vLLM Engine Schema

Engine version: 0.7.3
Discovered at: 2026-05-06T22:57:22+02:00
Discovery method: dataclasses.fields(EngineArgs) + msgspec.json.schema(SamplingParams)
Schema version: 1.0.0

Summary: 104 engine parameters, 31 sampling parameters.

Discovery limitations

• sampling_params - constraints (e.g. temperature>=0, top_p in (0,1]) live in imperative _verify_args() and are not introspectable from field metadata
• engine_params - per-field descriptions unavailable (vLLM EngineArgs has only a class docstring)

Engine Parameters

Field	Type	Default	Description
model	str	facebook/opt-125m
served_model_name	`str	list[str]	None`
tokenizer	`str	None`	-
task	Literal['auto', 'generate', 'embedding', 'embed', 'classify', 'score', 'reward', 'transcription']	auto
skip_tokenizer_init	bool	false
tokenizer_mode	str	auto
trust_remote_code	bool	false
allowed_local_media_path	str	""
download_dir	`str	None`	-
load_format	str	auto
config_format	ConfigFormat	auto
dtype	str	auto
kv_cache_dtype	str	auto
seed	int	0
max_model_len	`int	None`	-
distributed_executor_backend	`str	type[ExecutorBase]	None`
pipeline_parallel_size	int	1
tensor_parallel_size	int	1
max_parallel_loading_workers	`int	None`	-
block_size	`int	None`	-
enable_prefix_caching	`bool	None`	-
disable_sliding_window	bool	false
use_v2_block_manager	bool	true
swap_space	float	4
cpu_offload_gb	float	0
gpu_memory_utilization	float	0.9
max_num_batched_tokens	`int	None`	-
max_num_partial_prefills	`int	None`	1
max_long_partial_prefills	`int	None`	1
long_prefill_token_threshold	`int	None`	0
max_num_seqs	`int	None`	-
max_logprobs	int	20
disable_log_stats	bool	false
revision	`str	None`	-
code_revision	`str	None`	-
rope_scaling	`dict[str, Any]	None`	-
rope_theta	`float	None`	-
hf_overrides	`dict[str, Any]	Callable[[<class 'transformers.configuration_utils.PretrainedConfig'>], PretrainedConfig]	None`
tokenizer_revision	`str	None`	-
quantization	`str	None`	-
enforce_eager	`bool	None`	-
max_seq_len_to_capture	int	8192
disable_custom_all_reduce	bool	false
tokenizer_pool_size	int	0
tokenizer_pool_type	`str	type[ForwardRef('BaseTokenizerGroup')]`	ray
tokenizer_pool_extra_config	`dict[str, Any]	None`	-
limit_mm_per_prompt	`Mapping[str, int]	None`	-
mm_processor_kwargs	`dict[str, Any]	None`	-
disable_mm_preprocessor_cache	bool	false
enable_lora	bool	false
enable_lora_bias	bool	false
max_loras	int	1
max_lora_rank	int	16
enable_prompt_adapter	bool	false
max_prompt_adapters	int	1
max_prompt_adapter_token	int	0
fully_sharded_loras	bool	false
lora_extra_vocab_size	int	256
long_lora_scaling_factors	`tuple[float]	None`	-
lora_dtype	`str	dtype	None`
max_cpu_loras	`int	None`	-
device	str	auto
num_scheduler_steps	int	1
multi_step_stream_outputs	bool	true
ray_workers_use_nsight	bool	false
num_gpu_blocks_override	`int	None`	-
num_lookahead_slots	int	0
model_loader_extra_config	`dict	None`	-
ignore_patterns	`str	list[str]	None`
preemption_mode	`str	None`	-
scheduler_delay_factor	float	0.0
enable_chunked_prefill	`bool	None`	-
guided_decoding_backend	str	xgrammar
logits_processor_pattern	`str	None`	-
speculative_model	`str	None`	-
speculative_model_quantization	`str	None`	-
speculative_draft_tensor_parallel_size	`int	None`	-
num_speculative_tokens	`int	None`	-
speculative_disable_mqa_scorer	`bool	None`	false
speculative_max_model_len	`int	None`	-
speculative_disable_by_batch_size	`int	None`	-
ngram_prompt_lookup_max	`int	None`	-
ngram_prompt_lookup_min	`int	None`	-
spec_decoding_acceptance_method	str	rejection_sampler
typical_acceptance_sampler_posterior_threshold	`float	None`	-
typical_acceptance_sampler_posterior_alpha	`float	None`	-
qlora_adapter_name_or_path	`str	None`	-
disable_logprobs_during_spec_decoding	`bool	None`	-
otlp_traces_endpoint	`str	None`	-
collect_detailed_traces	`str	None`	-
disable_async_output_proc	bool	false
scheduling_policy	Literal['fcfs', 'priority']	fcfs
scheduler_cls	`str	type[object]`	vllm.core.scheduler.Scheduler
override_neuron_config	`dict[str, Any]	None`	-
override_pooler_config	`PoolerConfig	None`	-
compilation_config	`CompilationConfig	None`	-
worker_cls	str	auto
kv_transfer_config	`KVTransferConfig	None`	-
generation_config	`str	None`	-
override_generation_config	`dict[str, Any]	None`	-
enable_sleep_mode	bool	false
model_impl	str	auto
calculate_kv_scales	`bool	None`	-
additional_config	`dict[str, Any]	None`	-

Sampling Parameters

Field	Type	Default	Description
n	integer	1
best_of	unknown	-
_real_n	unknown	-
presence_penalty	number	0.0
frequency_penalty	number	0.0
repetition_penalty	number	1.0
temperature	number	1.0
top_p	number	1.0
top_k	integer	-1
min_p	number	0.0
seed	unknown	-
stop	unknown	-
stop_token_ids	unknown	-
bad_words	unknown	-
ignore_eos	boolean	false
max_tokens	unknown	16
min_tokens	integer	0
logprobs	unknown	-
prompt_logprobs	unknown	-
detokenize	boolean	true
skip_special_tokens	boolean	true
spaces_between_special_tokens	boolean	true
logits_processors	unknown	-
include_stop_str_in_output	boolean	false
truncate_prompt_tokens	unknown	-
output_kind	unknown	0
output_text_buffer_length	integer	0
_all_stop_token_ids	array	[]
guided_decoding	unknown	-
logit_bias	unknown	-
allowed_token_ids	unknown	-