Struct EngineConfig 

pub struct EngineConfig {

    pub backend: Backend,
    pub num_gpu_blocks: usize,
    pub block_size: usize,
    pub max_model_len: Option<usize>,
    pub max_num_seqs: usize,
    pub max_num_batched_tokens: usize,
    pub enable_prefix_caching: bool,
    pub enable_chunked_prefill: bool,
    pub speedup_ratio: f64,
    pub decode_speedup_ratio: f64,
    pub aic_nextn: Option<usize>,
    pub aic_nextn_accept_rates: Option<String>,
    pub aic_mtp_seed: u64,
    pub worker_type: WorkerType,
    pub preemption_mode: PreemptionMode,
    pub emit_kv_events: bool,
    pub emit_kv_token_ids: bool,
    pub kv_transfer_bytes_per_token: Option<usize>,
    pub kv_cache_bytes_per_token: Option<usize>,
    pub native_host_offload: Option<NativeHostOffloadConfig>,
    pub kv_transfer_bandwidth: Option<f64>,
    pub kv_transfer_timing_mode: TransferTimingMode,
    pub timing_model: TimingModelConfig,
    pub sglang: SglangConfig,
    pub trtllm: TrtllmConfig,
}

Serializable configuration for one scheduler rank.

Attention-DP size and worker identity belong to crate::engine::generalized::GeneralizedEngineConfig and crate::engine::generalized::EngineIdentity, not this rank-local configuration.

Default constructs a vLLM configuration. Changing only Self::backend afterward does not recompute backend-dependent fields such as Self::block_size; start with Self::for_backend when constructing a different backend in Rust. Deserialization selects the backend’s block-size default when block_size is omitted.

Fields

backend: Backend

Scheduler backend whose semantics this rank executes.

Use Self::for_backend instead of changing this field on Self::default when backend-dependent defaults are desired.

num_gpu_blocks: usize

Physical G1 capacity in blocks.

block_size: usize

KV block size in tokens.

max_model_len: Option<usize>

Optional model context limit.

max_num_seqs: usize

Maximum concurrently runnable sequences.

max_num_batched_tokens: usize

Per-pass token budget.

enable_prefix_caching: bool

Whether complete blocks remain reusable after request release.

enable_chunked_prefill: bool

Whether a prompt may be split across scheduler passes.

speedup_ratio: f64

Divisor applied to modeled prefill and decode latency.

decode_speedup_ratio: f64

Additional divisor applied to decode latency.

aic_nextn: Option<usize>

MTP/EAGLE draft-token count. One verification forward can emit up to aic_nextn + 1 output tokens.

aic_nextn_accept_rates: Option<String>

Conditional draft acceptance rates, comma-separated.

Entry i is the probability that draft i is accepted given that every preceding draft was accepted.

aic_mtp_seed: u64

Base seed for deterministic worker-local MTP acceptance sampling.

worker_type: WorkerType

Scheduler role.

preemption_mode: PreemptionMode

Decode preemption victim order.

emit_kv_events: bool

Retain and expose local token-block hashes in neutral KV events.

emit_kv_token_ids: bool

Retain block token IDs alongside neutral KV events.

kv_transfer_bytes_per_token: Option<usize>

Bytes transferred per prompt token for disaggregated handoff timing.

kv_cache_bytes_per_token: Option<usize>

Physical KV-cache bytes occupied by one token for host offload.

native_host_offload: Option<NativeHostOffloadConfig>

Optional framework-native host-offload simulation.

kv_transfer_bandwidth: Option<f64>

Modeled prefill-to-decode transfer bandwidth in decimal GB/s.

kv_transfer_timing_mode: TransferTimingMode

Prompt footprint used to model disaggregated transfer time.

timing_model: TimingModelConfig

Serializable timing-provider descriptor.

sglang: SglangConfig

SGLang-only scheduler controls.

trtllm: TrtllmConfig

TensorRT-LLM-only scheduler controls.

Implementations

impl EngineConfig

pub fn for_backend(backend: Backend) -> Self

Construct a configuration with the selected backend’s native defaults.

In particular, this selects Backend::default_block_size instead of inheriting the vLLM block size from Self::default.

Trait Implementations

impl Clone for EngineConfig

fn clone(&self) -> EngineConfig

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for EngineConfig

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for EngineConfig

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for EngineConfig

fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>

where D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl PartialEq for EngineConfig

fn eq(&self, other: &EngineConfig) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl Serialize for EngineConfig

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.

impl StructuralPartialEq for EngineConfig