Module web-engine

TimesFM Web Inference Engine — browser-compatible ONNX inference via onnxruntime-web (WASM / WebGPU / WebGL).

Implements the IInferenceEngine interface from @agentix-e/timesfm-core so it can be injected into TimesFMModel.fromPretrained().

Execution Providers

The engine tries providers in order: webgpu → wasm → webgl, falling back to the next available one on failure.

Model Loading

load() accepts:

• A URL string (fetched via fetch())
• An ArrayBuffer (pre-loaded model data)

Usage

import { TimesFMModel, createForecastConfig } from '@agentix-e/timesfm-core';
import { TimesFMWebInferenceEngine } from '@agentix-e/timesfm-web';

const engine = new TimesFMWebInferenceEngine(config);
await engine.load('/models/timesfm-2.5.onnx');

const model = await TimesFMModel.fromPretrained({
  modelPath: '/models/timesfm-2.5.onnx',
  engine,
});
model.compile(createForecastConfig({ maxContext: 512, maxHorizon: 128 }));
const result = await model.forecast(24, [inputData]);
Copy