Convert TensorFlow 2 BERT Transformer Models

The following examples demonstrate converting TensorFlow 2 models to Core ML using coremltools.

Convert the DistilBERT transformer model

The following example converts the DistilBERT model from Huggingface to Core ML.

📘

Install Transformers

You may need to first install Transformers version 2.10.0.

Follow these steps:

1. Add the import statements:

Python

import numpy as np
import coremltools as ct
import tensorflow as tf

from transformers import DistilBertTokenizer, TFDistilBertForMaskedLM

2. Load the DistilBERT model and tokenizer. This example uses the TFDistilBertForMaskedLM variant:

Python

tokenizer = DistilBertTokenizer.from_pretrained('distilbert-base-cased')
distilbert_model = TFDistilBertForMaskedLM.from_pretrained('distilbert-base-cased')

3. Describe and set the input layer, and then build the TensorFlow model (tf_model):

Python

max_seq_length = 10
input_shape = (1, max_seq_length) #(batch_size, maximum_sequence_length)

input_layer = tf.keras.layers.Input(shape=input_shape[1:], dtype=tf.int32, name='input')

prediction_model = distilbert_model(input_layer)
tf_model = tf.keras.models.Model(inputs=input_layer, outputs=prediction_model)

4. Convert the tf_model to the Core ML format (mlmodel):

Python

mlmodel = ct.convert(tf_model)

5. Create the input using tokenizer:

Python

# Fill the input with zeros to adhere to input_shape
input_values = np.zeros(input_shape)
# Store the tokens from our sample sentence into the input
input_values[0,:8] = np.array(tokenizer.encode("Hello, my dog is cute")).astype(np.int32)

6. Use mlmodel for prediction:

Python

mlmodel.predict({'input':input_values}) # 'input' is the name of our input layer from (3)

Convert the TF Hub BERT transformer model

The following example converts the BERT model from TensorFlow Hub. Follow these steps:

1. Add the import statements:

Python

import numpy as np
import tensorflow as tf
import tensorflow_hub as tf_hub

import coremltools as ct

2. Describe and set the input layer:

Python

max_seq_length = 384
input_shape = (1, max_seq_length)

input_words = tf.keras.layers.Input(
    shape=input_shape[1:], dtype=tf.int32, name='input_words')
input_masks = tf.keras.layers.Input(
    shape=input_shape[1:], dtype=tf.int32, name='input_masks')
segment_ids = tf.keras.layers.Input(
    shape=input_shape[1:], dtype=tf.int32, name='segment_ids')

3. Build the TensorFlow model (tf_model):

Python

bert_layer = tf_hub.KerasLayer("https://tfhub.dev/tensorflow/bert_en_uncased_L-12_H-768_A-12/1", trainable=False)

pooled_output, sequence_output = bert_layer(
    [input_words, input_masks, segment_ids])

tf_model = tf.keras.models.Model(
    inputs=[input_words, input_masks, segment_ids],
    outputs=[pooled_output, sequence_output])

4. Convert the tf_model to Core ML (mlmodel):

Python

mlmodel = ct.convert(tf_model, source='TensorFlow')