sokuto_runtime/engine/
mod.rs

1//! # 推論エンジンモジュール
2//!
3//! ONNX Runtime を用いた低遅延フォワードパス、セッション管理、
4//! および Execution Provider 解決機構を提供する。
5
6pub mod batch;
7pub mod cache;
8pub mod coarse;
9pub mod config;
10pub mod escalation;
11pub mod gating;
12pub mod provider;
13pub mod session;
14
15pub use batch::{BatchScratchpad, DEFAULT_MAX_BATCH_CHUNK_SIZE};
16pub use cache::{
17    AlignedBuffer64, CacheCompressionMode, CachedStateFeatures, CachedTensorPayload,
18    DEFAULT_DUAL_MAX_BATCH_CHUNK_SIZE, DualSessionEngine, StateCacheManager, is_aligned_64,
19};
20pub use coarse::{
21    CandidateEmbeddingCache, CoarseScorer, CoarseToFineConfig, DEFAULT_COARSE_THRESHOLD,
22    DEFAULT_NEGATIVE_KEYS, DEFAULT_TOP_M_CANDIDATES, EmbeddingCoarseScorer, FilteredCandidates,
23    HierarchicalCoarseScorer, LexicalCoarseScorer, ModelDrivenCoarseScorer, filter_top_candidates,
24    is_negative_candidate, reconstruct_probabilities,
25};
26pub use config::{ExecutionProvider, OptimizationLevel, SessionConfig};
27pub use escalation::{
28    EscalationPromptBuilder, EscalationTemplateConfig, build_rich_escalation_prompt,
29};
30pub use gating::{
31    apply_gating_to_answer, apply_gating_to_answer_with_energy, apply_gating_to_answer_with_logits,
32    apply_gating_to_answers, apply_gating_to_answers_with_logits, calculate_choice_energy,
33    resolve_gating_config,
34};
35pub use provider::register_execution_providers;
36pub use session::InferenceEngine;
sokuto_runtime/engine/mod.rs

sokuto_runtime/engine/
mod.rs
