Hi TF Users,
I am delighted to announce, I am releasing Tensorflow-Transformers version 2.0 soon. It comes with lot of speedup and designed specifically for Tensorflow users, who are interested in using SOTA Transformer models in NLP, CV, AUDIO etc.
Text-Generation is 90 times faster comparing to Huggingface TF implementation and other equivalent TF implementation in most cases. Designed with complete serialisation in mind, it offers a lot of other features.
We have custom tokenizers using tf-text, end to end serialization pipeline and many more.
It supports GPU, TPU, mixed precision, TFlite by default.
Currently supporting 12 major models Bert, Albert, Roberta, CLIP, VIT, T5, GPT2, mT5, BART and SentenceTransformers etc.
Have a look and share your feedback, open issues . Thanks.