Feature request

Hi! First of all, thank you for this amazing library.

I'd like to request support for a browser-compatible version (ONNX) of one of the following models:

• Salesforce/codet5p-770m
• microsoft/codebert-base
• bigcode/starcoderbase

Use case: I’m building an educational app that runs entirely in the browser (usingtransformers.js) and analyzes Python code written by students. Currently, only sentiment/text models are available, but it would be incredibly useful to have a model specifically trained on code.

If you could convert and publish one of these models under theXenova/ namespace, that would unlock huge potential for offline/secure code understanding in EdTech and frontend AI.

Thanks again for all your work!

Motivation

I’m building an educational platform that runs entirely in the browser using transformers.js and Pyodide. My goal is to provide students with instant feedback on Python code — without relying on external APIs or requiring backend servers.

Currently, only general-purpose NLP models (like sentiment classifiers) are available in ONNX format for transformers.js. But for code-related tasks (e.g., checking function structure, syntax errors, or task relevance), having a model like CodeT5+ or CodeBERT available in the Xenova namespace would be extremely valuable.

This would enable real-time, privacy-respecting code analysis directly in the browser — ideal for EdTech, offline environments, and secure classroom tools.

Your contribution

I’d be happy to help test the model in a browser-based educational setting once it's available. I can provide feedback on performance, latency, and educational usefulness. I’m also open to contributing documentation or example integrations using CodeMirror, Pyodide, and transformers.js in the browser.

While I’m not familiar with ONNX conversion myself, I’d be glad to help validate the integration in real-world use cases.