CODEWITHSUNDEEP
tensorflowtensorflow-litegoogle-mlkit
casolorz
casolorz

Reputation: 9544

Error occurred when initializing NLClassifier: Type mismatch for input tensor serving_default_input_type_ids:0. Requested STRING, got INT32

I'm trying to learn how to use some ML stuff for Android. I got the Text Classification demo working and seems to work fine. So then I tried creating my own model.

The code I used to create my own model was this:

import numpy as np
import os

from tflite_model_maker import model_spec
from tflite_model_maker import text_classifier
from tflite_model_maker.config import ExportFormat
from tflite_model_maker.text_classifier import AverageWordVecSpec
from tflite_model_maker.text_classifier import DataLoader

import tensorflow as tf
assert tf.__version__.startswith('2')
tf.get_logger().setLevel('ERROR')

spec = model_spec.get('mobilebert_classifier')

train_data = DataLoader.from_csv(
    filename='/path to file/train.csv',
    text_column='sentence',
    label_column='label',
    model_spec=spec,
    is_training=True)

model = text_classifier.create(train_data, model_spec=spec, epochs=10)

model.export(export_dir='average_word_vec')

The code appeared to run fine and it created a model.tflite file for me. I then replaced the demo tflite file with mine. But when I run the demo I get the following error:

 java.lang.AssertionError: Error occurred when initializing NLClassifier: Type mismatch for input tensor serving_default_input_type_ids:0. Requested STRING, got INT32.
        at org.tensorflow.lite.task.text.nlclassifier.NLClassifier.initJniWithByteBuffer(Native Method)
        at org.tensorflow.lite.task.text.nlclassifier.NLClassifier.access$100(NLClassifier.java:67)
        at org.tensorflow.lite.task.text.nlclassifier.NLClassifier$2.createHandle(NLClassifier.java:223)
        at org.tensorflow.lite.task.core.TaskJniUtils.createHandleFromLibrary(TaskJniUtils.java:91)
        at org.tensorflow.lite.task.text.nlclassifier.NLClassifier.createFromBufferAndOptions(NLClassifier.java:219)
        at org.tensorflow.lite.task.text.nlclassifier.NLClassifier.createFromFileAndOptions(NLClassifier.java:175)
        at org.tensorflow.lite.task.text.nlclassifier.NLClassifier.createFromFile(NLClassifier.java:150)
        at org.tensorflow.lite.examples.textclassification.client.TextClassificationClient.load(TextClassificationClient.java:44)
        at org.tensorflow.lite.examples.textclassification.MainActivity.lambda$onStart$1$MainActivity(MainActivity.java:67)
        at org.tensorflow.lite.examples.textclassification.-$$Lambda$MainActivity$eJaQnJq74KcmPEczFE5swJIGydg.run(Unknown Source:2)

What am I missing?

Upvotes: 0

Views: 523

Answers (1)

user3152729
user3152729

Reputation: 26

In your codes you trained a MobileBERT model, but saved to the path of average_word_vec? spec = model_spec.get('mobilebert_classifier') model.export(export_dir='average_word_vec')

One posssiblity is: you use the model of average_word_vec, but add MobileBERT metadata, thus the preprocessing doesn't match.

Could you follow the Model Maker tutorial and try again? https://colab.sandbox.google.com/github/tensorflow/tensorflow/blob/master/tensorflow/lite/g3doc/tutorials/model_maker_text_classification.ipynb Make sure change the export path.

Upvotes: 1

Related Questions