scenic icon indicating copy to clipboard operation
scenic copied to clipboard
verified publisher icon google-research

Nit: Specify units

Open stevietrouble opened this issue 3 years ago • 1 comments

Wasn't sure whether 77 meant words or characters. From the source, it looks like it's chars.

stevietrouble avatar Jun 07 '22 18:06 stevietrouble

CLIP uses the byte pair tokeniser implemented here: https://github.com/openai/CLIP/blob/main/clip/simple_tokenizer.py

The resulting tokens will not be characters.

AlexeyG avatar Jun 07 '22 18:06 AlexeyG