tensorflow/lite/g3doc/api_docs/python/tflite_model_maker/question_answer/BertQaSpec.md
page_type: reference description: A specification of BERT model for question answering.
<link rel="stylesheet" href="/site-assets/css/style.css"> <!-- DO NOT EDIT! Automatically generated file. --> <div itemscope itemtype="http://developers.google.com/ReferenceObject"> <meta itemprop="name" content="tflite_model_maker.question_answer.BertQaSpec" /> <meta itemprop="path" content="Stable" /> <meta itemprop="property" content="__init__"/> <meta itemprop="property" content="build"/> <meta itemprop="property" content="convert_examples_to_features"/> <meta itemprop="property" content="create_model"/> <meta itemprop="property" content="evaluate"/> <meta itemprop="property" content="get_config"/> <meta itemprop="property" content="get_default_quantization_config"/> <meta itemprop="property" content="get_name_to_features"/> <meta itemprop="property" content="predict"/> <meta itemprop="property" content="predict_tflite"/> <meta itemprop="property" content="reorder_input_details"/> <meta itemprop="property" content="reorder_output_details"/> <meta itemprop="property" content="save_vocab"/> <meta itemprop="property" content="select_data_from_record"/> <meta itemprop="property" content="train"/> <meta itemprop="property" content="compat_tf_versions"/> <meta itemprop="property" content="convert_from_saved_model_tf2"/> <meta itemprop="property" content="need_gen_vocab"/> </div>View source on GitHub
A specification of BERT model for question answering.
<pre class="devsite-click-to-copy prettyprint lang-py tfo-signature-link"> <code>tflite_model_maker.question_answer.BertQaSpec( uri='https://tfhub.dev/tensorflow/bert_en_uncased_L-12_H-768_A-12/1', model_dir=None, seq_len=384, query_len=64, doc_stride=128, dropout_rate=0.1, initializer_range=0.02, learning_rate=8e-05, distribution_strategy='mirrored', num_gpus=-1, tpu='', trainable=True, predict_batch_size=8, do_lower_case=True, is_tf2=True, tflite_input_name=None, tflite_output_name=None, init_from_squad_model=False, default_batch_size=16, name='Bert' ) </code></pre> <!-- Placeholder for "Used in" --> <!-- Tabular view --> <table class="responsive fixed orange"> <colgroup><col width="214px"><col></colgroup> <tr><th colspan="2"><h2 class="add-link">Args</h2></th></tr> <tr> <td> `uri`<a id="uri"></a> </td> <td> TF-Hub path/url to Bert module. </td> </tr><tr> <td> `model_dir`<a id="model_dir"></a> </td> <td> The location of the model checkpoint files. </td> </tr><tr> <td> `seq_len`<a id="seq_len"></a> </td> <td> Length of the sequence to feed into the model. </td> </tr><tr> <td> `query_len`<a id="query_len"></a> </td> <td> Length of the query to feed into the model. </td> </tr><tr> <td> `doc_stride`<a id="doc_stride"></a> </td> <td> The stride when we do a sliding window approach to take chunks of the documents. </td> </tr><tr> <td> `dropout_rate`<a id="dropout_rate"></a> </td> <td> The rate for dropout. </td> </tr><tr> <td> `initializer_range`<a id="initializer_range"></a> </td> <td> The stdev of the truncated_normal_initializer for initializing all weight matrices. </td> </tr><tr> <td> `learning_rate`<a id="learning_rate"></a> </td> <td> The initial learning rate for Adam. </td> </tr><tr> <td> `distribution_strategy`<a id="distribution_strategy"></a> </td> <td> A string specifying which distribution strategy to use. Accepted values are 'off', 'one_device', 'mirrored', 'parameter_server', 'multi_worker_mirrored', and 'tpu' -- case insensitive. 'off' means not to use Distribution Strategy; 'tpu' means to use TPUStrategy using `tpu_address`. </td> </tr><tr> <td> `num_gpus`<a id="num_gpus"></a> </td> <td> How many GPUs to use at each worker with the DistributionStrategies API. The default is -1, which means utilize all available GPUs. </td> </tr><tr> <td> `tpu`<a id="tpu"></a> </td> <td> TPU address to connect to. </td> </tr><tr> <td> `trainable`<a id="trainable"></a> </td> <td> boolean, whether pretrain layer is trainable. </td> </tr><tr> <td> `predict_batch_size`<a id="predict_batch_size"></a> </td> <td> Batch size for prediction. </td> </tr><tr> <td> `do_lower_case`<a id="do_lower_case"></a> </td> <td> boolean, whether to lower case the input text. Should be True for uncased models and False for cased models. </td> </tr><tr> <td> `is_tf2`<a id="is_tf2"></a> </td> <td> boolean, whether the hub module is in TensorFlow 2.x format. </td> </tr><tr> <td> `tflite_input_name`<a id="tflite_input_name"></a> </td> <td> Dict, input names for the TFLite model. </td> </tr><tr> <td> `tflite_output_name`<a id="tflite_output_name"></a> </td> <td> Dict, output names for the TFLite model. </td> </tr><tr> <td> `init_from_squad_model`<a id="init_from_squad_model"></a> </td> <td> boolean, whether to initialize from the model that is already retrained on Squad 1.1. </td> </tr><tr> <td> `default_batch_size`<a id="default_batch_size"></a> </td> <td> Default batch size for training. </td> </tr><tr> <td> `name`<a id="name"></a> </td> <td> Name of the object. </td> </tr> </table><a target="_blank" class="external" href="https://github.com/tensorflow/examples/blob/master/tensorflow_examples/lite/model_maker/core/task/model_spec/text_spec.py#L438-L445">View source</a>
<pre class="devsite-click-to-copy prettyprint lang-py tfo-signature-link"> <code>build() </code></pre>Builds the class. Used for lazy initialization.
<h3 id="convert_examples_to_features"><code>convert_examples_to_features</code></h3><a target="_blank" class="external" href="https://github.com/tensorflow/examples/blob/master/tensorflow_examples/lite/model_maker/core/task/model_spec/text_spec.py#L871-L885">View source</a>
<pre class="devsite-click-to-copy prettyprint lang-py tfo-signature-link"> <code>convert_examples_to_features( examples, is_training, output_fn, batch_size ) </code></pre>Converts examples to features and write them into TFRecord file.
<h3 id="create_model"><code>create_model</code></h3><a target="_blank" class="external" href="https://github.com/tensorflow/examples/blob/master/tensorflow_examples/lite/model_maker/core/task/model_spec/text_spec.py#L887-L899">View source</a>
<pre class="devsite-click-to-copy prettyprint lang-py tfo-signature-link"> <code>create_model() </code></pre>Creates the model for qa task.
<h3 id="evaluate"><code>evaluate</code></h3><a target="_blank" class="external" href="https://github.com/tensorflow/examples/blob/master/tensorflow_examples/lite/model_maker/core/task/model_spec/text_spec.py#L1016-L1083">View source</a>
<pre class="devsite-click-to-copy prettyprint lang-py tfo-signature-link"> <code>evaluate( model, tflite_filepath, dataset, num_steps, eval_examples, eval_features, predict_file, version_2_with_negative, max_answer_length, null_score_diff_threshold, verbose_logging, output_dir ) </code></pre>Evaluate QA model.
<!-- Tabular view --> <table class="responsive fixed orange"> <colgroup><col width="214px"><col></colgroup> <tr><th colspan="2">Args</th></tr> <tr> <td> `model` </td> <td> The keras model to be evaluated. </td> </tr><tr> <td> `tflite_filepath` </td> <td> File path to the TFLite model. </td> </tr><tr> <td> `dataset` </td> <td> tf.data.Dataset used for evaluation. </td> </tr><tr> <td> `num_steps` </td> <td> Number of steps to evaluate the model. </td> </tr><tr> <td> `eval_examples` </td> <td> List of `squad_lib.SquadExample` for evaluation data. </td> </tr><tr> <td> `eval_features` </td> <td> List of `squad_lib.InputFeatures` for evaluation data. </td> </tr><tr> <td> `predict_file` </td> <td> The input predict file. </td> </tr><tr> <td> `version_2_with_negative` </td> <td> Whether the input predict file is SQuAD 2.0 format. </td> </tr><tr> <td> `max_answer_length` </td> <td> The maximum length of an answer that can be generated. This is needed because the start and end predictions are not conditioned on one another. </td> </tr><tr> <td> `null_score_diff_threshold` </td> <td> If null_score - best_non_null is greater than the threshold, predict null. This is only used for SQuAD v2. </td> </tr><tr> <td> `verbose_logging` </td> <td> If true, all of the warnings related to data processing will be printed. A number of warnings are expected for a normal SQuAD evaluation. </td> </tr><tr> <td> `output_dir` </td> <td> The output directory to save output to json files: predictions.json, nbest_predictions.json, null_odds.json. If None, skip saving to json files. </td> </tr> </table> <!-- Tabular view --> <table class="responsive fixed orange"> <colgroup><col width="214px"><col></colgroup> <tr><th colspan="2">Returns</th></tr> <tr class="alt"> <td colspan="2"> A dict contains two metrics: Exact match rate and F1 score. </td> </tr> </table> <h3 id="get_config"><code>get_config</code></h3><a target="_blank" class="external" href="https://github.com/tensorflow/examples/blob/master/tensorflow_examples/lite/model_maker/core/task/model_spec/text_spec.py#L861-L869">View source</a>
<pre class="devsite-click-to-copy prettyprint lang-py tfo-signature-link"> <code>get_config() </code></pre>Gets the configuration.
<h3 id="get_default_quantization_config"><code>get_default_quantization_config</code></h3><a target="_blank" class="external" href="https://github.com/tensorflow/examples/blob/master/tensorflow_examples/lite/model_maker/core/task/model_spec/text_spec.py#L420-L424">View source</a>
<pre class="devsite-click-to-copy prettyprint lang-py tfo-signature-link"> <code>get_default_quantization_config() </code></pre>Gets the default quantization configuration.
<h3 id="get_name_to_features"><code>get_name_to_features</code></h3><a target="_blank" class="external" href="https://github.com/tensorflow/examples/blob/master/tensorflow_examples/lite/model_maker/core/task/model_spec/text_spec.py#L831-L845">View source</a>
<pre class="devsite-click-to-copy prettyprint lang-py tfo-signature-link"> <code>get_name_to_features( is_training ) </code></pre>Gets the dictionary describing the features.
<h3 id="predict"><code>predict</code></h3><a target="_blank" class="external" href="https://github.com/tensorflow/examples/blob/master/tensorflow_examples/lite/model_maker/core/task/model_spec/text_spec.py#L982-L984">View source</a>
<pre class="devsite-click-to-copy prettyprint lang-py tfo-signature-link"> <code>predict( model, dataset, num_steps ) </code></pre>Predicts the dataset for model.
<a target="_blank" class="external" href="https://github.com/tensorflow/examples/blob/master/tensorflow_examples/lite/model_maker/core/task/model_spec/text_spec.py#L996-L1014">View source</a>
<pre class="devsite-click-to-copy prettyprint lang-py tfo-signature-link"> <code>predict_tflite( tflite_filepath, dataset ) </code></pre>Predicts the dataset for TFLite model in tflite_filepath.
<a target="_blank" class="external" href="https://github.com/tensorflow/examples/blob/master/tensorflow_examples/lite/model_maker/core/task/model_spec/text_spec.py#L426-L436">View source</a>
<pre class="devsite-click-to-copy prettyprint lang-py tfo-signature-link"> <code>reorder_input_details( tflite_input_details ) </code></pre>Reorders the tflite input details to map the order of keras model.
<h3 id="reorder_output_details"><code>reorder_output_details</code></h3><a target="_blank" class="external" href="https://github.com/tensorflow/examples/blob/master/tensorflow_examples/lite/model_maker/core/task/model_spec/text_spec.py#L986-L994">View source</a>
<pre class="devsite-click-to-copy prettyprint lang-py tfo-signature-link"> <code>reorder_output_details( tflite_output_details ) </code></pre>Reorders the tflite output details to map the order of keras model.
<h3 id="save_vocab"><code>save_vocab</code></h3><a target="_blank" class="external" href="https://github.com/tensorflow/examples/blob/master/tensorflow_examples/lite/model_maker/core/task/model_spec/text_spec.py#L447-L452">View source</a>
<pre class="devsite-click-to-copy prettyprint lang-py tfo-signature-link"> <code>save_vocab( vocab_filename ) </code></pre>Prints the file path to the vocabulary.
<h3 id="select_data_from_record"><code>select_data_from_record</code></h3><a target="_blank" class="external" href="https://github.com/tensorflow/examples/blob/master/tensorflow_examples/lite/model_maker/core/task/model_spec/text_spec.py#L847-L859">View source</a>
<pre class="devsite-click-to-copy prettyprint lang-py tfo-signature-link"> <code>select_data_from_record( record ) </code></pre>Dispatches records to features and labels.
<h3 id="train"><code>train</code></h3><a target="_blank" class="external" href="https://github.com/tensorflow/examples/blob/master/tensorflow_examples/lite/model_maker/core/task/model_spec/text_spec.py#L901-L946">View source</a>
<pre class="devsite-click-to-copy prettyprint lang-py tfo-signature-link"> <code>train( train_ds, epochs, steps_per_epoch, **kwargs ) </code></pre>Run bert QA training.
<!-- Tabular view --> <table class="responsive fixed orange"> <colgroup><col width="214px"><col></colgroup> <tr><th colspan="2">Args</th></tr> <tr> <td> `train_ds` </td> <td> tf.data.Dataset, training data to be fed in tf.keras.Model.fit(). </td> </tr><tr> <td> `epochs` </td> <td> Integer, training epochs. </td> </tr><tr> <td> `steps_per_epoch` </td> <td> Integer or None. Total number of steps (batches of samples) before declaring one epoch finished and starting the next epoch. If `steps_per_epoch` is None, the epoch will run until the input dataset is exhausted. </td> </tr><tr> <td> `**kwargs` </td> <td> Other parameters used in the tf.keras.Model.fit(). </td> </tr> </table> <!-- Tabular view --> <table class="responsive fixed orange"> <colgroup><col width="214px"><col></colgroup> <tr><th colspan="2">Returns</th></tr> <tr class="alt"> <td colspan="2"> tf.keras.Model, the keras model that's already trained. </td> </tr> </table> <!-- Tabular view --> <table class="responsive fixed orange"> <colgroup><col width="214px"><col></colgroup> <tr><th colspan="2"><h2 class="add-link">Class Variables</h2></th></tr> <tr> <td> compat_tf_versions<a id="compat_tf_versions"></a> </td> <td> `[2]` </td> </tr><tr> <td> convert_from_saved_model_tf2<a id="convert_from_saved_model_tf2"></a> </td> <td> `True` </td> </tr><tr> <td> need_gen_vocab<a id="need_gen_vocab"></a> </td> <td> `False` </td> </tr> </table>