环境
vLLM 使用以下环境变量来配置系统：
警告 请注意，
VLLM_PORT和VLLM_HOST_IP仅用于 vLLM 内部使用，它们并不是 API 服务器的端口和 IP。如果你使用--host $VLLM_HOST_IP和--port $VLLM_PORT来启动 API 服务器，它将无法正常工作。
所有 vLLM 使用的环境变量都以 VLLM_ 作为前缀。Kubernetes 用户需要特别注意：请不要将服务命名为 vllm，否则 Kubernetes 设置的环境变量可能会与 vLLM 的环境变量发生冲突。因为 Kubernetes 会为每个服务设置环境变量，并使用大写的服务名称作为前缀。
environment_variables: dict[str, Callable[[], Any]] = {
# ================== Installation Time Env Vars ==================
# ================== 安装时环境变量 ==================
# Target device of vLLM, supporting [cuda (by default),
# rocm, neuron, cpu, openvino]
# vLLM 的目标设备，支持 [cuda（默认）、rocm、neuron、cpu、openvino]
"VLLM_TARGET_DEVICE":
lambda: os.getenv("VLLM_TARGET_DEVICE", "cuda"),
# Maximum number of compilation jobs to run in parallel.
# By default this is the number of CPUs
# 并行运行的编译任务的最大数量。默认是 CPU 的数量
"MAX_JOBS":
lambda: os.getenv("MAX_JOBS", None),
# Number of threads to use for nvcc
# By default this is 1.
# If set, `MAX_JOBS` will be reduced to avoid oversubscribing the CPU.
# 用于 nvcc 的线程数。默认为 1。如果设置，`MAX_JOBS` 将减少以避免 CPU 过载
"NVCC_THREADS":
lambda: os.getenv("NVCC_THREADS", None),
# If set, vllm will use precompiled binaries (*.so)
# 如果设置，vllm 将使用预编译的二进制文件 (*.so)
"VLLM_USE_PRECOMPILED":
lambda: bool(os.environ.get("VLLM_USE_PRECOMPILED")) or bool(
os.environ.get("VLLM_PRECOMPILED_WHEEL_LOCATION")),
# Whether to force using nightly wheel in python build.
# This is used for testing the nightly wheel in python build.
# 是否在 Python 构建中强制使用 nightly wheel。用于在 Python 构建中测试 nightly wheel
"VLLM_TEST_USE_PRECOMPILED_NIGHTLY_WHEEL":
lambda: bool(int(os.getenv("VLLM_TEST_USE_PRECOMPILED_NIGHTLY_WHEEL", "0"))
),
# CMake build type
# If not set, defaults to "Debug" or "RelWithDebInfo"
# Available options: "Debug", "Release", "RelWithDebInfo"
# CMake 构建类型
# 如果未设置，默认为 "Debug" 或 "RelWithDebInfo"
# 可用选项："Debug"、"Release"、"RelWithDebInfo"
"CMAKE_BUILD_TYPE":
lambda: os.getenv("CMAKE_BUILD_TYPE"),
# If set, vllm will print verbose logs during installation
# 如果设置，vllm 将在安装期间打印详细日志
"VERBOSE":
lambda: bool(int(os.getenv('VERBOSE', '0'))),
# Root directory for vLLM configuration files
# Defaults to `~/.config/vllm` unless `XDG_CONFIG_HOME` is set
# Note that this not only affects how vllm finds its configuration files
# during runtime, but also affects how vllm installs its configuration
# files during **installation**.
# vLLM 配置文件的根目录
# 默认为 `~/.config/vllm`，除非设置了 `XDG_CONFIG_HOME`
# 注意，这不仅影响 vllm 在运行时如何查找其配置文件，还影响 vllm 在**安装**期间如何安装其配置文件
"VLLM_CONFIG_ROOT":
lambda: os.path.expanduser(
os.getenv(
"VLLM_CONFIG_ROOT",
os.path.join(get_default_config_root(), "vllm"),
)),
# ================== Runtime Env Vars ==================
# ================== 运行时环境变量 ==================
# Root directory for vLLM cache files
# Defaults to `~/.cache/vllm` unless `XDG_CACHE_HOME` is set
# vLLM 缓存文件的根目录
# 默认为 `~/.cache/vllm`，除非设置了 `XDG_CACHE_HOME`
"VLLM_CACHE_ROOT":
lambda: os.path.expanduser(
os.getenv(
"VLLM_CACHE_ROOT",
os.path.join(get_default_cache_root(), "vllm"),
)),
# used in distributed environment to determine the ip address
# of the current node, when the node has multiple network interfaces.
# If you are using multi-node inference, you should set this differently
# on each node.
# 在分布式环境中用于确定当前节点的 IP 地址，当节点有多个网络接口时
# 如果你使用多节点推理，你应该在每个节点上设置不同的值
'VLLM_HOST_IP':
lambda: os.getenv('VLLM_HOST_IP', ""),
# used in distributed environment to manually set the communication port
# Note: if VLLM_PORT is set, and some code asks for multiple ports, the
# VLLM_PORT will be used as the first port, and the rest will be generated
# by incrementing the VLLM_PORT value.
# '0' is used to make mypy happy
# 在分布式环境中用于手动设置通信端口
# 注意：如果设置了 VLLM_PORT，并且某些代码请求多个端口，VLLM_PORT 将用作第一个端口，其余的将通过递增 VLLM_PORT 值生成
# '0' 用于使 mypy 满意
'VLLM_PORT':
lambda: int(os.getenv('VLLM_PORT', '0'))
if 'VLLM_PORT' in os.environ else None,
# path used for ipc when the frontend api server is running in
# multi-processing mode to communicate with the backend engine process.
# 当前端 API 服务器以多进程模式运行时用于 IPC 的路径，以便与后端引擎进程通信
'VLLM_RPC_BASE_PATH':
lambda: os.getenv('VLLM_RPC_BASE_PATH', tempfile.gettempdir()),
# If true, will load models from ModelScope instead of Hugging Face Hub.
# note that the value is true or false, not numbers
# 如果为 true，将从 ModelScope 加载模型，而不是 Hugging Face Hub
# 注意，值为 true 或 false，而不是数字
"VLLM_USE_MODELSCOPE":
lambda: os.environ.get("VLLM_USE_MODELSCOPE", "False").lower() == "true",
# Interval in seconds to log a warning message when the ring buffer is full
# 当环形缓冲区满时记录警告消息的时间间隔（秒）
"VLLM_RINGBUFFER_WARNING_INTERVAL":
lambda: int(os.environ.get("VLLM_RINGBUFFER_WARNING_INTERVAL", "60")),
# path to cudatoolkit home directory, under which should be bin, include,
# and lib directories.
# cudatoolkit 主目录的路径，其下应有 bin、include 和 lib 目录
"CUDA_HOME":
lambda: os.environ.get("CUDA_HOME", None),
# Path to the NCCL library file. It is needed because nccl>=2.19 brought
# by PyTorch contains a bug: https://github.com/NVIDIA/nccl/issues/1234
# NCCL 库文件的路径。这是必需的，因为 PyTorch 自带的 nccl>=2.19 包含一个 bug：https://github.com/NVIDIA/nccl/issues/1234
"VLLM_NCCL_SO_PATH":
lambda: os.environ.get("VLLM_NCCL_SO_PATH", None),
# when `VLLM_NCCL_SO_PATH` is not set, vllm will try to find the nccl
# library file in the locations specified by `LD_LIBRARY_PATH`
# 当 `VLLM_NCCL_SO_PATH` 未设置时，vllm 将尝试在 `LD_LIBRARY_PATH` 指定的位置查找 nccl 库文件
"LD_LIBRARY_PATH":
lambda: os.environ.get("LD_LIBRARY_PATH", None),
# flag to control if vllm should use triton flash attention
# 控制 vllm 是否使用 triton flash attention 的标志
"VLLM_USE_TRITON_FLASH_ATTN":
lambda: (os.environ.get("VLLM_USE_TRITON_FLASH_ATTN", "True").lower() in
("true", "1")),
# Force vllm to use a specific flash-attention version (2 or 3), only valid
# when using the flash-attention backend.
# 强制 vllm 使用特定的 flash-attention 版本（2 或 3），仅在使用 flash-attention 后端时有效
"VLLM_FLASH_ATTN_VERSION":
lambda: maybe_convert_int(os.environ.get("VLLM_FLASH_ATTN_VERSION", None)),
# Internal flag to enable Dynamo fullgraph capture
# 内部标志，用于启用 Dynamo fullgraph 捕获
"VLLM_TEST_DYNAMO_FULLGRAPH_CAPTURE":
lambda: bool(
os.environ.get("VLLM_TEST_DYNAMO_FULLGRAPH_CAPTURE", "1") != "0"),
# local rank of the process in the distributed setting, used to determine
# the GPU device id
# 分布式设置中进程的本地 rank，用于确定 GPU 设备 ID
"LOCAL_RANK":
lambda: int(os.environ.get("LOCAL_RANK", "0")),
# used to control the visible devices in the distributed setting
# 用于控制分布式设置中的可见设备
"CUDA_VISIBLE_DEVICES":
lambda: os.environ.get("CUDA_VISIBLE_DEVICES", None),
# timeout for each iteration in the engine
# 引擎中每次迭代的超时时间
"VLLM_ENGINE_ITERATION_TIMEOUT_S":
lambda: int(os.environ.get("VLLM_ENGINE_ITERATION_TIMEOUT_S", "60")),
# API key for vLLM API server
# vLLM API 服务器的 API 密钥
"VLLM_API_KEY":
lambda: os.environ.get("VLLM_API_KEY", None),
# S3 access information, used for tensorizer to load model from S3
# S3 访问信息，用于 tensorizer 从 S3 加载模型
"S3_ACCESS_KEY_ID":
lambda: os.environ.get("S3_ACCESS_KEY_ID", None),
"S3_SECRET_ACCESS_KEY":
lambda: os.environ.get("S3_SECRET_ACCESS_KEY", None),
"S3_ENDPOINT_URL":
lambda: os.environ.get("S3_ENDPOINT_URL", None),
# Usage stats collection
# 使用统计收集
"VLLM_USAGE_STATS_SERVER":
lambda: os.environ.get("VLLM_USAGE_STATS_SERVER", "https://stats.vllm.ai"),
"VLLM_NO_USAGE_STATS":
lambda: os.environ.get("VLLM_NO_USAGE_STATS", "0") == "1",
"VLLM_DO_NOT_TRACK":
lambda: (os.environ.get("VLLM_DO_NOT_TRACK", None) or os.environ.get(
"DO_NOT_TRACK", None) or "0") == "1",
"VLLM_USAGE_SOURCE":
lambda: os.environ.get("VLLM_USAGE_SOURCE", "production"),
# Logging configuration
# If set to 0, vllm will not configure logging
# If set to 1, vllm will configure logging using the default configuration
# or the configuration file specified by VLLM_LOGGING_CONFIG_PATH
# 日志配置
# 如果设置为 0，vllm 将不配置日志
# 如果设置为 1，vllm 将使用默认配置或 VLLM_LOGGING_CONFIG_PATH 指定的配置文件配置日志
"VLLM_CONFIGURE_LOGGING":
lambda: int(os.getenv("VLLM_CONFIGURE_LOGGING", "1")),
"VLLM_LOGGING_CONFIG_PATH":
lambda: os.getenv("VLLM_LOGGING_CONFIG_PATH"),
# this is used for configuring the default logging level
# 用于配置默认日志级别
"VLLM_LOGGING_LEVEL":
lambda: os.getenv("VLLM_LOGGING_LEVEL", "INFO"),
# if set, VLLM_LOGGING_PREFIX will be prepended to all log messages
# 如果设置，VLLM_LOGGING_PREFIX 将被添加到所有日志消息的前面
"VLLM_LOGGING_PREFIX":
lambda: os.getenv("VLLM_LOGGING_PREFIX", ""),
# if set, vllm will call logits processors in a thread pool with this many
# threads. This is useful when using custom logits processors that either
# (a) launch additional CUDA kernels or (b) do significant CPU-bound work
# while not holding the python GIL, or both.
# 如果设置，vllm 将在一个线程池中调用 logits processors，线程池的大小为此值
# 当使用自定义 logits processors 时，这很有用，这些 processors 可能（a）启动额外的 CUDA 内核，或（b）在不持有 Python GIL 的情况下执行大量 CPU 密集型工作，或两者兼有
"VLLM_LOGITS_PROCESSOR_THREADS":
lambda: int(os.getenv("VLLM_LOGITS_PROCESSOR_THREADS", "0"))
if "VLLM_LOGITS_PROCESSOR_THREADS" in os.environ else None,
# Trace function calls
# If set to 1, vllm will trace function calls
# Useful for debugging
# 跟踪函数调用
# 如果设置为 1，vllm 将跟踪函数调用
# 对调试很有用
"VLLM_TRACE_FUNCTION":
lambda: int(os.getenv("VLLM_TRACE_FUNCTION", "0")),
# Backend for attention computation
# Available options:
# - "TORCH_SDPA": use torch.nn.MultiheadAttention
# - "FLASH_ATTN": use FlashAttention
# - "XFORMERS": use XFormers
# - "ROCM_FLASH": use ROCmFlashAttention
# - "FLASHINFER": use flashinfer
# - "FLASHMLA": use FlashMLA
# 注意力计算的后端
# 可用选项：
# - "TORCH_SDPA"：使用 torch.nn.MultiheadAttention
# - "FLASH_ATTN"：使用 FlashAttention
# - "XFORMERS"：使用 XFormers
# - "ROCM_FLASH"：使用 ROCmFlashAttention
# - "FLASHINFER"：使用 flashinfer
# - "FLASHMLA"：使用 FlashMLA
"VLLM_ATTENTION_BACKEND":
lambda: os.getenv("VLLM_ATTENTION_BACKEND", None),
# If set, vllm will use flashinfer sampler
# 如果设置，vllm 将使用 flashinfer 采样器
"VLLM_USE_FLASHINFER_SAMPLER":
lambda: bool(int(os.environ["VLLM_USE_FLASHINFER_SAMPLER"]))
if "VLLM_USE_FLASHINFER_SAMPLER" in os.environ else None,
# If set, vllm will force flashinfer to use tensor cores;
# otherwise will use heuristic based on model architecture.
# 如果设置，vllm 将强制 flashinfer 使用张量核心；否则将基于模型架构使用启发式方法
"VLLM_FLASHINFER_FORCE_TENSOR_CORES":
lambda: bool(int(os.getenv("VLLM_FLASHINFER_FORCE_TENSOR_CORES", "0"))),
# Pipeline stage partition strategy
# 流水线阶段分区策略
"VLLM_PP_LAYER_PARTITION":
lambda: os.getenv("VLLM_PP_LAYER_PARTITION", None),
# (CPU backend only) CPU key-value cache space.
# default is 4GB
# （仅 CPU 后端）CPU 键值缓存空间
# 默认为 4GB
"VLLM_CPU_KVCACHE_SPACE":
lambda: int(os.getenv("VLLM_CPU_KVCACHE_SPACE", "0")),
# (CPU backend only) CPU core ids bound by OpenMP threads, e.g., "0-31",
# "0,1,2", "0-31,33". CPU cores of different ranks are separated by '|'.
# （仅 CPU 后端）OpenMP 线程绑定的 CPU 核心 ID，例如 "0-31"、"0,1,2"、"0-31,33"。不同 rank 的 CPU 核心用 '|' 分隔
"VLLM_CPU_OMP_THREADS_BIND":
lambda: os.getenv("VLLM_CPU_OMP_THREADS_BIND", "all"),
# (CPU backend only) whether to use prepack for MoE layer. This will be
# passed to ipex.llm.modules.GatedMLPMOE. On unsupported CPUs, you might
# need to set this to "0" (False).
# （仅 CPU 后端）是否对 MoE 层使用预打包。这将传递给 ipex.llm.modules.GatedMLPMOE。在不支持的 CPU 上，你可能需要将其设置为 "0"（False）。
"VLLM_CPU_MOE_PREPACK":
lambda: bool(int(os.getenv("VLLM_CPU_MOE_PREPACK", "1"))),
# OpenVINO device selection
# default is CPU
# OpenVINO 设备选择
# 默认为 CPU
"VLLM_OPENVINO_DEVICE":
lambda: os.getenv("VLLM_OPENVINO_DEVICE", "CPU").upper(),
# OpenVINO key-value cache space
# default is 4GB
# OpenVINO 键值缓存空间
# 默认为 4GB
"VLLM_OPENVINO_KVCACHE_SPACE":
lambda: int(os.getenv("VLLM_OPENVINO_KVCACHE_SPACE", "0")),
# OpenVINO KV cache precision
# default is bf16 if natively supported by platform, otherwise f16
# To enable KV cache compression, please, explicitly specify u8
# OpenVINO KV 缓存精度
# 如果平台原生支持，则默认为 bf16，否则为 f16
# 要启用 KV 缓存压缩，请显式指定 u8
"VLLM_OPENVINO_CPU_KV_CACHE_PRECISION":
lambda: os.getenv("VLLM_OPENVINO_CPU_KV_CACHE_PRECISION", None),
# Enables weights compression during model export via HF Optimum
# default is False
# 在通过 HF Optimum 导出模型时启用权重压缩
# 默认为 False
"VLLM_OPENVINO_ENABLE_QUANTIZED_WEIGHTS":
lambda:
(os.environ.get("VLLM_OPENVINO_ENABLE_QUANTIZED_WEIGHTS", "0").lower() in
("on", "true", "1")),
# If the env var is set, then all workers will execute as separate
# processes from the engine, and we use the same mechanism to trigger
# execution on all workers.
# Run vLLM with VLLM_USE_RAY_SPMD_WORKER=1 to enable it.
# 如果设置了环境变量，则所有工作者将作为与引擎分离的进程执行，并且我们使用相同的机制在所有工作者上触发执行。
# 运行 vLLM 时设置 VLLM_USE_RAY_SPMD_WORKER=1 以启用此功能。
"VLLM_USE_RAY_SPMD_WORKER":
lambda: bool(int(os.getenv("VLLM_USE_RAY_SPMD_WORKER", "0"))),
# If the env var is set, it uses the Ray's Compiled Graph
# (previously known as ADAG) API which optimizes the
# control plane overhead.
# Run vLLM with VLLM_USE_RAY_COMPILED_DAG=1 to enable it.
# 如果设置了环境变量，它将使用 Ray 的 Compiled Graph（以前称为 ADAG）API，该 API 优化了控制平面开销。
# 运行 vLLM 时设置 VLLM_USE_RAY_COMPILED_DAG=1 以启用此功能。
"VLLM_USE_RAY_COMPILED_DAG":
lambda: bool(int(os.getenv("VLLM_USE_RAY_COMPILED_DAG", "0"))),
# If the env var is set, it uses NCCL for communication in
# Ray's Compiled Graph. This flag is ignored if
# VLLM_USE_RAY_COMPILED_DAG is not set.
# 如果设置了环境变量，它将在 Ray 的 Compiled Graph 中使用 NCCL 进行通信。如果未设置 VLLM_USE_RAY_COMPILED_DAG，则忽略此标志。
"VLLM_USE_RAY_COMPILED_DAG_NCCL_CHANNEL":
lambda: bool(int(os.getenv("VLLM_USE_RAY_COMPILED_DAG_NCCL_CHANNEL", "1"))
),
# If the env var is set, it enables GPU communication overlap
# (experimental feature) in Ray's Compiled Graph. This flag is ignored if
# VLLM_USE_RAY_COMPILED_DAG is not set.
# 如果设置了环境变量，它将在 Ray 的 Compiled Graph 中启用 GPU 通信重叠（实验性功能）。如果未设置 VLLM_USE_RAY_COMPILED_DAG，则忽略此标志。
"VLLM_USE_RAY_COMPILED_DAG_OVERLAP_COMM":
lambda: bool(int(os.getenv("VLLM_USE_RAY_COMPILED_DAG_OVERLAP_COMM", "0"))
),
# Use dedicated multiprocess context for workers.
# Both spawn and fork work
# 为工作者使用专用的多进程上下文。
# spawn 和 fork 都有效
"VLLM_WORKER_MULTIPROC_METHOD":
lambda: os.getenv("VLLM_WORKER_MULTIPROC_METHOD", "fork"),
# Path to the cache for storing downloaded assets
# 存储下载资源的缓存路径
"VLLM_ASSETS_CACHE":
lambda: os.path.expanduser(
os.getenv(
"VLLM_ASSETS_CACHE",
os.path.join(get_default_cache_root(), "vllm", "assets"),
)),
# Timeout for fetching images when serving multimodal models
# Default is 5 seconds
# 为多模态模型服务时获取图像的超时时间
# 默认为 5 秒
"VLLM_IMAGE_FETCH_TIMEOUT":
lambda: int(os.getenv("VLLM_IMAGE_FETCH_TIMEOUT", "5")),
# Timeout for fetching videos when serving multimodal models
# Default is 30 seconds
# 为多模态模型服务时获取视频的超时时间
# 默认为 30 秒
"VLLM_VIDEO_FETCH_TIMEOUT":
lambda: int(os.getenv("VLLM_VIDEO_FETCH_TIMEOUT", "30")),
# Timeout for fetching audio when serving multimodal models
# Default is 10 seconds
# 为多模态模型服务时获取音频的超时时间
# 默认为 10 秒
"VLLM_AUDIO_FETCH_TIMEOUT":
lambda: int(os.getenv("VLLM_AUDIO_FETCH_TIMEOUT", "10")),
# Cache size (in GiB) for multimodal input cache
# Default is 8GiB
# 多模态输入缓存的缓存大小（以 GiB 为单位）
# 默认为 8GiB
"VLLM_MM_INPUT_CACHE_GIB":
lambda: int(os.getenv("VLLM_MM_INPUT_CACHE_GIB", "8")),
# Path to the XLA persistent cache directory.
# Only used for XLA devices such as TPUs.
# XLA 持久缓存目录的路径。
# 仅用于 XLA 设备（如 TPU）。
"VLLM_XLA_CACHE_PATH":
lambda: os.path.expanduser(
os.getenv(
"VLLM_XLA_CACHE_PATH",
os.path.join(get_default_cache_root(), "vllm", "xla_cache"),
)),
"VLLM_FUSED_MOE_CHUNK_SIZE":
lambda: int(os.getenv("VLLM_FUSED_MOE_CHUNK_SIZE", "32768")),
# If set, vllm will skip the deprecation warnings.
# 如果设置，vllm 将跳过弃用警告。
"VLLM_NO_DEPRECATION_WARNING":
lambda: bool(int(os.getenv("VLLM_NO_DEPRECATION_WARNING", "0"))),
# If set, the OpenAI API server will stay alive even after the underlying
# AsyncLLMEngine errors and stops serving requests
# 如果设置，即使底层的 AsyncLLMEngine 出错并停止服务请求，OpenAI API 服务器也将保持活动状态
"VLLM_KEEP_ALIVE_ON_ENGINE_DEATH":
lambda: bool(os.getenv("VLLM_KEEP_ALIVE_ON_ENGINE_DEATH", 0)),
# If the env var VLLM_ALLOW_LONG_MAX_MODEL_LEN is set, it allows
# the user to specify a max sequence length greater than
# the max length derived from the model's config.json.
# To enable this, set VLLM_ALLOW_LONG_MAX_MODEL_LEN=1.
# 如果设置了环境变量 VLLM_ALLOW_LONG_MAX_MODEL_LEN，则允许用户指定大于从模型 config.json 派生的最大长度的最大序列长度。
# 要启用此功能，请设置 VLLM_ALLOW_LONG_MAX_MODEL_LEN=1。
"VLLM_ALLOW_LONG_MAX_MODEL_LEN":
lambda:
(os.environ.get("VLLM_ALLOW_LONG_MAX_MODEL_LEN", "0").strip().lower() in
("1", "true")),
# If set, forces FP8 Marlin to be used for FP8 quantization regardless
# of the hardware support for FP8 compute.
# 如果设置，无论硬件是否支持 FP8 计算，都将强制使用 FP8 Marlin 进行 FP8 量化。
"VLLM_TEST_FORCE_FP8_MARLIN":
lambda:
(os.environ.get("VLLM_TEST_FORCE_FP8_MARLIN", "0").strip().lower() in
("1", "true")),
"VLLM_TEST_FORCE_LOAD_FORMAT":
lambda: os.getenv("VLLM_TEST_FORCE_LOAD_FORMAT", "dummy"),
# Time in ms for the zmq client to wait for a response from the backend
# server for simple data operations
# zmq 客户端等待后端服务器响应简单数据操作的时间（以毫秒为单位）
"VLLM_RPC_TIMEOUT":
lambda: int(os.getenv("VLLM_RPC_TIMEOUT", "10000")),
# a list of plugin names to load, separated by commas.
# if this is not set, it means all plugins will be loaded
# if this is set to an empty string, no plugins will be loaded
# 要加载的插件名称列表，以逗号分隔。
# 如果未设置，则表示将加载所有插件。
# 如果设置为空字符串，则不加载任何插件。
"VLLM_PLUGINS":
lambda: None if "VLLM_PLUGINS" not in os.environ else os.environ[
"VLLM_PLUGINS"].split(","),
# Enables torch profiler if set. Path to the directory where torch profiler
# traces are saved. Note that it must be an absolute path.
# 如果设置，则启用 torch profiler。torch profiler 跟踪保存的目录路径。注意，它必须是绝对路径。
"VLLM_TORCH_PROFILER_DIR":
lambda: (None if os.getenv("VLLM_TORCH_PROFILER_DIR", None) is None else os
.path.expanduser(os.getenv("VLLM_TORCH_PROFILER_DIR", "."))),
# If set, vLLM will use Triton implementations of AWQ.
# 如果设置，vLLM 将使用 Triton 实现的 AWQ。
"VLLM_USE_TRITON_AWQ":
lambda: bool(int(os.getenv("VLLM_USE_TRITON_AWQ", "0"))),
# If set, allow loading or unloading lora adapters in runtime,
# 如果设置，允许在运行时加载或卸载 lora 适配器，
"VLLM_ALLOW_RUNTIME_LORA_UPDATING":
lambda:
(os.environ.get("VLLM_ALLOW_RUNTIME_LORA_UPDATING", "0").strip().lower() in
("1", "true")),
# By default, vLLM will check the peer-to-peer capability itself,
# in case of broken drivers. See https://github.com/vllm-project/vllm/blob/a9b15c606fea67a072416ea0ea115261a2756058/vllm/distributed/device_communicators/custom_all_reduce_utils.py#L101-L108 for details. # noqa
# If this env var is set to 1, vLLM will skip the peer-to-peer check,
# and trust the driver's peer-to-peer capability report.
# 默认情况下，vLLM 会自行检查点对点能力，以防驱动程序损坏。有关详细信息，请参阅 https://github.com/vllm-project/vllm/blob/a9b15c606fea67a072416ea0ea115261a2756058/vllm/distributed/device_communicators/custom_all_reduce_utils.py#L101-L108。
# 如果此环境变量设置为 1，vLLM 将跳过点对点检查，并信任驱动程序的点对点能力报告。
"VLLM_SKIP_P2P_CHECK":
lambda: os.getenv("VLLM_SKIP_P2P_CHECK", "0") == "1",
# List of quantization kernels that should be disabled, used for testing
# and performance comparisons. Currently only affects MPLinearKernel
# selection
# (kernels: MacheteLinearKernel, MarlinLinearKernel, ExllamaLinearKernel)
# 应禁用的量化内核列表，用于测试和性能比较。目前仅影响 MPLinearKernel 选择
# （内核：MacheteLinearKernel、MarlinLinearKernel、ExllamaLinearKernel）
"VLLM_DISABLED_KERNELS":
lambda: [] if "VLLM_DISABLED_KERNELS" not in os.environ else os.environ[
"VLLM_DISABLED_KERNELS"].split(","),
# If set, use the V1 code path.
# 如果设置，则使用 V1 代码路径。
"VLLM_USE_V1":
lambda: bool(int(os.getenv("VLLM_USE_V1", "1"))),
# Pad the fp8 weights to 256 bytes for ROCm
# 为 ROCm 将 fp8 权重填充到 256 字节
"VLLM_ROCM_FP8_PADDING":
lambda: bool(int(os.getenv("VLLM_ROCM_FP8_PADDING", "1"))),
# Divisor for dynamic key scale factor calculation for FP8 KV Cache
# 用于 FP8 KV 缓存的动态键缩放因子计算的除数
"K_SCALE_CONSTANT":
lambda: int(os.getenv("K_SCALE_CONSTANT", "200")),
# Divisor for dynamic value scale factor calculation for FP8 KV Cache
# 用于 FP8 KV 缓存的动态值缩放因子计算的除数
"V_SCALE_CONSTANT":
lambda: int(os.getenv("V_SCALE_CONSTANT", "100")),
# If set, enable multiprocessing in LLM for the V1 code path.
# 如果设置，则在 V1 代码路径中启用 LLM 的多进程处理。
"VLLM_ENABLE_V1_MULTIPROCESSING":
lambda: bool(int(os.getenv("VLLM_ENABLE_V1_MULTIPROCESSING", "1"))),
"VLLM_LOG_BATCHSIZE_INTERVAL":
lambda: float(os.getenv("VLLM_LOG_BATCHSIZE_INTERVAL", "-1")),
"VLLM_DISABLE_COMPILE_CACHE":
lambda: bool(int(os.getenv("VLLM_DISABLE_COMPILE_CACHE", "0"))),
# If set, vllm will run in development mode, which will enable
# some additional endpoints for developing and debugging,
# e.g. `/reset_prefix_cache`
# 如果设置，vllm 将以开发模式运行，这将启用一些用于开发和调试的额外端点，例如 `/reset_prefix_cache`
"VLLM_SERVER_DEV_MODE":
lambda: bool(int(os.getenv("VLLM_SERVER_DEV_MODE", "0"))),
# Controls the maximum number of requests to handle in a
# single asyncio task when processing per-token outputs in the
# V1 AsyncLLM interface. It is applicable when handling a high
# concurrency of streaming requests.
# Setting this too high can result in a higher variance of
# inter-message latencies. Setting it too low can negatively impact
# TTFT and overall throughput.
# 控制在 V1 AsyncLLM 接口中处理每个 token 输出时，单个 asyncio 任务中处理的最大请求数。适用于处理高并发的流式请求。
# 将此值设置得太高可能会导致消息间延迟的更大差异。设置得太低可能会对 TTFT 和整体吞吐量产生负面影响。
"VLLM_V1_OUTPUT_PROC_CHUNK_SIZE":
lambda: int(os.getenv("VLLM_V1_OUTPUT_PROC_CHUNK_SIZE", "128")),
# If set, vLLM will disable the MLA attention optimizations.
# 如果设置，vLLM 将禁用 MLA 注意力优化。
"VLLM_MLA_DISABLE":
lambda: bool(int(os.getenv("VLLM_MLA_DISABLE", "0"))),
# If set, vLLM will use the Triton implementation of moe_align_block_size,
# i.e. moe_align_block_size_triton in fused_moe.py.
# 如果设置，vLLM 将使用 Triton 实现的 moe_align_block_size，即 fused_moe.py 中的 moe_align_block_size_triton。
"VLLM_ENABLE_MOE_ALIGN_BLOCK_SIZE_TRITON":
lambda: bool(int(os.getenv("VLLM_ENABLE_MOE_ALIGN_BLOCK_SIZE_TRITON", "0"))
),
# Number of GPUs per worker in Ray, if it is set to be a fraction,
# it allows ray to schedule multiple actors on a single GPU,
# so that users can colocate other actors on the same GPUs as vLLM.
# Ray 中每个工作者的 GPU 数量，如果设置为分数，则允许 Ray 在单个 GPU 上调度多个 actor，以便用户可以将其他 actor 与 vLLM 放在同一 GPU 上。
"VLLM_RAY_PER_WORKER_GPUS":
lambda: float(os.getenv("VLLM_RAY_PER_WORKER_GPUS", "1.0")),
# Bundle indices for Ray, if it is set, it can control precisely
# which indices are used for the Ray bundle, for every worker.
# Format: comma-separated list of integers, e.g. "0,1,2,3"
# Ray 的 bundle 索引，如果设置，它可以精确控制每个工作者使用的 Ray bundle 的索引。
# 格式：逗号分隔的整数列表，例如 "0,1,2,3"
"VLLM_RAY_BUNDLE_INDICES":
lambda: os.getenv("VLLM_RAY_BUNDLE_INDICES", ""),
# In some system, find_loaded_library() may not work. So we allow users to
# specify the path through environment variable VLLM_CUDART_SO_PATH.
# 在某些系统中，find_loaded_library() 可能无法工作。因此，我们允许用户通过环境变量 VLLM_CUDART_SO_PATH 指定路径。
"VLLM_CUDART_SO_PATH":
lambda: os.getenv("VLLM_CUDART_SO_PATH", None),
# Contiguous cache fetching to avoid using costly gather operation on
# Gaudi3. This is only applicable to HPU contiguous cache. If set to true,
# contiguous cache fetch will be used.
# 连续缓存获取，以避免在 Gaudi3 上使用昂贵的 gather 操作。这仅适用于 HPU 连续缓存。如果设置为 true，则将使用连续缓存获取。
"VLLM_USE_HPU_CONTIGUOUS_CACHE_FETCH":
lambda: os.environ.get("VLLM_CONTIGUOUS_PA", "true").lower() in
("1", "true"),
# Rank of the process in the data parallel setting
# 数据并行设置中进程的 rank
"VLLM_DP_RANK":
lambda: int(os.getenv("VLLM_DP_RANK", "0")),
# World size of the data parallel setting
# 数据并行设置中的 world size
"VLLM_DP_SIZE":
lambda: int(os.getenv("VLLM_DP_SIZE", "1")),
# IP address of the master node in the data parallel setting
# 数据并行设置中主节点的 IP 地址
"VLLM_DP_MASTER_IP":
lambda: os.getenv("VLLM_DP_MASTER_IP", "127.0.0.1"),
# Port of the master node in the data parallel setting
# 数据并行设置中主节点的端口
"VLLM_DP_MASTER_PORT":
lambda: int(os.getenv("VLLM_DP_MASTER_PORT", "0")),
# Whether to use S3 path for model loading in CI via RunAI Streamer
# 是否在 CI 中通过 RunAI Streamer 使用 S3 路径加载模型
"VLLM_CI_USE_S3":
lambda: os.environ.get("VLLM_CI_USE_S3", "0") == "1",
# Whether to use atomicAdd reduce in gptq/awq marlin kernel.
# 是否在 gptq/awq marlin 内核中使用 atomicAdd reduce。
"VLLM_MARLIN_USE_ATOMIC_ADD":
lambda: os.environ.get("VLLM_MARLIN_USE_ATOMIC_ADD", "0") == "1",
# Whether to turn on the outlines cache for V0
# This cache is unbounded and on disk, so it's not safe to use in
# an environment with potentially malicious users.
# 是否为 V0 启用 outlines 缓存
# 此缓存是无限制的并且存储在磁盘上，因此在可能存在恶意用户的环境中不安全使用。
"VLLM_V0_USE_OUTLINES_CACHE":
lambda: os.environ.get("VLLM_V0_USE_OUTLINES_CACHE", "0") == "1",
}