A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 13, 22:00 UTC
[Docs] Mock all imports for docs (#27873)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Harry Mellor committed
Nov 1, 2025 at 10:02 UTC
799ce45cc160ffc0a3e1a0f958cc8e260b751808
3 files changed
+49
-23
docs/mkdocs/hooks/generate_argparse.py
+47
-13
@@ -3,6 +3,7 @@
3
import importlib
4
import logging
5
import sys
6
+import traceback
7
from argparse import SUPPRESS, HelpFormatter
8
from pathlib import Path
9
from typing import Literal
@@ -16,7 +17,30 @@ ROOT_DIR = Path(__file__).parent.parent.parent.parent
17
ARGPARSE_DOC_DIR = ROOT_DIR / "docs/argparse"
18
19
sys.path.insert(0, str(ROOT_DIR))
20
+
21
+
22
+# Mock custom op code
23
+class MockCustomOp:
24
+ @staticmethod
25
+ def register(name):
26
+ def decorator(cls):
27
+ return cls
28
+
29
+ return decorator
30
+
31
+
32
+noop = lambda *a, **k: None
33
sys.modules["vllm._C"] = MagicMock()
34
+sys.modules["vllm.model_executor.custom_op"] = MagicMock(CustomOp=MockCustomOp)
35
+sys.modules["vllm.utils.torch_utils"] = MagicMock(direct_register_custom_op=noop)
36
+
37
+# Mock any version checks by reading from compiled CI requirements
38
+with open(ROOT_DIR / "requirements/test.txt") as f:
39
+ VERSIONS = dict(line.strip().split("==") for line in f if "==" in line)
40
+importlib.metadata.version = lambda name: VERSIONS.get(name) or "0.0.0"
41
+
42
+# Make torch.nn.Parameter safe to inherit from
43
+sys.modules["torch.nn"] = MagicMock(Parameter=object)
44
45
46
class PydanticMagicMock(MagicMock):
@@ -31,20 +55,17 @@ class PydanticMagicMock(MagicMock):
55
return core_schema.any_schema()
56
57
34
-def auto_mock(module, attr, max_mocks=50):
58
+def auto_mock(module, attr, max_mocks=100):
59
"""Function that automatically mocks missing modules during imports."""
60
logger.info("Importing %s from %s", attr, module)
61
for _ in range(max_mocks):
62
try:
63
# First treat attr as an attr, then as a submodule
40
- with patch("importlib.metadata.version", return_value="0.0.0"):
41
- return getattr(
42
- importlib.import_module(module),
43
- attr,
44
- importlib.import_module(f"{module}.{attr}"),
45
- )
46
- except importlib.metadata.PackageNotFoundError as e:
47
- raise e
64
+ return getattr(
65
+ importlib.import_module(module),
66
+ attr,
67
+ importlib.import_module(f"{module}.{attr}"),
68
+ )
69
except ModuleNotFoundError as e:
70
logger.info("Mocking %s for argparse doc generation", e.name)
71
sys.modules[e.name] = PydanticMagicMock(name=e.name)
@@ -139,10 +160,19 @@ def create_parser(add_cli_args, **kwargs) -> FlexibleArgumentParser:
160
Returns:
161
FlexibleArgumentParser: A parser with markdown formatting for the class.
162
"""
142
- parser = FlexibleArgumentParser(add_json_tip=False)
143
- parser.formatter_class = MarkdownFormatter
144
- with patch("vllm.config.DeviceConfig.__post_init__"):
145
- _parser = add_cli_args(parser, **kwargs)
163
+ try:
164
+ parser = FlexibleArgumentParser(add_json_tip=False)
165
+ parser.formatter_class = MarkdownFormatter
166
+ with patch("vllm.config.DeviceConfig.__post_init__"):
167
+ _parser = add_cli_args(parser, **kwargs)
168
+ except ModuleNotFoundError as e:
169
+ # Auto-mock runtime imports
170
+ if tb_list := traceback.extract_tb(e.__traceback__):
171
+ path = Path(tb_list[-1].filename).relative_to(ROOT_DIR)
172
+ auto_mock(module=".".join(path.parent.parts), attr=path.stem)
173
+ return create_parser(add_cli_args, **kwargs)
174
+ else:
175
+ raise e
176
# add_cli_args might be in-place so return parser if _parser is None
177
return _parser or parser
178
@@ -184,3 +214,7 @@ def on_startup(command: Literal["build", "gh-deploy", "serve"], dirty: bool):
214
with open(doc_path, "w", encoding="utf-8") as f:
215
f.write(super(type(parser), parser).format_help())
216
logger.info("Argparse generated: %s", doc_path.relative_to(ROOT_DIR))
217
+
218
+
219
+if __name__ == "__main__":
220
+ on_startup("build", False)
requirements/docs.txt
-8
@@ -9,12 +9,4 @@ mkdocs-git-revision-date-localized-plugin
9
mkdocs-minify-plugin
10
regex
11
ruff
12
-
13
-# Required for argparse hook only
14
--f https://download.pytorch.org/whl/cpu
15
-cachetools
16
-cloudpickle
17
-py-cpuinfo
18
-msgspec
12
pydantic
20
-torch
vllm/utils/cache.py
+2
-2
@@ -3,7 +3,7 @@
3
from collections import UserDict
4
from collections.abc import Callable, Hashable, Iterator, KeysView, Mapping
5
from types import MappingProxyType
6
-from typing import Generic, NamedTuple, TypeVar, cast, overload
6
+from typing import NamedTuple, TypeVar, cast, overload
7
8
import cachetools
9
@@ -48,7 +48,7 @@ class CacheInfo(NamedTuple):
48
)
49
50
51
-class LRUCache(cachetools.LRUCache[_K, _V], Generic[_K, _V]):
51
+class LRUCache(cachetools.LRUCache[_K, _V]):
52
def __init__(self, capacity: float, getsizeof: Callable[[_V], float] | None = None):
53
super().__init__(capacity, getsizeof)
54