# flake8: noqa
"""Tests experts_int8 quantization startup and generation, 
doesn't test correctness
"""
import pytest

from tests.quantization.utils import is_quant_method_supported

MODELS = ["ai21labs/Jamba-tiny-random"]


@pytest.mark.skipif(not is_quant_method_supported("experts_int8"),
                    reason="ExpertsInt8 is not supported on this GPU type.")
@pytest.mark.parametrize("model", MODELS)
@pytest.mark.parametrize("dtype", ["bfloat16"])
@pytest.mark.parametrize("max_tokens", [10])
def test_model_experts_int8_startup(
    hf_runner,
    aphrodite_runner,
    example_prompts,
    model: str,
    dtype: str,
    max_tokens: int,
) -> None:

    with aphrodite_runner(model, dtype=dtype,
                     quantization="experts_int8") as aphrodite_model:
        aphrodite_model.generate_greedy(example_prompts, max_tokens)