how to change model

[githubyounik]

Hello, I read your paper and I think it is a very good work.
I noticed that “deberta-v3-large” and “en_core_web_sm” are automatically downloaded in huggingface when using the library, and I was wondering if I can replace them with other models, or if I can use local models, and where should I change the model names or paths? I want to know if I can replace them with other models, or if I can use local models, and where should I change the model name or path?

[g185]

Hi @githubyounik,

I you want to train maverick models with a different encoder, you can do it by simply creating a new model configuration under conf/model with a specific name, and indicating it in the root.yaml file.
As you can see there is already some example for deberta-v3-base and other encoders, such as longformer.
Concretely, you can changehuggingface_model_name to link a configuration to a new encoder in huggingface.

After completing this step, to train a new system with a different encoder you can indicate it in conf/root.yaml model entry using the model configuration path.
E.G. to train a system with deberta-v3-base it is enough to change model: mes/deberta-large to model: mes/deberta-base.

If you want to use another model in inference instead, you can either select among our available models, or train one using the steps above and linking its checkpoint path as presented here.

Let me know if you have any other questions,
Giuliano