me vs. validator round 6

PiperOrigin-RevId: 312082138

tfhub_dev/assets/digitalepidemiologylab/digitalepidemiologylab.md

@@ -0,0 +1,7 @@
+# Publisher digitalepidemiologylab
+Digital Epidemiology Lab at EPFL, Switzerland
+
+## Digital Epidemiology Lab
+Research lab at EPFL working on Applied Machine Learning for Epidemiology.
+
+https://www.digitalepidemiologylab.org/

tfhub_dev/assets/digitalepidemiologylab/models/covid-twitter-bert/1.md

@@ -0,0 +1,64 @@
+# Module digitalepidemiologylab/covid-twitter-bert/1
+BERT-large-uncased model, pretrained on a corpus of messages from Twitter about COVID-19
+
+<!-- asset-path: https://crowdbreaks-public.s3.eu-central-1.amazonaws.com/models/covid-twitter-bert/v1/tfhub/covid-twitter-bert-v1.tar.gz -->
+<!-- module-type: text-embedding -->
+<!-- network-architecture: Transformer -->
+<!-- dataset: Twitter -->
+<!-- language: en -->
+<!-- fine-tunable: true -->
+<!-- format: saved_model_2 -->
+
+## Overview
+
+This model was trained on 160M tweets collected between January 12 and April 16, 2020 containing at least one of the keywords "wuhan", "ncov", "coronavirus", "covid", or "sars-cov-2". These tweets were filtered and preprocessed to reach a final sample of 22.5M tweets (containing 40.7M sentences and 633M tokens) which were used for training.
+
+This model was evaluated based on downstream classification tasks, but it could be used for any other NLP task which can leverage contextual embeddings. 
+
+In order to achieve best results, make sure to use the same text preprocessing as we did for pretraining. This involves replacing user mentions, urls and emojis. You can find a script on our projects [GitHub repo](https://github.com/digitalepidemiologylab/covid-twitter-bert).
+
+
+## Example use
+The saved model can be loaded directly:
+
+```python
+max_seq_length = 96  # Your choice here.
+input_word_ids = tf.keras.layers.Input(
+  shape=(max_seq_length,),
+  dtype=tf.int32,
+  name="input_word_ids")
+input_mask = tf.keras.layers.Input(
+  shape=(max_seq_length,),
+  dtype=tf.int32,
+  name="input_mask")
+input_type_ids = tf.keras.layers.Input(
+  shape=(max_seq_length,),
+  dtype=tf.int32,
+  name="input_type_ids")
+bert_layer = hub.KerasLayer("https://tfhub.dev/digitalepidemiologylab/covid-twitter-bert/1", trainable=True)
+pooled_output, sequence_output = bert_layer([input_word_ids, input_mask, input_type_ids])
+```
+
+From there you can create a classifier model the following way:
+```python
+num_labels = 3
+initializer = tf.keras.initializers.TruncatedNormal(stddev=0.2)
+output = tf.keras.layers.Dropout(rate=0.1)(pooled_output)
+output = tf.keras.layers.Dense(num_labels, kernel_initializer=initializer, name='output')(output)
+classifier_model = tf.keras.Model(
+  inputs={
+          'input_word_ids': input_word_ids,
+          'input_mask': input_mask,
+          'input_type_ids': input_type_ids}, 
+  outputs=output)
+```
+
+You can load the tokenizer the following way. The vocab is equivalent to the official bert-large-uncased vocab:
+```python
+vocab_file = bert_layer.resolved_object.vocab_file.asset_path.numpy()
+do_lower_case = bert_layer.resolved_object.do_lower_case.numpy()
+tokenizer = tokenization.FullTokenizer(vocab_file, do_lower_case)
+```
+
+## References
+[1] Martin Müller, Marcel Salaté, Per E Kummervold. "COVID-Twitter-BERT: A Natural Language Processing Model to Analyse COVID-19 Content on Twitter" arXiv preprint arXiv:2005.07503 (2020).