Model from nascarr.

PiperOrigin-RevId: 293784873

tfhub_dev/assets/nascarr/models/nq/1.md

@@ -0,0 +1,49 @@
+# Module nascarr/nq/1
+Question Answering model trained on Natural Questions Dataset [1, 2].
+
+
+<!-- asset-path: https://github.com/nascarr/natural-questions/raw/master/models/v1/model.tar.gz -->
+<!-- module-type: text-question-answering -->
+<!-- network-architecture: Transformer -->
+<!-- dataset: Natural Questions -->
+<!-- language: en -->
+<!-- fine-tunable: true -->
+<!-- format: saved_model_2 -->
+
+## Overview
+
+Model was created as part of a Gold Medal solution at TensorFlow 2.0 Question Answering competition [3].
+Solution develops ideas from "A BERT Baseline for the Natural Questions" paper [4] and uses Bert Large Uncased (Whole Word Masking)
+as a pretrained model [5]. Model is fine-tuned on Natural Questions Dataset.
+More detailed description of the solution can be found on Kaggle forum [6].
+
+
+#### Example use
+
+```
+seq_len = 512
+unique_id  = tf.keras.Input(shape=(1,),dtype=tf.int64,name='unique_id')
+input_ids   = tf.keras.Input(shape=(seq_len,),dtype=tf.int32,name='input_ids') # input word ids
+input_mask  = tf.keras.Input(shape=(seq_len,),dtype=tf.int32,name='input_mask')
+segment_ids = tf.keras.Input(shape=(seq_len,),dtype=tf.int32,name='segment_ids')
+model = hub.KerasLayer("https://tfhub.dev/nascarr/nq/1", trainable=True)
+unique_id, start_logits, end_logits, ans_type = model([unique_id, input_ids, input_mask, segment_ids])
+```
+
+
+#### References
+
+
+[1] Tom Kwiatkowski, Jennimaria Palomaki, Olivia Redfield, Michael Collins, Ankur Parikh, Chris Alberti, Danielle Epstein, Illia Polosukhin, Matthew Kelcey, Jacob Devlin, Kenton Lee, Kristina N. Toutanova, Llion Jones, Ming-Wei Chang, Andrew Dai, Jakob Uszkoreit, Quoc Le, Slav Petrov.
+[Natural Questions: a Benchmark for Question Answering Research](https://research.google/pubs/pub47761/)
+
+[2] https://ai.google.com/research/NaturalQuestions
+
+[3] https://www.kaggle.com/c/tensorflow2-question-answering
+
+[4] Chris Alberti, Kenton Lee, Michael Collins.
+[A BERT Baseline for the Natural Questions](https://arxiv.org/abs/1901.08634)
+
+[5] https://github.com/google-research/bert
+
+[6] https://www.kaggle.com/c/tensorflow2-question-answering/discussion/128278

tfhub_dev/assets/nascarr/nascarr.md

@@ -0,0 +1,7 @@
+# Publisher nascarr
+Anastasiya Karpovich
+
+## Anastasiya Karpovich
+NLP Engineer. Kaggle competitor.
+
+https://www.kaggle.com/user189546