Chapter 2. vLLM server usage

The model tag to serve. Optional if specified in the config.

Default: None

Allow credentials.

Default: False

Allowed headers.

Default: ['*']

Allowed methods.

Default: ['*']

Allowed origins.

Default: ['*'])

If provided, the server will require this key to be presented in the header.

Default: None

How many API server processes to run.

Default: 1

The file path to the chat template, or the template in single-line form for the specified model.

Default: None

The format to render message content within a chat template. * "string" will render the content as a string. Example: "Hello World" * "openai" will render the content as a list of dictionaries, similar to OpenAI schema. Example: [{"type": "text", "text": "Hello world!"}]

Default: auto

Read CLI options from a config file.Must be a YAML with the following options: https://docs.vllm.ai/en/latest/serving/openai_compatible_server.html#cli-reference

Default:

Starting data parallel rank for secondary nodes.

Default: 0

Disable FastAPI’s OpenAPI schema, Swagger UI, and ReDoc endpoint.

Default: False

If specified, will run the OpenAI frontend server in the same process as the model serving engine.

Default: False

Disable logging requests.

Default: False

Disable logging statistics.

Default: False

Disable uvicorn access log.

Default: False

Enable auto tool choice for supported models. Use --tool-call-parser to specify which parser to use.

Default: False

If set to True, enable prompt_tokens_details in usage.

Default: False

If specified, API server will add X-Request-Id header to responses. Caution: this hurts performance at high QPS.

Default: False

If set to True, enable tracking server_load_metrics in the app state.

Default: False

Refresh SSL Context when SSL certificate files change

Default: False

Run in headless mode. See multi-node data parallel documentation for more details.

Default: False

Host name.

Default: None

Path to logging config JSON file for both vllm and uvicorn

Default: None

LoRA module configurations in either name=path formator JSON format. Example (old format): name=path. Example (new format): {"name": "name", "path": "lora_path", "base_model_name": "id"}

Default: None

Max number of prompt characters or prompt ID numbers being printed in log. The default of None means unlimited.

Default: None

Additional ASGI middleware to apply to the app. We accept multiple --middleware arguments. The value should be an import path. If a function is provided, vLLM will add it to the server using @app.middleware('http'). If a class is provided, vLLM will add it to the server using app.add_middleware().

Default: []

Prompt adapter configurations in the format name=path. Multiple adapters can be specified.

Default: None

The role name to return if request.add_generation_prompt=true.

Default: assistant

When --max-logprobs is specified, represents single tokens as strings of the form 'token_id:{token_id}' so that tokens that are not JSON-encodable can be identified.

Default: False

FastAPI root_path when app is behind a path based routing proxy.

Default: None

The CA certificates file.

Default: None

Whether client certificate is required (see stdlib ssl module).

Default: 0

The file path to the SSL cert file.

Default: None

The file path to the SSL key file.

Default: None

Or name registered in --tool-parser-plugin. Select the tool call parser depending on the model that you’re using. This is used to parse the model-generated tool call into OpenAI API format. Required for --enable-auto-tool-choice.

Default: None

Special the tool parser plugin write to parse the model-generated tool into OpenAI API format, the name register in this plugin can be used in --tool-call-parser.

Default:

DEPRECATED: block manager v1 has been removed and SelfAttnBlockSpaceManager (i.e. block manager v2) is now the default. Setting this flag to True or False has no effect on vLLM behavior.

Default: True

Allowing API requests to read local images or videos from directories specified by the server file system. This is a security risk. Should only be enabled in trusted environments.

Default:

The specific revision to use for the model code on the Hugging Face Hub. It can be a branch name, a tag name, or a commit id. If unspecified, will use the default version.

Default: None

The format of the model config to load: * auto will try to load the config in hf format if available else it will try to load in mistral format. * hf will load the config in hf format. * mistral will load the config in mistral format.

Default: auto

Disable async output processing. This may result in lower performance.

Default: False

Disable cascade attention for V1. While cascade attention does not change the mathematical correctness, disabling it could be useful for preventing potential numerical issues. Note that even if this is set to False, cascade attention will be only used when the heuristic tells that it’s beneficial.

Default: False

Whether to disable sliding window. If True, we will disable the sliding window functionality of the model, capping to sliding window size. If the model does not support sliding window, this argument is ignored.

Default: False

Data type for model weights and activations: * auto will use FP16 precision for FP32 and FP16 models, and BF16 precision for BF16 models. * half for FP16. Recommended for AWQ quantization. * float16 is the same as "half". * bfloat16 for a balance between precision and range. * float is shorthand for FP32 precision. * float32 for FP32 precision.

Default: auto

If True, enables passing text embeddings as inputs via the prompt_embeds key. Note that enabling this will double the time required for graph compilation.

Default: False

Enable sleep mode for the engine (only CUDA platform is supported).

Default: False

Whether to always use eager-mode PyTorch. If True, we will disable CUDA graph and always execute the model in eager mode. If False, we will use CUDA graph and eager execution in hybrid for maximal performance and flexibility.

Default: False

The folder path to the generation config. Defaults to auto, the generation config will be loaded from model path. If set to vllm, no generation config is loaded, vLLM defaults will be used. If set to a folder path, the generation config will be loaded from the specified folder path. If max_new_tokens is specified in generation config, then it sets a server-wide limit on the number of output tokens for all requests.

Default: auto

Name or path of the Hugging Face config to use. If unspecified, model name or path will be used.

Default: None

If a dictionary, contains arguments to be forwarded to the HuggingFace config. If a callable, it is called to update the HuggingFace config.

Default: {}

The token to use as HTTP bearer authorization for remote files . If True, will use the token generated when running huggingface-cli login (stored in ~/.huggingface).

Default: None

Optional regex pattern specifying valid logits processor qualified names that can be passed with the logits_processors extra completion argument. Defaults to None, which allows no processors.

Default: None

Maximum number of log probabilities to return when logprobs is specified in SamplingParams. The default value comes the default for the OpenAI Chat Completions API.

Default: 20

Model context length (prompt and output). If unspecified, will be automatically derived from the model config. When passing via --max-model-len, supports k/m/g/K/M/G in human-readable format. Examples: * 1k 1000 * 1K 1024 * 25.6k 25,600

Default: None

Maximum sequence len covered by CUDA graphs. When a sequence has context length larger than this, we fall back to eager mode. Additionally for encoder-decoder models, if the sequence length of the encoder input is larger than this, we fall back to the eager mode.

Default: 8192

Name or path of the Hugging Face model to use. It is also used as the content for model_name tag in metrics output when served_model_name is not specified.

Default: facebook/opt-125m

Which implementation of the model to use: * auto will try to use the vLLM implementation, if it exists, and fall back to the Transformers implementation if no vLLM implementation is available. * vllm will use the vLLM model implementation. * transformers will use the Transformers model implementation.

Default: auto

Overrides or sets generation config. e.g. {"temperature": 0.5}. If used with --generation-config auto, the override parameters will be merged with the default config from the model. If used with --generation-config vllm, only the override parameters are used. 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"}}' * --json-arg.key1 value1 --json-arg.key2.key3 value2

Default: {}

Initialize non-default neuron config or override default neuron config that are specific to Neuron devices, this argument will be used to configure the neuron config that can not be gathered from the vllm arguments. e.g. {"cast_logits_dtype": "bfloat16"}. 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"}}' * --json-arg.key1 value1 --json-arg.key2.key3 value2

Default: {}

Initialize non-default pooling config or override default pooling config for the pooling model. For example, {"pooling_type": "mean", "normalize": false}.

Default: None

Method used to quantize the weights. If None, we first check the quantization_config attribute in the model config file. If that is None, we assume the model weights are not quantized and use dtype to determine the data type of the weights.

Default: None

The specific model version to use. It can be branch name, a tag name, or a commit id. If unspecified, will use the default version.

Default: None

RoPE scaling configuration. For example, {"rope_type":"dynamic","factor":2.0}. 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"}}' * --json-arg.key1 value1 --json-arg.key2.key3 value2

Default: {}

RoPE theta. Use with rope_scaling. In some cases, changing the RoPE theta improves the performance of the scaled model.

Default: None

Random seed for reproducibility. Initialized to None in V0, but initialized to 0 in V1.

Default: None

The model name(s) used in the API. If multiple names are provided, the server will respond to any of the provided names. The model name in the model field of a response will be the first name in this list. If not specified, the model name will be the same as the --model argument. Noted that this name(s) will also be used in model_name tag content of prometheus metrics, if multiple names provided, metrics tag will take the first one.

Default: None

Skip initialization of tokenizer and detokenizer. Expects valid prompt_token_ids and None for prompt from the input. The generated output will contain token ids.

Default: False

The task to use the model for. Each vLLM instance only supports one task, even if the same model can be used for multiple tasks. When the model only supports one task, auto can be used to select it; otherwise, you must specify explicitly which task to use.

Default: auto

Name or path of the Hugging Face tokenizer to use. If unspecified, model name or path will be used.

Default: None

Tokenizer mode: * auto will use the fast tokenizer if available. * slow will always use the slow tokenizer. * mistral will always use the tokenizer from mistral_common. * custom will use --tokenizer to select the preregistered tokenizer.

Default: auto

The specific revision to use for the tokenizer on the Hugging Face Hub. It can be a branch name, a tag name, or a commit id. If unspecified, will use the default version.

Default: None

Trust remote code (e.g., from HuggingFace) when downloading the model and tokenizer.

Default: False

Directory to download and load the weights, default to the default cache directory of Hugging Face.

Default: None

The list of patterns to ignore when loading the model. Defaults to "original/*/" to avoid repeated loading of llama’s checkpoints.

Default: None

The format of the model weights to load:

Default: auto

Extra config for model loader. This will be passed to the model loader corresponding to the chosen load_format.

Default: {}

PT_LOAD_MAP_LOCATION: the map location for loading pytorch checkpoint, to support loading checkpoints can only be loaded on certain devices like "CUDA", this is equivalent to {"": "CUDA"}. Another supported format is mapping from different devices like from GPU 1 to GPU 0: {"CUDA:1": "CUDA:0"}. Note that when passed from command line, the strings in dictionary needs to be double quoted for JSON parsing. For more details, see original doc for map_location in https://pytorch.org/docs/stable/generated/torch.load.html

Default: cpu

The --qlora-adapter-name-or-path has no effect, do not set it, and it will be removed in v0.10.0.

Default: None

Whether to enable tqdm for showing progress bar when loading model weights.

Default: True

[DEPRECATED] The --enable-reasoning flag is deprecated as of v0.9.0. Use --reasoning-parser to specify the reasoning parser backend instead. This flag (--enable-reasoning) will be removed in v0.10.0. When --reasoning-parser is specified, reasoning mode is automatically enabled.

Default: None

Which engine will be used for guided decoding (JSON schema / regex etc) by default. With auto, we will make opinionated choices based on request contents and what the backend libraries currently support, so the behavior is subject to change in each release.

Default: auto

If True, the guidance backend will not use additionalProperties in the JSON schema. This is only supported for the guidance backend and is used to better align its behaviour with outlines and xgrammar.

Default: False

If True, the model will not generate any whitespace during guided decoding. This is only supported for xgrammar and guidance backends.

Default: False

If True, vLLM will not fallback to a different backend on error.

Default: False

Select the reasoning parser depending on the model that you’re using. This is used to parse the reasoning content into OpenAI API format.

Default:

Address of data parallel cluster head-node.

Default: None

Backend for data parallel, either mp or ray.

Default: mp

Port for data parallel RPC communication.

Default: None

Number of data parallel groups. MoE layers will be sharded according to the product of the tensor parallel size and data parallel size.

Default: 1

Number of data parallel replicas to run on this node.

Default: None

Disable the custom all-reduce kernel and fall back to NCCL.

Default: False

Backend to use for distributed model workers, either ray or mp (multiprocessing). If the product of pipeline_parallel_size and tensor_parallel_size is less than or equal to the number of GPUs available, mp will be used to keep processing on a single host. Otherwise, this will default to ray if Ray is installed and fail otherwise. Note that TPU and HPU only support Ray for distributed inference.

Default: None

Use expert parallelism instead of tensor parallelism for MoE layers.

Default: False

Use data parallelism instead of tensor parallelism for vision encoder. Only support LLama4 for now

Default: False

Maximum number of parallel loading workers when loading model sequentially in multiple batches. To avoid RAM OOM when using tensor parallel and large models.

Default: None

Number of pipeline parallel groups.

Default: 1

Whether to profile Ray workers with nsight, see https://docs.ray.io/en/latest/ray-observability/user-guides/profiling.html#profiling-nsight-profiler.

Default: False

Number of tensor parallel groups.

Default: 1

The full name of the worker class to use. If auto, the worker class will be determined based on the platform.

Default: auto

The full name of the worker extension class to use. The worker extension class is dynamically inherited by the worker class. This is used to inject new attributes and methods to the worker class for use in collective_rpc calls.

Default:

Size of a contiguous cache block in number of tokens. This is ignored on neuron devices and set to --max-model-len. On CUDA devices, only block sizes up to 32 are supported. On HPU devices, block size defaults to 128. This config has no static default. If left unspecified by the user, it will be set in Platform.check_and_update_configs() based on the current platform.

Default: None

This enables dynamic calculation of k_scale and v_scale when kv_cache_dtype is fp8. If False, the scales will be loaded from the model checkpoint if available. Otherwise, the scales will default to 1.0.

Default: False

The space in GiB to offload to CPU, per GPU. Default is 0, which means no offloading. Intuitively, this argument can be seen as a virtual way to increase the GPU memory size. For example, if you have one 24 GB GPU and set this to 10, virtually you can think of it as a 34 GB GPU. Then you can load a 13B model with BF16 weight, which requires at least 26GB GPU memory. Note that this requires fast CPU-GPU interconnect, as part of the model is loaded from CPU memory to GPU memory on the fly in each model forward pass.

Default: 0

Whether to enable prefix caching. Disabled by default for V0. Enabled by default for V1.

Default: None

The fraction of GPU memory to be used for the model executor, which can range from 0 to 1. For example, a value of 0.5 would imply 50% GPU memory utilization. If unspecified, will use the default value of 0.9. This is a per-instance limit, and only applies to the current vLLM instance. It does not matter if you have another vLLM instance running on the same GPU. For example, if you have two vLLM instances running on the same GPU, you can set the GPU memory utilization to 0.5 for each instance.

Default: 0.9

Data type for kv cache storage. If auto, will use model data type. CUDA 11.8+ supports fp8 (=fp8_e4m3) and fp8_e5m2. ROCm (AMD GPU) supports fp8 (=fp8_e4m3).

Default: auto

Number of GPU blocks to use. This overrides the profiled num_gpu_blocks if specified. Does nothing if None. Used for testing preemption.

Default: None

Set the hash algorithm for prefix caching: * "builtin" is Python’s built-in hash. * "sha256" is collision resistant but with certain overheads.

Default: builtin

Size of the CPU swap space per GPU (in GiB).

Default: 4

If True, disable caching of the processed multi-modal inputs.

Default: False

The maximum number of input items allowed per prompt for each modality. Defaults to 1 (V0) or 999 (V1) for each modality. For example, to allow up to 16 images and 2 videos per prompt: {"images": 16, "videos": 2} 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"}}' * --json-arg.key1 value1 --json-arg.key2.key3 value2

Default: {}

Overrides for the multi-modal processor obtained from transformers.AutoProcessor.from_pretrained. The available overrides depend on the model that is being run. For example, for Phi-3-Vision: {"num_crops": 4}. 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"}}' * --json-arg.key1 value1 --json-arg.key2.key3 value2

Default: None

If True, enable handling of LoRA adapters.

Default: None

Enable bias for LoRA adapters.

Default: False

By default, only half of the LoRA computation is sharded with tensor parallelism. Enabling this will use the fully sharded layers. At high sequence length, max rank or tensor parallel size, this is likely faster.

Default: False

Specify multiple scaling factors (which can be different from base model scaling factor) to allow for multiple LoRA adapters trained with those scaling factors to be used at the same time. If not specified, only adapters trained with the base model scaling factor are allowed.

Default: None

Data type for LoRA. If auto, will default to base model dtype.

Default: auto

Maximum size of extra vocabulary that can be present in a LoRA adapter (added to the base model vocabulary).

Default: 256

Maximum number of LoRAs to store in CPU memory. Must be greater than max_loras.

Default: None

Max LoRA rank.

Default: 16

Max number of LoRAs in a single batch.

Default: 1

If True, enable handling of PromptAdapters.

Default: None

Max number of PromptAdapters tokens.

Default: 0

Max number of PromptAdapters in a batch.

Default: 1

Device type for vLLM execution. This parameter is deprecated and will be removed in a future release. It will now be set automatically based on the current platform.

Default: auto

The configurations for speculative decoding. Should be a JSON string.

Default: None

It makes sense to set this only if --otlp-traces-endpoint is set. If set, it will collect detailed traces for the specified modules. This involves use of possibly costly and or blocking operations and hence might have a performance impact. Note that collecting detailed timing information for each request can be expensive.

Default: None

Target URL to which OpenTelemetry traces will be sent.

Default: None

Enable deprecated Prometheus metrics that have been hidden since the specified version. For example, if a previously deprecated metric has been hidden since the v0.7.0 release, you use --show-hidden-metrics-for-version=0.7 as a temporary escape hatch while you migrate to new metrics. The metric is likely to be removed completely in an upcoming release.

Default: None

Cuda graph capture sizes, default is 512. 1. if one value is provided, then the capture list would follow the pattern: [1, 2, 4] + [i for i in range(8, CUDA_graph_sizes + 1, 8)] 2. more than one value (e.g. 1 2 128) is provided, then the capture list will follow the provided list.

Default: 512

If set to true and chunked prefill is enabled, we do not want to partially schedule a multimodal item. Only used in V1 This ensures that if a request has a mixed prompt (like text tokens TTTT followed by image tokens IIIIIIIIII) where only some image tokens can be scheduled (like TTTTIIIII, leaving IIIII), it will be scheduled as TTTT in one step and IIIIIIIIII in the next.

Default: False

If set to True, KV cache manager will allocate the same size of KV cache for all attention layers even if there are multiple type of attention layers like full attention and sliding window attention.

Default: False

If True, prefill requests can be chunked based on the remaining max_num_batched_tokens.

Default: None

For chunked prefill, a request is considered long if the prompt is longer than this number of tokens.

Default: 0

For chunked prefill, the maximum number of prompts longer than long_prefill_token_threshold that will be prefilled concurrently. Setting this less than max_num_partial_prefills will allow shorter prompts to jump the queue in front of longer prompts in some cases, improving latency.

Default: 1

Maximum number of tokens to be processed in a single iteration. This config has no static default. If left unspecified by the user, it will be set in EngineArgs.create_engine_config based on the usage context.

Default: None

For chunked prefill, the maximum number of sequences that can be partially prefilled concurrently.

Default: 1

Maximum number of sequences to be processed in a single iteration. This config has no static default. If left unspecified by the user, it will be set in EngineArgs.create_engine_config based on the usage context.

Default: None

If False, then multi-step will stream outputs at the end of all steps

Default: True

The number of slots to allocate per sequence per step, beyond the known token ids. This is used in speculative decoding to store KV activations of tokens which may or may not be accepted.

+ Default: 0

Maximum number of forward steps per scheduler call.

Default: 1

Whether to perform preemption by swapping or recomputation. If not specified, we determine the mode as follows: We use recomputation by default since it incurs lower overhead than swapping. However, when the sequence group has multiple sequences (e.g., beam search), recomputation is not currently supported. In such a case, we use swapping instead.

Default: None

The scheduler class to use. "vllm.core.scheduler.Scheduler" is the default scheduler. Can be a class directly or the path to a class of form "mod.custom_class".

Default: vllm.core.scheduler.Scheduler

Apply a delay (of delay factor multiplied by previous prompt latency) before scheduling next prompt.

Default: 0.0

The scheduling policy to use: * "fcfs" means first come first served, i.e. requests are handled in order of arrival. * "priority" means requests are handled based on given priority (lower value means earlier handling) and time of arrival deciding any ties).

Default: fcfs

Additional config for specified platform. Different platforms may support different configs. Make sure the configs are valid for the platform you are using. Contents must be hashable.

Default: {}

torch.compile configuration for the model. When it is a number (0, 1, 2, 3), it will be interpreted as the optimization level.

Following the convention of traditional compilers, using -O without space is also supported. -O3 is equivalent to -O 3. You can specify the full compilation config like so: {"level": 3, "CUDAgraph_capture_sizes": [1, 2, 4, 8]}. 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"}}' * --json-arg.key1 value1 --json-arg.key2.key3 value2

Default:

{
  "level": 0,
  "debug_dump_path": "",
  "cache_dir": "",
  "backend": "",
  "custom_ops": [],
  "splitting_ops": [],
  "use_inductor": true,
  "compile_sizes": null,
  "inductor_compile_config": {
    "enable_auto_functionalized_v2": false
  },
  "inductor_passes": {},
  "use_CUDAgraph": true,
  "CUDAgraph_num_of_warmups": 0,
  "CUDAgraph_capture_sizes": null,
  "CUDAgraph_copy_inputs": false,
  "full_CUDA_graph": false,
  "max_capture_size": null,
  "local_cache_dir": null
}

{
  "level": 0,
  "debug_dump_path": "",
  "cache_dir": "",
  "backend": "",
  "custom_ops": [],
  "splitting_ops": [],
  "use_inductor": true,
  "compile_sizes": null,
  "inductor_compile_config": {
    "enable_auto_functionalized_v2": false
  },
  "inductor_passes": {},
  "use_CUDAgraph": true,
  "CUDAgraph_num_of_warmups": 0,
  "CUDAgraph_capture_sizes": null,
  "CUDAgraph_copy_inputs": false,
  "full_CUDA_graph": false,
  "max_capture_size": null,
  "local_cache_dir": null
}

The configurations for event publishing. 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"}}' * --json-arg.key1 value1 --json-arg.key2.key3 value2

Default: None

The configurations for distributed KV cache transfer. 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"}}' * --json-arg.key1 value1 --json-arg.key2.key3 value2

Default: None