Restoring checkpoint in distributed tensorflow


Restoring checkpoint in distributed tensorflow



Using a setup similar to https://github.com/tensorflow/models/tree/master/inception, the chief worker automatically saves a checkpoint file periodically on the node this process is running on. I'm running two ps on two different nodes. Two workers are also running on the two nodes each, with one out of 4 workers being the chief.



When restarting training without any modification, the Supervisor automatically tries to restore the last checkpoint file, but ends up giving an error that it could not find the ckpt on the second node (the node other than the chief worker), because the chief never saved the ckpt on the second node.


W tensorflow/core/framework/op_kernel.cc:936] Not found: Unsuccessful TensorSliceReader constructor: Failed to find any matching files for /home/muneebs/tf_train/model.ckpt-275



If I copy the ckpt directory to the second node, it restores fine. Is it a bug? Should the saver be initialized as sharded=True? If so, is that the only way, and we can't have the ckpt as a single file in case the number of nodes change later on?




1 Answer
1



A distributed file system like hdfs would help.



U can save the model (ckpt) to a directory in hdfs, thus avoiding the question of restoring ckpt.



Another method is launch the ps and worker whose task_index=0 in a same machine.





hdfs is not an option. task_index doesn't matter because each ps on each machine has to restore the part of the model parameters that it's supposed to manage.
– muneeb
Aug 11 '17 at 18:49






By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Comments

Post a Comment

Popular posts from this blog

paramiko-expect timeout is happening after executing the command

paramiko-expect timeout is happening after executing the command Can someone help me why timeout is happening after executing the command .I am trying to SSH to a machine and execute a command.I want to store the result of the executed command . I have referred paramiko-timeout but i didn't get my expected result. import traceback try: import paramiko except ImportError: raise Exception("Please install paramiko , pip install paramiko") import pexpect from paramiko_expect import SSHClientInteraction def main(): HOSTNAME = "minion5.net" USERNAME = "root" PASSWORD = "help@432" PROMPT = "root@*:~$s+" try: client = paramiko.SSHClient() client.load_system_host_keys() client.set_missing_host_key_policy(paramiko.AutoAddPolicy()) client.connect(hostname=HOSTNAME, username=USERNAME, password=PASSWORD) with SSHClientInteraction(client, timeout=20, display=True) as interact: ...
Read more

Opening a url is failing in Swift

Opening a url is failing in Swift breakpointing inside the callback and logging success shows false. I added itms-apps to my plist: LSApplicationQueriesSchemes and put this in my app delegate: func application(_ app: UIApplication, open url: URL, options: [UIApplicationOpenURLOptionsKey : Any] = [:]) -> Bool { return true } func moreTapped() { let url = NSURL(string: "itms-apps://itunes.com/developer/quantum-productions/id979315877") if #available(iOS 10.0, *) { UIApplication.shared.open(url! as URL, options: [:], completionHandler: {(success: Bool) in }) } else { UIApplication.shared.openURL(url! as URL) } } Are you testing on a real device? – rmaddy Jun 21 at 23:57 I try on a device, get "Cannot Connect to App Store". Device is online ...
Read more

Export result set on Dbeaver to CSV

Export result set on Dbeaver to CSV Normally I use Dbeaver for windows and always export my result set like this: run my query --> select the result --> export the result set --> select export to clipboard --> done This step by step puts my result set in my clipboard and I can paste it wherever I want to work with it. The problem is that now I am using dbeaver for mac and this guide is not working. I can go on until the moment that I select my result set like in the image below: exporting data set But once I go further in the process, in the last step I get: no query Note that in "source" it was suppose to show the query that originated the result set, but instead it says just "select. As a result it does't select my result or anything (besides being "successful"). Normally my query would show up there automatically and I couldn't find any option that corrects this problem in the menus. Please help me! Thanks. Aft...
Read more