[Doc] Add engine args back in to the docs (#20674)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>

Author: hmellor

.gitignore

@@ -146,6 +146,7 @@ venv.bak/
 
 # mkdocs documentation
 /site
+docs/argparse
 docs/examples
 
 # mypy

docs/configuration/engine_args.md

@@ -1,15 +1,20 @@
+---
+toc_depth: 3
+---
+
 # Engine Arguments
 
 Engine arguments control the behavior of the vLLM engine.
 
 - For [offline inference](../serving/offline_inference.md), they are part of the arguments to [LLM][vllm.LLM] class.
 - For [online serving](../serving/openai_compatible_server.md), they are part of the arguments to `vllm serve`.
 
-You can look at [EngineArgs][vllm.engine.arg_utils.EngineArgs] and [AsyncEngineArgs][vllm.engine.arg_utils.AsyncEngineArgs] to see the available engine arguments.
+The engine argument classes, [EngineArgs][vllm.engine.arg_utils.EngineArgs] and [AsyncEngineArgs][vllm.engine.arg_utils.AsyncEngineArgs], are a combination of the configuration classes defined in [vllm.config][]. Therefore, if you are interested in developer documentation, we recommend looking at these configuration classes as they are the source of truth for types, defaults and docstrings.
+
+## `EngineArgs`
 
-However, these classes are a combination of the configuration classes defined in [vllm.config][]. Therefore, we would recommend you read about them there where they are best documented.
+--8<-- "docs/argparse/engine_args.md"
 
-For offline inference you will have access to these configuration classes and for online serving you can cross-reference the configs with `vllm serve --help`, which has its arguments grouped by config.
+## `AsyncEngineArgs`
 
-!!! note
-    Additional arguments are available to the [AsyncLLMEngine][vllm.engine.async_llm_engine.AsyncLLMEngine] which is used for online serving. These can be found by running `vllm serve --help`
+--8<-- "docs/argparse/async_engine_args.md"

docs/mkdocs/hooks/generate_argparse.py

@@ -0,0 +1,105 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+import logging
+import sys
+from argparse import SUPPRESS, HelpFormatter
+from pathlib import Path
+from typing import Literal
+from unittest.mock import MagicMock, patch
+
+ROOT_DIR = Path(__file__).parent.parent.parent.parent
+ARGPARSE_DOC_DIR = ROOT_DIR / "docs/argparse"
+
+sys.path.insert(0, str(ROOT_DIR))
+sys.modules["aiohttp"] = MagicMock()
+sys.modules["blake3"] = MagicMock()
+sys.modules["vllm._C"] = MagicMock()
+
+from vllm.engine.arg_utils import AsyncEngineArgs, EngineArgs  # noqa: E402
+from vllm.utils import FlexibleArgumentParser  # noqa: E402
+
+logger = logging.getLogger("mkdocs")
+
+
+class MarkdownFormatter(HelpFormatter):
+    """Custom formatter that generates markdown for argument groups."""
+
+    def __init__(self, prog):
+        super().__init__(prog,
+                         max_help_position=float('inf'),
+                         width=float('inf'))
+        self._markdown_output = []
+
+    def start_section(self, heading):
+        if heading not in {"positional arguments", "options"}:
+            self._markdown_output.append(f"\n### {heading}\n\n")
+
+    def end_section(self):
+        pass
+
+    def add_text(self, text):
+        if text:
+            self._markdown_output.append(f"{text.strip()}\n\n")
+
+    def add_usage(self, usage, actions, groups, prefix=None):
+        pass
+
+    def add_arguments(self, actions):
+        for action in actions:
+
+            option_strings = f'`{"`, `".join(action.option_strings)}`'
+            self._markdown_output.append(f"#### {option_strings}\n\n")
+
+            if choices := action.choices:
+                choices = f'`{"`, `".join(str(c) for c in choices)}`'
+                self._markdown_output.append(
+                    f"Possible choices: {choices}\n\n")
+
+            self._markdown_output.append(f"{action.help}\n\n")
+
+            if (default := action.default) != SUPPRESS:
+                self._markdown_output.append(f"Default: `{default}`\n\n")
+
+    def format_help(self):
+        """Return the formatted help as markdown."""
+        return "".join(self._markdown_output)
+
+
+def create_parser(cls, **kwargs) -> FlexibleArgumentParser:
+    """Create a parser for the given class with markdown formatting.
+    
+    Args:
+        cls: The class to create a parser for
+        **kwargs: Additional keyword arguments to pass to `cls.add_cli_args`.
+
+    Returns:
+        FlexibleArgumentParser: A parser with markdown formatting for the class.
+    """
+    parser = FlexibleArgumentParser()
+    parser.formatter_class = MarkdownFormatter
+    with patch("vllm.config.DeviceConfig.__post_init__"):
+        return cls.add_cli_args(parser, **kwargs)
+
+
+def on_startup(command: Literal["build", "gh-deploy", "serve"], dirty: bool):
+    logger.info("Generating argparse documentation")
+    logger.debug("Root directory: %s", ROOT_DIR.resolve())
+    logger.debug("Output directory: %s", ARGPARSE_DOC_DIR.resolve())
+
+    # Create the ARGPARSE_DOC_DIR if it doesn't exist
+    if not ARGPARSE_DOC_DIR.exists():
+        ARGPARSE_DOC_DIR.mkdir(parents=True)
+
+    # Create parsers to document
+    parsers = {
+        "engine_args": create_parser(EngineArgs),
+        "async_engine_args": create_parser(AsyncEngineArgs,
+                                           async_args_only=True),
+    }
+
+    # Generate documentation for each parser
+    for stem, parser in parsers.items():
+        doc_path = ARGPARSE_DOC_DIR / f"{stem}.md"
+        with open(doc_path, "w") as f:
+            f.write(parser.format_help())
+        logger.info("Argparse generated: %s", doc_path.relative_to(ROOT_DIR))

docs/mkdocs/hooks/generate_examples.py

@@ -161,8 +161,8 @@ def on_startup(command: Literal["build", "gh-deploy", "serve"], dirty: bool):
     for example in sorted(examples, key=lambda e: e.path.stem):
         example_name = f"{example.path.stem}.md"
         doc_path = EXAMPLE_DOC_DIR / example.category / example_name
-        logger.debug("Example generated: %s", doc_path.relative_to(ROOT_DIR))
         if not doc_path.parent.exists():
             doc_path.parent.mkdir(parents=True)
         with open(doc_path, "w+") as f:
             f.write(example.generate())
+        logger.debug("Example generated: %s", doc_path.relative_to(ROOT_DIR))

docs/mkdocs/overrides/partials/toc-item.html

@@ -0,0 +1,21 @@
+<!-- Enables the use of toc_depth in document frontmatter https://github.com/squidfunk/mkdocs-material/issues/4827#issuecomment-1869812019 -->
+<li class="md-nav__item">
+    <a href="{{ toc_item.url }}" class="md-nav__link">
+      <span class="md-ellipsis">
+        {{ toc_item.title }}
+      </span>
+    </a>
+  
+    <!-- Table of contents list -->
+    {% if toc_item.children %}
+      <nav class="md-nav" aria-label="{{ toc_item.title | striptags }}">
+        <ul class="md-nav__list">
+          {% for toc_item in toc_item.children %}
+          {% if not page.meta.toc_depth or toc_item.level <= page.meta.toc_depth %}
+            {% include "partials/toc-item.html" %}
+          {% endif %}
+          {% endfor %}
+        </ul>
+      </nav>
+    {% endif %}
+  </li>

mkdocs.yaml

@@ -3,6 +3,7 @@ site_url: https://docs.vllm.ai
 repo_url: https://github.com/vllm-project/vllm
 edit_uri: edit/main/docs/
 exclude_docs: |
+  argparse
   *.inc.md
   *.template.md
 theme:
@@ -47,6 +48,7 @@ theme:
 hooks:
   - docs/mkdocs/hooks/remove_announcement.py
   - docs/mkdocs/hooks/generate_examples.py
+  - docs/mkdocs/hooks/generate_argparse.py
   - docs/mkdocs/hooks/url_schemes.py
 
 # Required to stop api-autonav from raising an error

requirements/docs.txt

@@ -7,3 +7,18 @@ mkdocs-awesome-nav
 python-markdown-math
 regex
 ruff
+
+# Required for argparse hook only
+-f https://download.pytorch.org/whl/cpu
+cachetools
+cloudpickle
+fastapi
+msgspec
+openai
+pillow
+psutil
+pybase64
+pydantic
+torch
+transformers
+zmq

vllm/engine/arg_utils.py

@@ -12,8 +12,9 @@
 import warnings
 from dataclasses import MISSING, dataclass, fields, is_dataclass
 from itertools import permutations
-from typing import (Annotated, Any, Callable, Dict, List, Literal, Optional,
-                    Type, TypeVar, Union, cast, get_args, get_origin)
+from typing import (TYPE_CHECKING, Annotated, Any, Callable, Dict, List,
+                    Literal, Optional, Type, TypeVar, Union, cast, get_args,
+                    get_origin)
 
 import regex as re
 import torch
@@ -33,20 +34,26 @@
                          SchedulerConfig, SchedulerPolicy, SpeculativeConfig,
                          TaskOption, TokenizerMode, TokenizerPoolConfig,
                          VllmConfig, get_attr_docs, get_field)
-from vllm.executor.executor_base import ExecutorBase
 from vllm.logger import init_logger
-from vllm.model_executor.layers.quantization import QuantizationMethods
 from vllm.platforms import CpuArchEnum, current_platform
 from vllm.plugins import load_general_plugins
 from vllm.reasoning import ReasoningParserManager
 from vllm.test_utils import MODEL_WEIGHTS_S3_BUCKET, MODELS_ON_S3
 from vllm.transformers_utils.utils import check_gguf_file
-from vllm.usage.usage_lib import UsageContext
 from vllm.utils import (STR_DUAL_CHUNK_FLASH_ATTN_VAL, FlexibleArgumentParser,
                         GiB_bytes, get_ip, is_in_ray_actor)
 
 # yapf: enable
 
+if TYPE_CHECKING:
+    from vllm.executor.executor_base import ExecutorBase
+    from vllm.model_executor.layers.quantization import QuantizationMethods
+    from vllm.usage.usage_lib import UsageContext
+else:
+    ExecutorBase = Any
+    QuantizationMethods = Any
+    UsageContext = Any
+
 logger = init_logger(__name__)
 
 # object is used to allow for special typing forms
@@ -200,14 +207,17 @@ def _compute_kwargs(cls: ConfigType) -> dict[str, Any]:
         kwargs[name] = {"default": default, "help": help}
 
         # Set other kwargs based on the type hints
-        json_tip = """\n\nShould either be a valid JSON string or JSON keys
-        passed individually. For example, the following sets of arguments are
-        equivalent:\n\n
-        - `--json-arg '{"key1": "value1", "key2": {"key3": "value2"}}'`\n
-        - `--json-arg.key1 value1 --json-arg.key2.key3 value2`\n
-        Additionally, list elements can be passed individually using '+':
-        - `--json-arg '{"key4": ["value3", "value4", "value5"]}'`\n
-        - `--json-arg.key4+ value3 --json-arg.key4+='value4,value5'`\n\n"""
+        json_tip = """Should either be a valid JSON string or JSON keys
+passed individually. For example, the following sets of arguments are
+equivalent:
+
+- `--json-arg '{"key1": "value1", "key2": {"key3": "value2"}}'`\n
+- `--json-arg.key1 value1 --json-arg.key2.key3 value2`
+
+Additionally, list elements can be passed individually using `+`:
+
+- `--json-arg '{"key4": ["value3", "value4", "value5"]}'`\n
+- `--json-arg.key4+ value3 --json-arg.key4+='value4,value5'`"""
         if dataclass_cls is not None:
 
             def parse_dataclass(val: str, cls=dataclass_cls) -> Any:
@@ -219,7 +229,7 @@ def parse_dataclass(val: str, cls=dataclass_cls) -> Any:
                     raise argparse.ArgumentTypeError(repr(e)) from e
 
             kwargs[name]["type"] = parse_dataclass
-            kwargs[name]["help"] += json_tip
+            kwargs[name]["help"] += f"\n\n{json_tip}"
         elif contains_type(type_hints, bool):
             # Creates --no-<name> and --<name> flags
             kwargs[name]["action"] = argparse.BooleanOptionalAction
@@ -255,7 +265,7 @@ def parse_dataclass(val: str, cls=dataclass_cls) -> Any:
             kwargs[name]["type"] = union_dict_and_str
         elif contains_type(type_hints, dict):
             kwargs[name]["type"] = parse_type(json.loads)
-            kwargs[name]["help"] += json_tip
+            kwargs[name]["help"] += f"\n\n{json_tip}"
         elif (contains_type(type_hints, str)
               or any(is_not_builtin(th) for th in type_hints)):
             kwargs[name]["type"] = str
@@ -1545,7 +1555,6 @@ def _set_default_args_v0(self, model_config: ModelConfig) -> None:
             # Enable chunked prefill by default for long context (> 32K)
             # models to avoid OOM errors in initial memory profiling phase.
             elif use_long_context:
-                from vllm.platforms import current_platform
                 is_gpu = current_platform.is_cuda()
                 use_sliding_window = (model_config.get_sliding_window()
                                       is not None)
@@ -1653,6 +1662,7 @@ def _set_default_args_v1(self, usage_context: UsageContext,
         # NOTE(Kuntai): Setting large `max_num_batched_tokens` for A100 reduces
         # throughput, see PR #17885 for more details.
         # So here we do an extra device name check to prevent such regression.
+        from vllm.usage.usage_lib import UsageContext
         if device_memory >= 70 * GiB_bytes and "a100" not in device_name:
             # For GPUs like H100 and MI300x, use larger default values.
             default_max_num_batched_tokens = {

vllm/entrypoints/chat_utils.py

@@ -38,7 +38,6 @@
 
 from vllm.config import ModelConfig
 from vllm.logger import init_logger
-from vllm.model_executor.model_loader import get_model_cls
 from vllm.model_executor.models import SupportsMultiModal
 from vllm.multimodal import MULTIMODAL_REGISTRY, MultiModalDataDict
 from vllm.multimodal.utils import MediaConnector
@@ -524,6 +523,7 @@ def model_config(self) -> ModelConfig:
 
     @cached_property
     def model_cls(self):
+        from vllm.model_executor.model_loader import get_model_cls
         return get_model_cls(self.model_config)
 
     @property

vllm/inputs/registry.py

@@ -13,14 +13,21 @@
 from vllm.jsontree import JSONTree, json_map_leaves
 from vllm.logger import init_logger
 from vllm.transformers_utils.processor import cached_processor_from_config
-from vllm.transformers_utils.tokenizer import AnyTokenizer
 from vllm.utils import resolve_mm_processor_kwargs
 
 if TYPE_CHECKING:
     from vllm.config import ModelConfig
     from vllm.multimodal import (MultiModalDataDict, MultiModalPlaceholderDict,
                                  MultiModalRegistry)
     from vllm.sequence import SequenceData
+    from vllm.transformers_utils.tokenizer import AnyTokenizer
+else:
+    ModelConfig = Any
+    MultiModalDataDict = Any
+    MultiModalPlaceholderDict = Any
+    MultiModalRegistry = Any
+    SequenceData = Any
+    AnyTokenizer = Any
 
 _T = TypeVar("_T")
 _C = TypeVar("_C", bound=PretrainedConfig, default=PretrainedConfig)
@@ -36,7 +43,7 @@ class InputContext:
     modify the inputs.
     """
 
-    model_config: "ModelConfig"
+    model_config: ModelConfig
     """The configuration of the model."""
 
     def get_hf_config(
@@ -200,9 +207,9 @@ class DummyData(NamedTuple):
     Note: This is only used in V0.
     """
 
-    seq_data: "SequenceData"
-    multi_modal_data: Optional["MultiModalDataDict"] = None
-    multi_modal_placeholders: Optional["MultiModalPlaceholderDict"] = None
+    seq_data: SequenceData
+    multi_modal_data: Optional[MultiModalDataDict] = None
+    multi_modal_placeholders: Optional[MultiModalPlaceholderDict] = None
 
 
 class InputRegistry:
@@ -212,9 +219,9 @@ class InputRegistry:
 
     def dummy_data_for_profiling(
         self,
-        model_config: "ModelConfig",
+        model_config: ModelConfig,
         seq_len: int,
-        mm_registry: "MultiModalRegistry",
+        mm_registry: MultiModalRegistry,
         is_encoder_data: bool = False,
     ) -> DummyData:
         """