How to use tokenizer.json in ExecuTorch Android demo (without tokenizer.model)?

[jordanqi]

📚 The doc issue

I'm trying to deploy a language or vision-language model on Android using the ExecuTorch Android demo app.
The model I'm working with only provides tokenizer.json, but the current Android implementation appears to expect a tokenizer.model file instead.

Is tokenizer.model mandatory for the ExecuTorch demo app?

If I only have a tokenizer.json file (from HuggingFace), is there any recommended way to convert or load it in the app?

Suggest a potential alternative/fix

No response

cc @kirklandsign @cbilgin

[kirklandsign]

The app can now recognize .json file

https://github.com/pytorch/executorch/blob/main/examples/demo-apps/android/LlamaDemo/app/src/main/java/com/example/executorchllamademo/SettingsActivity.java#L387-L389

[kirklandsign]

BTW it really depends on your tokenizer type. Please let us know if you see an error. Right now the supported tokenizers are her: https://github.com/pytorch-labs/tokenizers

[jordanqi]

BTW it really depends on your tokenizer type. Please let us know if you see an error. Right now the supported tokenizers are her: https://github.com/pytorch-labs/tokenizers

I’m trying to deploy a VITA-1.5 multimodal model (supports audio, vision, and text) using ExecuTorch.

The tokenizer is in Hugging Face tokenizer.json format, and I’d like to ask:
1. Is there any suggested way to convert the model into .pte format for ExecuTorch?
2. Since this is a new architecture, do you have any guidance or examples for adding custom models?
3. Can I still use the LlamaDemo Android app with this multimodal？

[kirklandsign]

The tokenizer is in Hugging Face tokenizer.json format

This should be supported already.

Is there any suggested way to convert the model into .pte format for ExecuTorch?

Mind opening a new issue?

Since this is a new architecture, do you have any guidance or examples for adding custom models?

Same as above.

Can I still use the LlamaDemo Android app with this multimodal？

Most likely yes. We can run some HF models out of box.