aisimulate-core 0.12.0

Engine-neutral inference simulation, deterministic replay, and performance modeling
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43

// SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
// SPDX-License-Identifier: Apache-2.0

//! Runtime-neutral mock inference schedulers and attention-DP composition.

mod cache;
mod common;
mod config;
pub mod generalized;
mod handoff;
mod host_offload;
mod kv_manager;
mod protocol;
mod runtime;
mod scheduler;
mod timing;
mod trace;

pub(crate) use host_offload::{
    HostBlockKey, HostOffloadObservation, HostOffloadObservationData, HostOffloadObserver,
};

pub use common::running_mean::RunningMean;
pub use common::speculative::normalize_conditional_accept_rates;
pub use config::{
    Backend, EngineConfig, NativeHostOffloadConfig, PreemptionMode, SglangConfig,
    SglangSchedulePolicy, TrtllmCapacityPolicy, TrtllmConfig, WorkerType,
};
pub use handoff::{HandoffId, HandoffTransferTiming, TransferTimingMode, prefill_handoff_delay_ms};
pub use protocol::{
    Admission, CacheTierAttribution, Command, CommandEffects, CommandResult, ForwardPassMetrics,
    KvBlock, KvEvent, KvEventData, LifecycleEvent, Metrics, Output, PassCompletionEffects,
    PassStartEffects, PressureEvent, PressureKind, PressureState, Request, StoredBlocks,
};
pub use runtime::{Engine, EngineFactory};
pub use scheduler::SchedulerRank;
pub use timing::{TimingModel, TimingModelConfig};

#[doc(hidden)]
pub use protocol::PendingPass;
pub(in crate::engine) use timing::modeled_duration_ms;
#[doc(hidden)]
pub use trace::g1_parent_chain_events;