Understanding and modifying config files for different architecture implemenations

[shrinidhin]

Hi!I have been using spacy for an entity based sentiment analysis project and have been loving it!I am still exploring the guides and documentations along with my implementations and have received a lot of help from the spacy team on github here!Thank you!

I have been wondering how to write the config or rather what to include in the config file, If I were to use different model architectures for my custom trained spacy pipelines. Taking the Healthsea project as an example, For the Entity recognition bit after training on simple tok2vec embedding, If I want to use a transformer architecture and evaluate the results, I am not able to understand what to include in the config. Though the github repo has the config files available. Is it required for any project to try out different model architectures? How do I understand the different architectures to make the relevant changes in the config files?

[thomashacker]

Hello,
If you want to use transformer embeddings for your components you can use the quickstart widget in our docs to get the pipeline configuration you want and then use spacy init config (https://spacy.io/api/cli#init) to auto-fill all default values.