Skip to content

Transforms complex documents like PDFs and Office docs into LLM-ready markdown/JSON for your Agentic workflows.

feat: update llama server binary handling and improve preload model functions

myhlolif095d20df4e0
diff --git a/mineru/kit/vlm_server/llama_cpp_server.py b/mineru/kit/vlm_server/llama_cpp_server.py
index a6a5bba7..889d84c3 100644
--- a/mineru/kit/vlm_server/llama_cpp_server.py
+++ b/mineru/kit/vlm_server/llama_cpp_server.py
@@ -40,7 +40,9 @@ def llama_server_binary() -> Path:
"""定位 mineru-llama-cpp 包内分发的 llama-server 可执行文件。"""
import mineru_llama_cpp
- return Path(mineru_llama_cpp.__file__).resolve().parent / "bin" / "llama-server"
+ # Windows wheel 分发的是 llama-server.exe,其余平台无后缀。
+ binary_name = "llama-server.exe" if sys.platform == "win32" else "llama-server"
+ return Path(mineru_llama_cpp.__file__).resolve().parent / "bin" / binary_name
def _has_arg(args: list[str], *flags: str) -> bool:
diff --git a/tests/unittest/test_llama_cpp_server.py b/tests/unittest/test_llama_cpp_server.py
new file mode 100644
index 00000000..5632405e
--- /dev/null
+++ b/tests/unittest/test_llama_cpp_server.py
@@ -0,0 +1,27 @@
+from __future__ import annotations
+
+import sys
+import types
+from pathlib import Path
+
+import pytest
+
+from mineru.kit.vlm_server import llama_cpp_server
+
+
+ ("platform", "expected_name"),
+ [("win32", "llama-server.exe"), ("darwin", "llama-server"), ("linux", "llama-server")],
+)
+def test_llama_server_binary_matches_platform_executable_name(
+ monkeypatch: pytest.MonkeyPatch, tmp_path: Path, platform: str, expected_name: str
+) -> None:
+ """Windows wheel 分发 llama-server.exe,其余平台无后缀。"""
+ fake_pkg = types.ModuleType("mineru_llama_cpp")
+ fake_pkg.__file__ = str(tmp_path / "mineru_llama_cpp" / "__init__.py")
+ monkeypatch.setitem(sys.modules, "mineru_llama_cpp", fake_pkg)
+ monkeypatch.setattr(sys, "platform", platform)
+
+ binary = llama_cpp_server.llama_server_binary()
+
+ assert binary == tmp_path / "mineru_llama_cpp" / "bin" / expected_name
diff --git a/tests/unittest/test_parser_model_preload.py b/tests/unittest/test_parser_model_preload.py
index c3edf1b9..f9c7b187 100644
--- a/tests/unittest/test_parser_model_preload.py
+++ b/tests/unittest/test_parser_model_preload.py
@@ -30,7 +30,7 @@ def test_preload_local_models_initializes_conditional_model_families(monkeypatch
fake_runtime.HybridLocalModelContextSingleton = _ContextSingleton
monkeypatch.setitem(sys.modules, "mineru.model.runtime.hybrid", fake_runtime)
- api_server._preload_local_models("ch")
+ api_server._preload_local_models()
assert calls == [
("context", None),
@@ -44,17 +44,17 @@ def test_preload_local_models_initializes_conditional_model_families(monkeypatch
def test_preload_basic_models_initializes_only_local_models(monkeypatch: pytest.MonkeyPatch) -> None:
local_calls: list[str] = []
- monkeypatch.setattr(api_server, "_preload_local_models", local_calls.append)
+ monkeypatch.setattr(api_server, "_preload_local_models", lambda: local_calls.append("local"))
- result = api_server._preload_server_models("basic", language="ch")
+ result = api_server._preload_server_models("basic")
assert result == api_server._ModelPreloadResult(tier="basic", engine="hybrid-local")
- assert local_calls == ["ch"]
+ assert local_calls == ["local"]
def test_preload_standard_models_initializes_platform_engine_and_local_models(monkeypatch: pytest.MonkeyPatch) -> None:
calls: list[object] = []
- monkeypatch.setattr(api_server, "_preload_local_models", lambda language: calls.append(("local", language)))
+ monkeypatch.setattr(api_server, "_preload_local_models", lambda: calls.append("local"))
from mineru.model.vlm import selector as engine_utils
@@ -70,10 +70,10 @@ def test_preload_standard_models_initializes_platform_engine_and_local_models(mo
fake_runtime.ModelSingleton = _ModelSingleton
monkeypatch.setitem(sys.modules, "mineru.model.vlm.runtime", fake_runtime)
- result = api_server._preload_server_models("standard", language="en")
+ result = api_server._preload_server_models("standard")
assert result == api_server._ModelPreloadResult(tier="standard", engine="lmdeploy-engine")
- assert calls == [("vlm", "lmdeploy-engine", None, None), ("local", "en")]
+ assert calls == [("vlm", "lmdeploy-engine", None, None), "local"]
def test_local_preload_and_parse_reuse_llama_predictor(monkeypatch: pytest.MonkeyPatch, tmp_path: Path) -> None:
@@ -100,9 +100,9 @@ def test_local_preload_and_parse_reuse_llama_predictor(monkeypatch: pytest.Monke
monkeypatch.setattr(type(runtime.MINERU_2_5_PRO_2605_1_2B), "ensure", ensure_gguf)
engine_selector = MagicMock(return_value="llama-cpp-engine")
monkeypatch.setattr(selector, "get_vlm_engine", engine_selector)
- monkeypatch.setattr(api_server, "_preload_local_models", lambda language: None)
+ monkeypatch.setattr(api_server, "_preload_local_models", lambda: None)
- api_server._preload_server_models("standard", language="ch", vlm_config=VlmConfig())
+ api_server._preload_server_models("standard", vlm_config=VlmConfig())
predictor, engine = get_vlm_predictor(VlmConfig())
assert predictor is predictor_factory.return_value
assert engine == "llama-cpp-engine"
diff --git a/tests/unittest/test_vlm_remote.py b/tests/unittest/test_vlm_remote.py
index 3b1397f6..778c35f5 100644
--- a/tests/unittest/test_vlm_remote.py
+++ b/tests/unittest/test_vlm_remote.py
@@ -199,17 +199,17 @@ def test_remote_preflight_and_preload(openai_server: _OpenAIServer, monkeypatch:
monkeypatch.setattr(
api_server, "ensure_tier_runtime_dependencies", lambda tier, **kwargs: checks.append((tier, kwargs["vlm_config"]))
)
- monkeypatch.setattr(api_server, "_preload_local_models", local_loads.append)
+ monkeypatch.setattr(api_server, "_preload_local_models", lambda: local_loads.append("local"))
settings = _settings(openai_server)
api_server._preflight_tier_dependencies("standard", settings)
- result = api_server._preload_server_models("standard", language="en", vlm_config=settings)
+ result = api_server._preload_server_models("standard", vlm_config=settings)
assert result.engine == "http-client"
assert checks == [("standard", settings)]
- assert local_loads == ["en"]
+ assert local_loads == ["local"]
request_count = len(openai_server.requests)
get_vlm_predictor(settings)
assert len(openai_server.requests) == request_count
- api_server._preload_server_models("basic", language="ch", vlm_config=settings)
+ api_server._preload_server_models("basic", vlm_config=settings)
assert len(openai_server.requests) == request_count