Frontend Machine Learning Libraries

Running ML in the browser is becoming standard. Here are the major players.

Library	Best For	Size	Complexity
Transformers.js	NLP, LLMs, Audio	Medium	Low (Easy API)
TensorFlow.js	Training, Custom Models	Large	High
ONNX Runtime Web	Production, Optimization	Small	Medium
MediaPipe	Vision, Pose, Face	Small	Low

Decision Matrix

• Need to run a HuggingFace model? -> Use Transformers.js.
• Need to train a model in the browser? -> Use TensorFlow.js.
• Need specific computer vision (Hand tracking)? -> Use MediaPipe.
• Need maximum speed for a custom model? -> Use ONNX Runtime + WebGPU.

Performance (WebGPU vs WASM)

Modern libraries use WebAssembly (WASM) for CPU execution and WebGPU for GPU execution. WebGPU is 10-100x faster for large matrix multiplications (like LLMs).

Next Steps

• Transformers.js Guide (Recommended start)
• TensorFlow.js Guide
• ONNX Guide