如何部署tensorflow系统？ 张量流服务？

Question

I have built a recommender system with tf.keras.

I'd like to deploy this to a live environment.

One thing I could do is, create a python-based webserver (django) that takes (http) rest request, do the prediction with the trained model and gives back the result as a rest response.

I wonder if there's recommended alternative than doing the above?

I can guess tensorflow serving might be related to what I'm trying to do, but couldn't tell for sure..

• I'm subsclassing tf.keras.models.Model to create my model, and it doesn't support model.save() only model.save_weights .. It might affect my options I guess.

Answer 1

Tesnsorflow serving is the way to go here: https://github.com/tensorflow/serving

It has some very good features: https://youtu.be/CxUc5FJF_9w?t=1056 And this video is very informative in general about TensorFlow serving and TFX.