Hello, have you ever considered using wasm technology to implement this library? For example, rust's wgpu-rs and c++'s dawn are both implementations of webgpu. They can be converted to wasm and can also be accelerated with simd.
This library is based on ONNX Runtime for Web, which in turn uses WebAssembly (wasm). Do you mean something else or more wasm for slow functions in Transformers.js?
Question
Hello, have you ever considered using wasm technology to implement this library? For example, rust's wgpu-rs and c++'s dawn are both implementations of webgpu. They can be converted to wasm and can also be accelerated with simd.