For NAM users out there, there’s a free service that recently launched that offers free GPU model training. You simply upload the captured audio (assuming you used the standard NAM training sample, if not, you can upload a dry/wet pair), and you get back the trained model (after some time). The training is done purely on the server side, so no client side resources are required.
(No affiliation in any way.)
Apparently, it’s a side project of a VC-funded startup, so who knows how long this free service will remain free. As they say, make hay while the sun shines. 
If you wanted to try out capturing your own gear, but haven’t had the stomach to wade through python scripts and/or don’t have a GPU, this is your chance!