flan-t5 support about fastt5

[Ki6an]

can you please provide the reproducible code and output you are getting?

[Ki6an]

does it work with python ort or are you facing the issue only with the js version of ort?

[loretoparisi]

facing

I'm facing this error in transformer-js that is using ONNX converted model here

[loretoparisi]

can you please provide the reproducible code and output you are getting?

Yes, I will fork the original repo and apply the changes.

[loretoparisi]

@Ki6an here my fork where you can try it

This wil install the app and convert flan-t5-small to onnx:

git clone https://github.com/loretoparisi/transformers-js.git
cd transformers-js/
pip install transformers
python tools/convert_model.py 

You will find then the quantized models in the /models folder.

to run it

make demo
make run

now points to http://localhost:8152/?model_id=google/flan-t5-small

The tokenizer code is located here.