Closed cul8rmom1 closed 2 months ago
I can't go through this all immediately. However, know that you should only need the extension installed on a single instance. The other instances merely need their api exposed so that you can add them in the main instance.
The solution in this case was to increase batch size before iterations/batch count
Awesome extension. I have copied the entire install directory between all of the machines to make sure everything is apples to apples. This is the start command line on the 2 remote workers. set COMMANDLINE_ARGS= --theme dark --xformers --listen --api --api-server-stop.
Running windows 10/11 with all virus scan/firewall etc turned off.
Everything works great until...
I can do the benchmark and it runs on all nodes. All good in the hood. Now I change the resolution. Then it says "DISTRIBUTED | DEBUG distributed has nothing to do, returning control to webui distributed.py:303"
I turn complimentary back on and it creates a crap ton of extra on the slower nodes.
I will jump in discord and see who says what. Let me know if there is any other information I can provide.
Thanks!!