ariacompute-inference 1.6.5

Aria bundle load + session inference
Documentation
//! Aria inference: bundle loader, family registry, Session.

mod bundle;
mod chat;
mod family;
pub mod fixture;
pub mod multimodal;
mod pack;
mod profile;
mod session;
mod tensor_names;
mod tokenizer;

pub use aria_kernel::{ComputePref, EngineError};
pub use bundle::{dequantize, load_bundle, Bundle, ModelConfig, QuantTensor, TensorData};
pub use family::{
    arch_class_representatives, effective_rope_theta, family_phase, graph_hook, infer_family_path,
    lookup_family, require_runnable, require_stage_a, require_stage_b, ArchClass, Family,
    FamilyPhase, FAMILY_REGISTRY,
};
pub use multimodal::{action_head, asr_transcribe_pcm16le, rag_pack_context, vision_encode};
pub use profile::{EngineProfile, GenerateProfile, LoadProfile};
pub use session::{
    confidence_from_logits, GenerateOpts, Generation, Session, SessionBuilder,
};
pub use chat::{apply_chat_template, strip_assistant_visible, ChatTurn};
pub use tokenizer::{decode_placeholders, encode_naive, BundleTokenizer};