Creates extra-processes
See original GitHub issueI’ve configured accelerate to handle 4,6,8,10 processes(different setups for testing). And it creates 7, 11, 15 and 19 processes respectively instead. The number is N-1 +N, because it creates N-1 extra processes on the 4th card. I.e.
| 0 N/A N/A 58825 C python3 28175MiB |
| 1 N/A N/A 58826 C python3 28195MiB |
| 2 N/A N/A 58827 C python3 28199MiB |
| 3 N/A N/A 58825 C python3 1405MiB |
| 3 N/A N/A 58826 C python3 1405MiB |
| 3 N/A N/A 58827 C python3 1405MiB |
| 3 N/A N/A 58828 C python3 28179MiB |
And it doesn’t matter which card will be 4th. I changed the order in CUDA_VISIBLE_DEVICES and it still creates on the 4th in order.
Is this a bug or I’m doing something wrong?
Issue Analytics
- State:
- Created 2 years ago
- Comments:6 (2 by maintainers)
Top Results From Across the Web
Why so many multiple processes in Windows 10?
Hi, I noticed in the task manager that several processes occure more than one time, and some processes occure multiple times, ...
Read more >Google Chrome Creates Multiple Windows Processes
You can view the Chrome.exe processes in the Windows Task Manager. You may wonder why multiple processes are needed. What does each of...
Read more >Creating multiple processes in C - Stack Overflow
I am writing a program that needs to create multiple processes. Let's say this number is 3. I want each of these processes...
Read more >Can an application create multiple processes? - Super User
Yes. The Chrome web browser is a good example of this. Open Chrome and you'll find dozens of chrome.exe processes.
Read more >Create Multiple Processes - Synchro Arts
Processes are created on successive output tracks, starting with the one selected, and numbered sequentially. NOTE: Each process created in this step will ......
Read more >Top Related Medium Post
No results found
Top Related StackOverflow Question
No results found
Troubleshoot Live Code
Lightrun enables developers to add logs, metrics and snapshots to live code - no restarts or redeploys required.
Start FreeTop Related Reddit Thread
No results found
Top Related Hackernoon Post
No results found
Top Related Tweet
No results found
Top Related Dev.to Post
No results found
Top Related Hashnode Post
No results found
Top GitHub Comments
Well, I fixed It with adding device in loading
torch.load(self.continue_from, self.accelerator.device)
It created extra objects on inappropriate gpuCUDA_VISIBLE_DEVICES=0,1,2,3,5,7,13,15 accelerate launch train.py