I have a TensorFlow model that I use to make predictions via API. Currently, I have to use the full version of TensorFlow, which is nearly 1GB! I am only using:
from tensorflow.keras.models import load_model
model = load_model(
"/tmp/current_model",
custom_objects=None,
compile=True
)
predictions = model.predict(inputs, batch_size=1, verbose=1, steps=None, callbacks=None)
Is there a version or strategy to load a model & predict() that doesn't require the entire 1GB Tensorflow package? This no longer requires any of the training functionality.