[Scheduler] Add configurable decode interval after prefill (#35017)

This commit is contained in:
Po-Han Huang (NVIDIA)
2026-08-19 12:01:36 -07:00
committed by GitHub
parent 4f8ecf6ae9
commit 6f69f927da
5 changed files with 123 additions and 0 deletions
@@ -43,6 +43,15 @@ _mock_device.start()
class TestPrepareServerArgs(CustomTestCase):
def test_prefill_decode_interval(self):
args = ServerArgs(model_path="dummy", prefill_decode_interval=16)
self.assertEqual(args.prefill_decode_interval, 16)
with self.assertRaisesRegex(
ValueError, "--prefill-decode-interval must be non-negative"
):
ServerArgs(model_path="dummy", prefill_decode_interval=-1)
def test_return_hidden_states_mode_configuration(self):
disabled = ServerArgs(model_path="dummy")
self.assertFalse(disabled.enable_return_hidden_states)