How do Float16Array typed arrays optimize machine learning inference in browsers?

// Compact ML model storage
const weights = new Float16Array(1048576); // 2MB vs 4MB Float32
const input = new Float32Array(512);

// WebGPU bind group with Float16 storage buffer
const weightBuffer = device.createBuffer({
  size: weights.byteLength,
  usage: GPUBufferUsage.STORAGE,
  mappedAtCreation: true
});
new Uint8Array(weightBuffer.getMappedRange()).set(weights.buffer);
weightBuffer.unmap();