Read .npy arrays saved with NumPy directly in modern JavaScript runtimes.
npm install npyjs # or yarn add npyjsSupports Node ≥18, modern browsers, and Deno/Bun.
// Modern named export (recommended) import { load } from "npyjs"; // Back-compatibility class (matches legacy docs/tests) import npyjs from "npyjs";import { load } from "npyjs"; const arr = await load("my-array.npy"); // arr has { data, shape, dtype, fortranOrder } console.log(arr.shape); // e.g., [100, 784]import npyjs from "npyjs"; // Default options const n = new npyjs(); // Disable float16→float32 conversion const n2 = new npyjs({ convertFloat16: false }); const arr = await n.load("my-array.npy");npyjs returns flat typed arrays with a shape. npyjs also ships a small helper to turn the flat data + shape into nested JS arrays.
import { load } from "npyjs"; import { reshape } from "npyjs/reshape"; const { data, shape, fortranOrder } = await load("my-array.npy"); const nested = reshape(data, shape, fortranOrder); // -> arrays nested by dimsFor C-order arrays (the NumPy default), pass fortranOrder = false (default).
For Fortran-order arrays, pass true and the helper will return the natural row-major nested structure.
Or pair it with ndarray or TensorFlow.js:
import ndarray from "ndarray"; import { load } from "npyjs"; const { data, shape } = await load("my-array.npy"); const tensor = ndarray(data, shape); console.log(tensor.get(10, 15));int8,uint8int16,uint16int32,uint32int64,uint64(asBigInt)float32float64float16(converted to float32 by default)
// Default: converts float16 → float32 const n1 = new npyjs(); // Keep raw Uint16Array const n2 = new npyjs({ convertFloat16: false });npm run build # Build to dist/ npm test # Run Vitest npm run typecheck # TypeScript type checkingApache-2.0 © JHU APL
Made with ♥ at 