It depends on what a "large model" means to you.
If we are talking about "ridiculously large," like in ImageNet-type-large, then you probably won't be able to do it for free.
But then, why would you want to train such a massive model?
If we are talking about "ridiculously large," like in ImageNet-type-large, then you probably won't be able to do it for free.
But then, why would you want to train such a massive model?
https://twitter.com/AnandforReal/status/1367503897582051336
If we get it down a couple of notches, we get into the realm of "a few days of training" as long as you have a GPU.
You could train this locally, assuming you have a computer with the right hardware. But that computer won't be cheap either.
You could train this locally, assuming you have a computer with the right hardware. But that computer won't be cheap either.
If we are talking about hours of training, then Google Colab might be all you need, and you can access it for free.
It won't be the best experience, but again, hard to beat free.
It won't be the best experience, but again, hard to beat free.
Finally, as a student, and assuming you are learning this stuff, you don't need to obsess on training large models.
Focus on learning the ropes, and grow into a position where a company will worry about paying the bills.
Focus on learning the ropes, and grow into a position where a company will worry about paying the bills.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
