Transforms complex documents like PDFs and Office docs into LLM-ready markdown/JSON for your Agentic workflows.
Python ai4sciencedocument-analysisdocxextract-datalayout-analysisocrparserpdfpdf-converterpdf-extractor-llmpdf-extractor-pretrainpdf-extractor-ragpdf-parserpptxpythonxlsx
from __future__ import annotations
import sys
from contextlib import contextmanager
from pathlib import Path
from types import SimpleNamespace
from typing import Iterator
import pytest
from mineru.kit.commands import vlm_server
from mineru.kit.vlm_server import mlx_vlm_server
from mineru_vl_utils import mlx_compat
@pytest.mark.parametrize("model_args", [["--model", "custom"], ["--model=custom"]])
def test_native_forwarding_and_cleanup(monkeypatch: pytest.MonkeyPatch, model_args: list[str]) -> None:
"""原生服务退出时恢复 argv 并结束兼容目录生命周期。"""
events = []
original = sys.argv
@contextmanager
def prepare(model: str) -> Iterator[Path]:
"""记录路径准备与清理顺序。"""
events.append(model)
try:
yield Path("/prepared")
finally:
events.append("cleanup")
def native() -> None:
"""模拟原生解析失败,确认异常不会泄漏资源。"""
events.append(sys.argv[1:])
raise SystemExit(2)
monkeypatch.setattr(mlx_compat, "prepare_mlx_model_path", prepare)
monkeypatch.setitem(sys.modules, "mlx_vlm.server", SimpleNamespace(main=native))
with pytest.raises(SystemExit, match="2"):
mlx_vlm_server.main(args=[*model_args, "--port=1234", "--api-key", "test"], prog_name="test", standalone_mode=False)
assert sys.argv is original
assert events == [
"custom",
["--model", "/prepared", "--host", "127.0.0.1", "--port", "1234", "--max-num-seqs", "8", "--api-key", "test"],
"cleanup",
]
def test_help_skips_model_preparation(monkeypatch: pytest.MonkeyPatch) -> None:
"""帮助直接委托原生入口,不解析或下载模型。"""
seen = []
monkeypatch.setattr(mlx_vlm_server, "_run_native", lambda *args: seen.append(args))
mlx_vlm_server.main(args=["--help"], prog_name="test", standalone_mode=False)
assert seen == [(["--help"], "test", False)]
@pytest.mark.parametrize("version, available", [("0.3.9", False), ("0.7.0", True), ("0.8.0", False)])
def test_version_gate(monkeypatch: pytest.MonkeyPatch, version: str, available: bool) -> None:
"""自动检测只接受已验证的版本范围。"""
monkeypatch.setattr(vlm_server.platform, "system", lambda: "Darwin")
monkeypatch.setattr(vlm_server.platform, "machine", lambda: "arm64")
monkeypatch.setattr(vlm_server, "is_mac_os_version_supported", lambda _: True)
monkeypatch.setattr(vlm_server.importlib.metadata, "version", lambda _: version)
monkeypatch.setattr(vlm_server.importlib.util, "find_spec", lambda _: object())
assert vlm_server._mlx_server_available() is available
@pytest.mark.parametrize("system,machine", [("Linux", "aarch64"), ("Darwin", "x86_64")])
def test_platform_gate(monkeypatch: pytest.MonkeyPatch, system: str, machine: str) -> None:
"""不支持的平台给出平台提示,不尝试导入 MLX。"""
monkeypatch.setattr(vlm_server.platform, "system", lambda: system)
monkeypatch.setattr(vlm_server.platform, "machine", lambda: machine)
assert "Apple Silicon" in vlm_server._mlx_server_error()
def test_default_model_and_options(monkeypatch: pytest.MonkeyPatch) -> None:
"""默认模型来自注册表,原生调用期间兼容目录保持有效。"""
seen = []
@contextmanager
def prepare(model: str) -> Iterator[Path]:
"""记录默认模型并提供临时模型路径。"""
seen.append(model)
yield Path("/prepared")
monkeypatch.setattr(mlx_vlm_server, "MINERU_2_5_PRO_2605_1_2B", SimpleNamespace(ensure=lambda: Path("/default")))
monkeypatch.setattr(mlx_compat, "prepare_mlx_model_path", prepare)
monkeypatch.setattr(mlx_vlm_server, "_run_native", lambda args, *rest: seen.append(args))
mlx_vlm_server.main(args=[], prog_name="test", standalone_mode=False)
assert seen == ["/default", ["--model", "/prepared", "--host", "127.0.0.1", "--port", "8080", "--max-num-seqs", "8"]]
@pytest.mark.parametrize(
"args,env,expected",
[([], None, "8"), ([], "4", "4"), (["--max-num-seqs", "12"], "4", "12"), (["--max-num-seqs=16"], None, "16")],
)
def test_server_sequence_limit_precedence(
monkeypatch: pytest.MonkeyPatch, args: list[str], env: str | None, expected: str
) -> None:
"""活跃并发默认八,显式 CLI 与环境配置按优先级覆盖默认值。"""
from contextlib import nullcontext
if env is None:
monkeypatch.delenv("MLX_VLM_MAX_NUM_SEQS", raising=False)
else:
monkeypatch.setenv("MLX_VLM_MAX_NUM_SEQS", env)
monkeypatch.setattr(mlx_compat, "prepare_mlx_model_path", lambda model: nullcontext(Path("/prepared")))
seen = []
monkeypatch.setattr(mlx_vlm_server, "_run_native", lambda argv, *rest: seen.extend(argv))
mlx_vlm_server.main(args=["--model", "test", *args], prog_name="test", standalone_mode=False)
assert seen.count("--max-num-seqs") == 1
assert seen[seen.index("--max-num-seqs") + 1] == expected