Artificial Intelligence - Tensorflow

Back to Course

Lesson Description

Lession - #1067 Tensorflow Distributed Computing

TensorFlow supports distributed computing

TensorFlow supports distributed computing, permitting bits of the diagram to be figured on various cycles, which might be on totally various servers! Moreover, this can be utilized to distributed computing to servers with strong GPUs, and have different calculations done on servers with more memory, etc.

This part will zero in on the best way to get everything rolling with circulated TensorFlow. The point is to assist engineers with understanding the fundamental disseminated TF ideas that are repeating, like TF servers. We will utilize the Jupyter Notebook for assessing conveyed TensorFlow. The execution of conveyed registering with TensorFlow is referenced beneath −

Stage 1 − Import the essential modules compulsory for conveyed registering −

import tensorflow as tf

Stage 2 − Create a TensorFlow bunch with one hub. Allow this hub to be liable for a task that that has name "specialist" and that will work one take at localhost:2222.

cluster_spec = tf.train.ClusterSpec({'worker' : ['localhost:2222']}>
server = tf.train.Server(cluster_spec>

The above scripts produce the accompanying result −

The server is right now running.

Stage 3 − The server setup with particular meeting can be determined by executing the accompanying order −


group {
   work {
      name: "specialist"
      undertakings {
         esteem: "localhost:2222"
job_name: "specialist"
convention: "grpc"

Stage 4 − Launch a TensorFlow meeting with the execution motor being the server. Utilize TensorFlow to make a neighborhood server and use lsof to figure out the area of the server.

sess = tf.Session(target = server.target>
server = tf.train.Server.create_local_server(>

Stage 5 − View gadgets accessible in this meeting and close the particular meeting.

gadgets = sess.list_devices(>
for d in gadgets: print(d.name>

The above order produces the accompanying result −