I wonder if it's possible for a huge number of hobbyists to team up and train a model together in a distributed manner like seti@home or folding@home. Or does this kind of workload not really lend itself to that approach?
Those things were of course characterised by the ability to spread the work into pretty self-contained work packages. Not sure if that can be done with model training.
Comments
I wonder if it's possible for a huge number of hobbyists to team up and train a model together in a distributed manner like seti@home or folding@home. Or does this kind of workload not really lend itself to that approach?
Those things were of course characterised by the ability to spread the work into pretty self-contained work packages. Not sure if that can be done with model training.
Not likely to work. Not many (any) hobbyists can get 400gbps network throughput between each other's GPUs...