Specifically, I am writing an academic paper and I cannot find any coherent source on which it is clearly presented which type of embedding models are used within the pipelines. I understand that tok2vec is the model used to embed the words, but what is the architecture behind? Does it implements word2vec? GloVe? An encoder?
If anyone know any source where it is clearly stated which type of architecture it uses it would be extremely helpful. I am desperate

I went on the website and looker for a solution but it only present the different ways you can build your model and does not explains how are built theirs. Specifically what I am looking to understand is the architecture behind it_core_news_lg model.