Training AI applications

Can glitch be used for AI training? I am aware of the 12 hour running limit, but resource wise, is feasible or can disturb other people?

Hey @CodingEngineer I think that would be super interesting!

Since each Glitch project is isolated by virtue of being in its own container you’re unlikely to affect other users at all, and if your project did get out of hand somehow our systems would automatically suspend it until you contacted us, so I’m not at all concerned about that. I’d honestly be more worried about whether or not a Glitch project container would offer you the resources you need to do what you’re hoping to - I know for example that some combinations of tensorflow packages can easily exceed the shared space available for packages.

Please let us know how it goes!


I am just worried that this could make other projects slower. The resources are very limited, however, I am currently doing this in my own pc and that means that I need to leave the pc for large periods. I will try out some stuff and will tell the results here. Thanks!

I got the numbers! Using the code in this repo!/ai-testing, I got an average of 50.788 ms on my computer, with 12Gb of ram and a Ryzen 5 2500U cpu, and on glitch I got 58.627 ms on average. I made 3 tests. A problem that I found is that glitch restarts the project as soon it finishes, do I need express to stop it? Anyway, the diference is not that big, and I have the advantage of being able to use my computer, so I can call this a success. Thanks!

1 Like