javascript inference

[deepcode-debug]

Thanks a lot for the contribution, really amazing library.
Is there anyway to use it directly in javascript? I am not sure if the onnx file could be directly importable

[codelion]

JavaScript inference is not directly supported, but it's possible to wire it up manually.

The two pieces you need:

• Use @xenova/transformers which ships ONNX versions of the underlying sentence-transformer models (e.g. all-MiniLM-L6-v2) and handles tokenization out of the box. For Node.js you can also use onnxruntime-node directly.

• After training, export the learned prototypes (class centroids) to JSON from Python, load them in JS, and compute cosine similarity against them to get predictions.

The classifier logic itself is just vector math so it translates cleanly. The main work is the one-time export step on the Python side. Happy to accept a PR if someone wants to add a export_for_js() helper and a reference inference.js example.