApplyFp8LinearScaleDispatch(CustomTestCase):
) )
) )
for capability in ( for capability in (
"_is_sm90_supported", "is_sm90",
"_is_sm100_supported", "is_sm100",
"_is_sm120_supported", "is_sm120",
): ):
with self.subTest(capability=capability): with self.subTest(capability=capability):
input, qinput, weight, input_scale, weight_scale = self._make_inputs() input, qinput, weight, input_scale, weight_scale = self._make_inputs()
@@ -92,14 +92,20 @@ class TestApplyFp8LinearScaleDispatch(CustomTestCase):
) )
capabilities = { capabilities = {
"_is_sm90_supported": False, "is_sm90": False,
"_is_sm100_supported": False, "is_sm100": False,
"_is_sm120_supported": False, "is_sm120": False,
} }
capabilities[capability] = True capabilities[capability] = True
with patch.multiple(fp8_utils, **capabilities), patch.object( with patch.object(
fp8_utils,
"get_platform",
return_value=SimpleNamespace(**capabilities),
), patch.object(
fp8_utils, "fp8_scaled_mm", side_effect=fake_fp8_scaled_mm fp8_utils, "fp8_scaled_mm", side_effect=fake_fp8_scaled_mm
), patch.object(fp8_utils, "get_exec", return_value=exec_config): ), patch.object(
fp8_utils, "get_exec", return_value=exec_config
):
fp8_utils.apply_fp8_linear( fp8_utils.apply_fp8_linear(
input, input,
weight, weight,
@@ -151,11 +157,14 @@ class TestApplyFp8LinearScaleDispatch(CustomTestCase):
(mat_a.shape[0], mat_b.shape[1]), dtype=out_dtype, device=mat_a.device (mat_a.shape[0], mat_b.shape[1]), dtype=out_dtype, device=mat_a.device
) )
with patch.multiple( with patch.object(
fp8_utils, fp8_utils,
_is_sm90_supported=False, "get_platform",
_is_sm100_supported=False, return_value=SimpleNamespace(
_is_sm120_supported=False, is_sm90=False,
is_sm100=False,
is_sm120=False,
),
), patch.object(fp8_utils, "fp8_scaled_mm", side_effect=fake_fp8_scaled_mm): ), patch.object(fp8_utils, "fp8_scaled_mm", side_effect=fake_fp8_scaled_mm):
fp8_utils.apply_fp8_linear( fp8_utils.apply_fp8_linear(
input, input,