Transformers.js in 30 seconds #MachineLearning #AI #WebAI

Key Takeaways

Transformers JS brings state-of-the-art machine learning to JavaScript. Let's see what that means in 30 seconds. For inference, it uses ONNX, a standard that stores the model as a computation graph plus trained weights in a binary file format. It then uses ONNX runtime to run the calculations in the execution provider you choose. But before that happens, Transformers JS figures out what files are needed, downloads and caches them, then creates the ONNX inference session, and returns it as a pipe. Once you run the pipe, it converts your input into tensors in the shape the model expects, runs inference, and converts the output tensors into the format you need.