Skip to content

Releases

v0.1.1

Choose a tag to compare

@Juwan-Hwang Juwan-Hwang released this 18 Nov 05:57

Changes

  • ModelError now implements Show: a caught load error prints as InvalidModel: ... or Unsupported: ... (println("\{e}") works).
  • README: current size of the minified JS bundle (270 KB, 77 KB gzipped, training included).

Install

moon add Juwan-Hwang/moonboost@0.1.1

Prebuilt files

File What it is
moonboost.js ES module for browsers and Node (prediction for XGBoost JSON/UBJSON and LightGBM text models, plus training), minified with esbuild
moonboost.d.ts, moonbit.d.ts TypeScript declarations for moonboost.js
webgc.wasm + train_wasm.mjs WebAssembly GC training module and its loader (Chrome 119+, Firefox 120+, Safari 18.2+, Node 22+)
moonboost-linux-x86_64, moonboost-linux-arm64 CLI (info, predict, bench, train), glibc 2.34+
SHA256SUMS Checksums
import { load_model, predict } from "./moonboost.js";
const model = load_model(jsonText);
predict(model, [5.1, 3.5, 1.4, 0.2]);
./moonboost-linux-x86_64 predict model.json data.csv --header

moonboost 0.1.0

Choose a tag to compare

@Juwan-Hwang Juwan-Hwang released this 18 Nov 05:42

First release, also on mooncakes: moon add Juwan-Hwang/moonboost.

  • Loads XGBoost JSON and UBJSON models (gbtree, dart, gblinear) and LightGBM text models; predictions match the original libraries bit for bit on native, wasm, wasm-gc and js.
  • Tree-walking kernel checked with moon prove: terminates and never reads out of bounds for any model it accepts.
  • Branch-free batch and single-row prediction; faster than the XGBoost C API, TL2cgen, lleaves and onnxruntime-web in the benchmarks in the README.
  • Histogram tree training (reg:squarederror, binary:logistic, multi:softprob, multi:softmax) that writes XGBoost JSON; runs in browsers as JS or a 37 KB WebAssembly GC module.
  • ES module for browsers and Node, a CLI (info, predict, train) and a browser demo.