Function test_model_cache

tests/model_loader/test_model_cache.py:79–130

(
    fd_runner,
    model_name_or_path: str,
    tensor_parallel_size: int,
    max_num_seqs: int,
    max_model_len: int,
    max_tokens: int,
    quantization: str,
    env,
    monkeypatch,
)

    params,
)
def test_model_cache(
    fd_runner,
    model_name_or_path: str,
    tensor_parallel_size: int,
    max_num_seqs: int,
    max_model_len: int,
    max_tokens: int,
    quantization: str,
    env,
    monkeypatch,
) -> None:
    model_path = get_paddle_model_path(model_name_or_path)

    fd_outputs_v1 = run_with_timeout(
        target=form_model_get_output_topp0,
        args=(
            fd_runner,
            model_path,
            tensor_parallel_size,
            max_num_seqs,
            max_model_len,
            max_tokens,
            quantization,
            "default_v1",
            prompts,
        ),
    )

    if env:
        for k, v in env.items():
            monkeypatch.setenv(k, v)

    fd_outputs_v1_with_cache = run_with_timeout(
        target=form_model_get_output_topp0,
        args=(
            fd_runner,
            model_path,
            tensor_parallel_size,
            max_num_seqs,
            max_model_len,
            max_tokens,
            quantization,
            "default_v1",
            prompts,
        ),
    )
    check_tokens_id_and_text_close(
        outputs_0_lst=fd_outputs_v1,
        outputs_1_lst=fd_outputs_v1_with_cache,
        name_0="default_v1 loader",
        name_1="default_v1 loader using cache",
    )