How to load keras models in parralel?

Research & Models
,
#1

I am loading two keras models, right now I am doing it sequentially:

page_encoder = tf.keras.models.load_model(PAGE_MODEL_PATH, compile=False)
query_encoder = tf.keras.models.load_model(QUERY_MODEL_PATH, compile=False)

As the models are fairly large - I would like to do this in parallel, any help would be appreciated!

#3

Hi @dendog ,

I think you can load Keras models in parallel using Python multiprocessing module to load each model in a separate process.

def model_loading(models_path):
    return tf.keras.models.load_model(models_path)

model1 = 'path_to_model1'
model2 = 'path_to_model2'

__main__:
    model1_parallel = multiprocessing.Process(target=model_loading, args=(model1,))
    model2_parallel = multiprocessing.Process(target=model_loading, args=(model2,))

    model1_parallel.start()
    model2_parallel.start()


    model1_parallel.join()
    model2_parallel.join()

For more information you can use this python multiprocessing documentation.