[Docs] Reduce requirements for docs build (#23651)

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

docs/mkdocs/hooks/generate_argparse.py

@@ -1,5 +1,6 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+import importlib
 import logging
 import sys
 from argparse import SUPPRESS, HelpFormatter
@@ -7,25 +8,52 @@ from pathlib import Path
 from typing import Literal
 from unittest.mock import MagicMock, patch
 
+from pydantic_core import core_schema
+
+logger = logging.getLogger("mkdocs")
+
 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.benchmarks import latency  # noqa: E402
-from vllm.benchmarks import serve  # noqa: E402
-from vllm.benchmarks import throughput  # noqa: E402
-from vllm.engine.arg_utils import AsyncEngineArgs, EngineArgs  # noqa: E402
-from vllm.entrypoints.cli.openai import ChatCommand  # noqa: E402
-from vllm.entrypoints.cli.openai import CompleteCommand  # noqa: E402
-from vllm.entrypoints.openai import cli_args  # noqa: E402
-from vllm.entrypoints.openai import run_batch  # noqa: E402
-from vllm.utils import FlexibleArgumentParser  # noqa: E402
 
-logger = logging.getLogger("mkdocs")
+class PydanticMagicMock(MagicMock):
+    """`MagicMock` that's able to generate pydantic-core schemas."""
+
+    def __get_pydantic_core_schema__(self, source_type, handler):
+        return core_schema.any_schema()
+
+
+def auto_mock(module, attr, max_mocks=50):
+    """Function that automatically mocks missing modules during imports."""
+    logger.info("Importing %s from %s", attr, module)
+    for _ in range(max_mocks):
+        try:
+            # First treat attr as an attr, then as a submodule
+            return getattr(importlib.import_module(module), attr,
+                           importlib.import_module(f"{module}.{attr}"))
+        except importlib.metadata.PackageNotFoundError as e:
+            raise e
+        except ModuleNotFoundError as e:
+            logger.info("Mocking %s for argparse doc generation", e.name)
+            sys.modules[e.name] = PydanticMagicMock()
+
+    raise ImportError(
+        f"Failed to import {module}.{attr} after mocking {max_mocks} imports")
+
+
+latency = auto_mock("vllm.benchmarks", "latency")
+serve = auto_mock("vllm.benchmarks", "serve")
+throughput = auto_mock("vllm.benchmarks", "throughput")
+AsyncEngineArgs = auto_mock("vllm.engine.arg_utils", "AsyncEngineArgs")
+EngineArgs = auto_mock("vllm.engine.arg_utils", "EngineArgs")
+ChatCommand = auto_mock("vllm.entrypoints.cli.openai", "ChatCommand")
+CompleteCommand = auto_mock("vllm.entrypoints.cli.openai", "CompleteCommand")
+cli_args = auto_mock("vllm.entrypoints.openai", "cli_args")
+run_batch = auto_mock("vllm.entrypoints.openai", "run_batch")
+FlexibleArgumentParser = auto_mock("vllm.utils", "FlexibleArgumentParser")
 
 
 class MarkdownFormatter(HelpFormatter):

requirements/docs.txt

@@ -14,20 +14,6 @@ ruff
 # Required for argparse hook only
 -f https://download.pytorch.org/whl/cpu
 cachetools
-cbor2
-cloudpickle
-fastapi
 msgspec
-openai
-openai-harmony
-partial-json-parser
-pillow
-psutil
-pybase64
 pydantic
-setproctitle
 torch
-transformers
-zmq
-uvloop
-prometheus-client

vllm/sequence.py

@@ -16,14 +16,17 @@ import msgspec
 import torch
 
 from vllm.inputs import SingletonInputs
-from vllm.lora.request import LoRARequest
 from vllm.multimodal import MultiModalKwargs, MultiModalPlaceholderDict
 from vllm.pooling_params import PoolingParams
 from vllm.sampling_params import RequestOutputKind, SamplingParams
 
 if TYPE_CHECKING:
+    from vllm.lora.request import LoRARequest
     from vllm.v1.worker.kv_connector_model_runner_mixin import (
         KVConnectorOutput)
+else:
+    LoRARequest = Any
+    KVConnectorOutput = Any
 
 VLLM_TOKEN_ID_ARRAY_TYPE = "l"
 
@@ -1138,7 +1141,7 @@ class IntermediateTensors:
     """
 
     tensors: dict[str, torch.Tensor]
-    kv_connector_output: Optional["KVConnectorOutput"]
+    kv_connector_output: Optional[KVConnectorOutput]
 
     def __init__(self, tensors):
         # manually define this function, so that

vllm/transformers_utils/config.py

@@ -27,19 +27,6 @@ from transformers.utils import CONFIG_NAME as HF_CONFIG_NAME
 
 from vllm import envs
 from vllm.logger import init_logger
-# yapf conflicts with isort for this block
-# yapf: disable
-from vllm.transformers_utils.configs import (ChatGLMConfig, DeepseekVLV2Config,
-                                             EAGLEConfig, JAISConfig,
-                                             KimiVLConfig, MedusaConfig,
-                                             MLPSpeculatorConfig,
-                                             Nemotron_Nano_VL_Config,
-                                             NemotronConfig, OvisConfig,
-                                             RWConfig, SpeculatorsConfig,
-                                             Step3TextConfig, Step3VLConfig,
-                                             UltravoxConfig)
-# yapf: enable
-from vllm.transformers_utils.configs.mistral import adapt_config_dict
 from vllm.transformers_utils.utils import check_gguf_file
 
 if envs.VLLM_USE_MODELSCOPE:
@@ -67,24 +54,31 @@ def _get_hf_token() -> Optional[str]:
     return None
 
 
-_CONFIG_REGISTRY: dict[str, type[PretrainedConfig]] = {
+class LazyConfigDict(dict):
-    "chatglm": ChatGLMConfig,
+
-    "deepseek_vl_v2": DeepseekVLV2Config,
+    def __getitem__(self, key):
-    "kimi_vl": KimiVLConfig,
+        import vllm.transformers_utils.configs as configs
-    "Llama_Nemotron_Nano_VL": Nemotron_Nano_VL_Config,
+        return getattr(configs, super().__getitem__(key))
-    "RefinedWeb": RWConfig,  # For tiiuae/falcon-40b(-instruct)
+
-    "RefinedWebModel": RWConfig,  # For tiiuae/falcon-7b(-instruct)
+
-    "jais": JAISConfig,
+_CONFIG_REGISTRY: dict[str, type[PretrainedConfig]] = LazyConfigDict(
-    "mlp_speculator": MLPSpeculatorConfig,
+    chatglm="ChatGLMConfig",
-    "medusa": MedusaConfig,
+    deepseek_vl_v2="DeepseekVLV2Config",
-    "eagle": EAGLEConfig,
+    kimi_vl="KimiVLConfig",
-    "speculators": SpeculatorsConfig,
+    Llama_Nemotron_Nano_VL="Nemotron_Nano_VL_Config",
-    "nemotron": NemotronConfig,
+    RefinedWeb="RWConfig",  # For tiiuae/falcon-40b(-instruct)
-    "ovis": OvisConfig,
+    RefinedWebModel="RWConfig",  # For tiiuae/falcon-7b(-instruct)
-    "ultravox": UltravoxConfig,
+    jais="JAISConfig",
-    "step3_vl": Step3VLConfig,
+    mlp_speculator="MLPSpeculatorConfig",
-    "step3_text": Step3TextConfig,
+    medusa="MedusaConfig",
-}
+    eagle="EAGLEConfig",
+    speculators="SpeculatorsConfig",
+    nemotron="NemotronConfig",
+    ovis="OvisConfig",
+    ultravox="UltravoxConfig",
+    step3_vl="Step3VLConfig",
+    step3_text="Step3TextConfig",
+)
 
 _CONFIG_ATTRS_MAPPING: dict[str, str] = {
     "llm_config": "text_config",
@@ -461,6 +455,8 @@ def get_config(
                 model, revision, **kwargs)
             config_dict["max_position_embeddings"] = max_position_embeddings
 
+        from vllm.transformers_utils.configs.mistral import adapt_config_dict
+
         config = adapt_config_dict(config_dict)
 
         # Mistral configs may define sliding_window as list[int]. Convert it