What Is ONNX? (And Why Transformers.js Uses It)

Short summary

ONNX is a portable, standardized format for storing ML model structure and weights independently of the execution environment. ONNX Runtime handles the execution layer, supporting multiple backends including WebGPU for GPU acceleration in browsers, WASM for lightweight cross-platform execution, and native code. Transformers.js leverages this separation of concerns to provide a single, unified JavaScript API for running the same model across any platform: browsers, servers, and edge devices.

• •ONNX decouples model storage (graphs and weights) from execution runtime
• •ONNX Runtime abstracts backends: WebGPU (browser), WASM (lightweight), native (performance)
• •Transformers.js provides one JavaScript API for cross-device inference using ONNX

Generated with AI, which can make mistakes.