diff --git a/tests/test_config.py b/tests/test_config.py index ebf76c76a0..425d55f293 100644 --- a/tests/test_config.py +++ b/tests/test_config.py @@ -3486,6 +3486,12 @@ def test_scheduler_config_init(): print(SchedulerConfig.default_factory().max_model_len) +def test_scheduler_config_rejects_zero_max_num_scheduled_tokens(): + """A zero token budget never schedules anything, so generation would hang.""" + with pytest.raises(ValidationError, match="max_num_scheduled_tokens"): + SchedulerConfig.default_factory(max_num_scheduled_tokens=0) + + @pytest.mark.parametrize( ( "model_id", diff --git a/vllm/config/scheduler.py b/vllm/config/scheduler.py index 0d10103ed4..1dbe4b72eb 100644 --- a/vllm/config/scheduler.py +++ b/vllm/config/scheduler.py @@ -52,7 +52,7 @@ class SchedulerConfig: In real usage, this should be set in `EngineArgs.create_engine_config`. """ - max_num_scheduled_tokens: int | None = Field(default=None, ge=0) + max_num_scheduled_tokens: int | None = Field(default=None, ge=1) """Maximum number of tokens that the scheduler may issue in a single iteration. This is usually equal to max_num_batched_tokens, but can be smaller in cases