[Model Loading] Overlap checkpoint staging with CUDA graph capture during startup (#32017)

Co-authored-by: Wenhui Zhu <wzhu59@asu.edu>
Co-authored-by: Alex Nails <alex.nails@radixark.ai>
This commit is contained in:
Han Yu
2026-08-13 12:26:25 -07:00
committed by GitHub
co-authored by Wenhui Zhu Alex Nails
parent 0772e79ee7
commit 6b94d39f13
16 changed files with 2020 additions and 79 deletions
@@ -93,6 +93,25 @@ class TestServerArgsAnnotatedCli(CustomTestCase):
sa = self._parse(["--image-processor-backend", backend])
self.assertEqual(sa.image_processor_backend, backend)
def test_startup_weight_load_mode(self):
"""The startup loading mode keeps serial as the safe default."""
serial = self._parse([])
overlap = self._parse(["--startup-weight-load-mode", "overlap"])
self.assertEqual(serial.startup_weight_load_mode, "serial")
self.assertFalse(serial.is_startup_weight_load_overlap)
self.assertEqual(overlap.startup_weight_load_mode, "overlap")
self.assertTrue(overlap.is_startup_weight_load_overlap)
with self.assertRaises(SystemExit):
self.parser.parse_args(
[
"--model",
"dummy",
"--startup-weight-load-mode",
"unsupported",
]
)
def test_deprecated_flags_still_work(self):
"""Deprecated flags set the correct dest field."""
sa = self._parse(["--stream-output"])