This visualization presents the relationships between the input token embeddings in various text embedding models. The input token embeddings are projected into a 2D space using TSNE. For more details, please refer to the article about Word Injection fine-tuning..
The search box supports regular expressions. It is possible to find multiple tokens at a time.
^[0-9]+$ - will find all tokens that contain only digits
^the$ - will find the token "the"
the - will find all tokens that contain the substring "the"
^the$|^a$|^an$ - will find the tokens "the", "a", "an"
^## - will find all tokens that start with "##" (subword tokens in WordPiece)