Module common 

Cross-model traits and helpers (Model-as-Code shared infrastructure).

Re-exports

pub use families::AudioBuffer;

pub use families::AudioTokens;

pub use families::EmbeddingModel;

pub use families::EncoderDecoderLM;

pub use families::EncoderState;

pub use families::ImageBuffer;

pub use families::MultimodalLLM;

pub use families::PcmSamples;

pub use families::SpeakerRef;

pub use families::Transcriber;

pub use families::TranscriptSegment;

pub use families::TtsModel;

pub use families::VisualTokens;

pub use llm::DecoderOnlyLLM;

pub use llm::LlmRuntimeConfig;

Modules

decoder_unified

Shared helpers for decoder-only unified mixed-batch forward.

families

Model-family traits beyond DecoderOnlyLLM.

llm

DecoderOnlyLLM trait — the “model family” interface that every decoder-only language model (Qwen3 / Llama / Mistral / DeepSeek / …) implements, independent of backend and weight format.

paged_pool

Multi-sequence paged-KV pool — Phase 4 of Metal paged attention.