【QUESTION】How can I export the original nanochat model cards to an functional ONNX model like this?

#1
by HayatoHongo - opened

Hello, thank you very much for sharing this wonderful nanochat ONNX file.

I really appreciate your work — it is extremely helpful and inspiring.
My name is Hayato Hongo, and I am an LLM hobbyist currently experimenting with lightweight model deployment.

I would like to convert my nanochat model
(https://huggingface.co/HayatoHongo/sydney-filesystem-nanochat2/tree/main)
into an ONNX model so that I can run it in a lightweight runtime environment.

I attempted to export it myself, but encountered several issues (such as RMSNorm export errors), which I was unable to resolve.
I also checked this repository and searched the discussion threads, but could not find a conversion guide.

If possible, could you please let me know how to export the original .pt file to .onnx file like the one in this space?

I completely understand if you are busy — even access to an unpolished script or experimental code on GitHub would be incredibly helpful.

Thank you again for your great work. I truly appreciate your time!

Sign up or log in to comment