A high-performance inference engine for AI models
"""Launch the ML Drift adapter, which builds the patched runner through run.sh."""
import pytest
from .engine_process import EngineProcess
from .engine_tests import EngineTests
MODEL = "Qwen/Qwen3-0.6B"
def add_options(parser: pytest.Parser) -> None:
group = parser.getgroup("ML Drift")
group.addoption("--mldrift-model", default=MODEL, help="Hugging Face checkpoint of an ML Drift model preset.")
class TestMLDrift(EngineTests):
engine_name = "mldrift"
@staticmethod
def command(config: pytest.Config) -> list[str]:
return ["./engine-mldrift/run.sh", "--model", config.getoption("mldrift_model")]
def test_sampling(self, engine: EngineProcess) -> None:
pytest.skip("ML Drift decodes greedily on the GPU and rejects sampling")