How to feed multiple inputs to tf.keras.Model.predict?

Question

Given this simple tensorflow toy model

import tensorflow as tf
inputs = {
    "a":tf.keras.Input(shape=(), name="input_a"),
    "b":tf.keras.Input(shape=(), name="input_b")
}
outputs = tf.keras.layers.Add()([inputs["a"], inputs["b"]])
model = tf.keras.Model(inputs=inputs, outputs=outputs)

I can get its output by invoking it with inputs as defined, so the following:

model({"a":2,"b":3})

Gives the output:

<tf.Tensor: shape=(), dtype=float32, numpy=5.0>

But invoking the predict function:

model.predict({"a":2,"b":3})

Gives the following error:

ValueError: Failed to find data adapter that can handle input: (<class 'dict'> containing {"<class 'str'>"} keys and {"<class 'int'>"} values), <class 'NoneType'>

So how do I correctly invoke the predict function when my model has more than a single input as in this case?

Answer 1

It seems that the predict method can handle a dictionary of BATCHES of numpy arrays, this works: model.predict({'a': np.full((1,), 3), 'b': np.full((1,), 2)}) and outputs <tf.Tensor: shape=(1,), dtype=float32, numpy=array([5.], dtype=float32)> In this case, the batch size is equal to 1.