Module engine 

Inference engine interfaces — split per modality.

Phase 5a step 2 splits the historical mega-trait (which mixed LLM generation, embedding, transcription, and TTS in one) into a base lifecycle trait and four modality-specific supertraits. Each engine impl now implements exactly the trait its modality needs; no more inert “unsupported” stubs.

Traits

AdvancedInferenceEngine

Advanced engine capabilities — opt-in addition to LLM engines that support batching / speculation / runtime reconfig / diagnostics.

EmbedEngine

Embedding engine (CLIP, BERT, etc.).

InferenceEngine

Lifecycle / status methods shared by every engine kind.

LlmInferenceEngine

LLM text-generation engine.

TranscribeEngine

Speech-to-text (Whisper) engine.

TtsEngine

Text-to-speech (Qwen3-TTS, etc.) engine.

Type Aliases

HardwareConstraints

Hardware constraints alias.

LatencyRequirements

Latency requirements alias.

RequestCharacteristics

Request characteristics alias.

SpeculationConfig

Speculation configuration for speculative decoding.