Show HN: Kiln – A WebGPU alternative to transformers.js, 2.2× faster (median) (github.com)

Kiln is a new JavaScript library that allows users to run AI models directly in the browser using WebGPU, achieving a median speed improvement of 2.2 times compared to transformers.js. It streamlines the process by reading original model weights from the Hugging Face Hub or a user’s own server, eliminating the need for ONNX exports or inference servers. Kiln supports various tasks, including text generation, image classification, and speech recognition, while maintaining a familiar API design akin to transformers.js, making migration for developers straightforward. This development is significant for the AI/ML community as it enables better utilization of local hardware, especially older devices that typically underutilize their resources. By allowing models to run directly in the browser, Kiln promotes privacy since model data never leaves the user's device. Additionally, its lightweight structure—just a 241 KB file with no external runtime dependencies—further encourages developers to incorporate AI capabilities into web applications easily. With the commitment to provide accurate outputs, Kiln checks all models against established frameworks, ensuring reliability while enhancing performance.