Crate kime 

The public crate. It re-exports the in process API from kime-engine and the request and answer types from kime-core, so a Rust program depends on this one crate and nothing else. See spec/07-engine.md and spec/14-sdks-cli.md.

use kime::{Device, Kime, Request, Response, State};
let kime = Kime::builder()
    .model("laya")                  // alias, local path or hf:// ref
    .device(Device::Auto)           // Cuda(0), Cpu { threads }, Auto
    .preload(true)
    .build()?;

let req = Request::new(State::json(&value))
    .choice("department", "Which team should handle this", [("billing", "Payment issues"), ("technical", "Bugs")])
    .score("urgency", "How urgent", ["not urgent", "soon", "critical"])
    .noul("churn", "The customer threatens to leave");

let res: Response = kime.decide(&req)?;          // blocking
let many = kime.decide_batch(&[req1, req2])?;    // batch
let res = kime.decide_async(&req).await?;        // async, any executor

Modules

answer

Answers, built from a question’s option logits and act logits the way Laya builds them.

confidence

The confidence formulas from spec/04-semantics.md.

hub

Finding a model by name, in the Hugging Face cache layout so downloads made by Laya or huggingface_hub are reused and ours are reused by them.

pyjson

Python’s json.dumps, byte for byte, for the values a request can hold.

render

The text the model reads for a question, before tokenizing.

request

The /v1/systemone request, parsed from JSON and validated the way spec/03-api.md describes.

round

Rounding a distribution for the wire, per spec/03-api.md.

Structs

Builder

Settings for Kime, from Kime::builder.

Decision

The future Kime::decide_async returns.

Kime

A loaded model on a device. Clones share it, and it can be used from any thread.

Request

A validated request.

Response

The answers to one request.

State

What a request asks about: text, or any JSON value, which is rendered the way the model family expects.

Enums

Answer

One question’s answer.

Device

Where the model runs.

Error

What can go wrong.

Precision

The number format on a GPU. The CPU always computes in FP32.