Building models with tf.text (TF World '19)

preview_player
Показать описание

Presented by: Robby Neale

  1. TensorFlow
  2. building models with tf.text
  3. tf.text
  4. tf.data
  5. RaggedTensors
  6. text-based ops
Рекомендации по теме
Building models with 0:39:33

Building models with tf.text (TF World '19)

TensorFlow Hub: Making 0:09:03

TensorFlow Hub: Making model discovery easy (TF Dev Summit '20)

Building a Regression 0:09:37

Building a Regression model using Keras - Pt-11 | Machine Learning using TF

Building the custom 0:06:16

Building the custom spam detection (text classification) TF lite model

Natural Language Processing|TF-IDF 0:08:27

Natural Language Processing|TF-IDF Intuition| Text Prerocessing

Tensorflow Input Pipeline 0:33:19

Tensorflow Input Pipeline | tf Dataset | Deep Learning Tutorial 44 (Tensorflow, Keras & Python)

Responsible AI with 0:21:52

Responsible AI with TensorFlow (TF Dev Summit '20)

Training and deploying 0:11:41

Training and deploying ML models on edge devices (TF Fall 2020 Updates)

TF-Ranking with TFR-BERT 0:12:16

TF-Ranking with TFR-BERT

Performant, scalable models 0:40:35

Performant, scalable models in TensorFlow 2 with tf.data, tf.function & tf.distribute (TF World ...

How to Build 0:13:46

How to Build a Neural Network with TensorFlow and Keras in 10 Minutes

Tutorial 1-Transformer And 0:24:30

Tutorial 1-Transformer And Bert Implementation With Huggingface

Learning to read 0:15:36

Learning to read with TensorFlow and Keras (TF Dev Summit '20)

TF Data and 2:08:34

TF Data and Deployment - Advanced Deployment Scenarios with TensorFlow Part 1/2

Understanding TF-IDF Model 0:16:45

Understanding TF-IDF Model - Hands On NLP using Python Demo

LEGO, but AI 0:07:22

LEGO, but AI controls what I BUILD...

Building a Visual 0:05:19

Building a Visual Debugging Tool for ML - TF.js in Interactive Visual Analytics (TF Dev Summit &apos...

Build Natural Language 0:35:10

Build Natural Language Processing (NLP) model with Python| TF-IDF, N-gram, Text processing

tf serving tutorial 0:19:52

tf serving tutorial | tensorflow serving tutorial | Deep Learning Tutorial 48 (Tensorflow, Python)

Leverage TF Lite 0:25:23

Leverage TF Lite Pre-trained Models

Inside TensorFlow: tf.Keras 0:53:37

Inside TensorFlow: tf.Keras (Part 1)

Building a ranking 0:05:20

Building a ranking model with TF Recommenders (Building recommendation systems with TensorFlow)

TensorFlow Dev Summit 0:31:12

TensorFlow Dev Summit 2020 Keynote

Text Representation Using 0:36:18

Text Representation Using TF-IDF: NLP Tutorial For Beginners - S2 E6

Комментарии
Автор

Throughout the BERT GitHub repo, almost a lot of people wanted to have this within TF as one of the API. Finally is here. I was one of the first people to write a preprocessor to do fine-tuning from BERT and raised this question at first. And, also posted a pr on the Github issue request.
Glad to see this happening. It's been a year that Google Research gave birth to BERT.

jugsma
Автор

This is something we waited really long.

aiwithr
Автор

Listen to around 30 minute 45 second mark (near the end of the main presentation). Hol up ... Wait a min. So for sequence data (in addition to text data) that is non-text, e.g., numeric time series, tensorflow_text is the right thing to use because it contains the ragged tensors? Then why name it tensorflow_text? Did you think this through? This is a lot like the keras attention layer which although useful for sequences generally in concept, Google apparently wrote and documented as if it's only for text sequences and encoder-decoder text models. Google product managers need to consider to generalize their thinking -- and software tools -- a little more.

geoffreyanderson
Автор

Are the preprocessing functions such as bert_preprocess() built-in and done automatically, or do we still write those? Wasn't sure if you were showing the code for what happens behind the scenes or if they were examples of what an engineer might write.

jeffreydeason