build: enable hf-xet model downloads

docs/zh/usage/model_source.md

@@ -6,6 +6,14 @@ MinerU使用 `HuggingFace` 和 `ModelScope` 作为模型仓库，用户可以根
 - `HuggingFace` 在全球范围内提供了优异的加载速度和极高稳定性。
 - `ModelScope` 是中国大陆地区用户的最佳选择，提供了无缝兼容的SDK模块，适用于无法访问`HuggingFace`的用户。
 
+使用 `HuggingFace` 下载模型时，MinerU 默认通过 `hf_xet` 加速传输。如果当前网络或代理无法访问 Xet CAS，设置
+`HF_HUB_DISABLE_XET=1` 可回退到普通 HTTP 下载。
+
+```powershell
+$env:HF_HUB_DISABLE_XET = "1"
+mineru-kit models download --tier standard --source huggingface
+```
+
 ## 模型源的切换方法
 
 ### 通过环境变量切换

pyproject.toml

@@ -45,7 +45,7 @@ dependencies = [
     "reportlab",
     "pdftext>=0.6.3,<0.8.0",
     "modelscope>=1.26.0",
-    "huggingface-hub>=0.32.4",
+    "huggingface-hub[hf_xet]>=0.32.4",
     "filelock>=3.0.0",
     "json-repair>=0.46.2",
     "opencv-python>=4.11.0.86",

tests/unittest/test_doclib_app_startup.py

@@ -8,6 +8,7 @@ from pathlib import Path
 import pytest
 from fastapi.testclient import TestClient
 from loguru import logger as loguru_logger
+from packaging.requirements import Requirement
 
 from mineru.config import LogConfig, PatchedConfig, config as mineru_config
 from mineru.doclib import app as doclib_app
@@ -55,6 +56,15 @@ def test_doclib_runtime_dependencies_are_in_base_install() -> None:
     assert "watchfiles" in dependency_names
 
 
+def test_huggingface_hub_base_dependency_enables_xet() -> None:
+    pyproject_path = Path(__file__).resolve().parents[2] / "pyproject.toml"
+    pyproject = tomllib.loads(pyproject_path.read_text(encoding="utf-8"))
+    dependencies = (Requirement(raw_dependency) for raw_dependency in pyproject["project"]["dependencies"])
+    huggingface_hub = next(dependency for dependency in dependencies if dependency.name == "huggingface-hub")
+
+    assert huggingface_hub.extras == {"hf_xet"}
+
+
 def test_basic_extra_includes_preflight_runtime_dependencies() -> None:
     pyproject_path = Path(__file__).resolve().parents[2] / "pyproject.toml"
     pyproject = tomllib.loads(pyproject_path.read_text(encoding="utf-8"))