Combining VOCABS for training and inference

[mohamedsaeed8223]

I was wondering if this is feasible at all or not
I tried creating a new vocab the same way the multilingual vocab was written
here's a snippet:

VOCABS["englisharabic"] = "".join(
    dict.fromkeys(
        VOCABS["english"]
        + VOCABS["arabic"]
        + "§"
    )
)

I added this in the doctr/datasets/vocabs.py in my cloned repo
and tried to train a model using it but the train script didn't recognize it
I was wondering if I should add my new vocab to the package code in my env instead or if it isn't doable at all

[mohamedsaeed8223]

update: I updated the vocabs.py in the path miniconda3/envs/doctr/lib/python3.10/site-packages/doctr/datasets with my new vocab and the training initiated 😅

[felixdittrich92]

Hi @mohamedsaeed8223 👋,

You can install the repo as editable then you can modify the vocab directly in you IDE

cd doctr
pip3 install -e .[torch]  # for example