You can add location information to your Tweets, such as your city or precise location, from the web and via third-party applications. You always have the option to delete your Tweet location history. Learn more
New example on http://keras.io : a neural machine translation example with a seq2seq Transformer. Starts from raw data, and includes data preparation, building the Transformer (from scratch!) & training it, and inference. Less than 200 lines total.https://keras.io/examples/nlp/neural_machine_translation_with_transformer/ …
I *think* this is the most concise & readable seq2seq Transformer you'll find anywhere (that doesn't involve a pre-made model).
Dont you think using a pre made model with transfer learning is much better for the environment?
Thanks for sharing.
Does it support caching to make inference faster?
Hi Francis, just wanted to know, why you think @huggingface implementations for model.generate is way too slower than @PyTorch . I am an active user of @TensorFlow , but no one has actually addressed it. Even the implementation you had doesn't support caching If I am not wrong.
Hey! Tensorflow maintainer at Hugging Face here - that's not @fchollet's fault, it's mine! Can you tell me what model you were using and what the speeds were like with TF and PT so I can try to reproduce the issue and file a ticket?
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.
