mlbench

This looks like a very nice abstraction. Good job Ralf and team, can't wait to try it out! On Fri

yea storing all checkpoints will not always be necessary. i was bringing this idea up basically just

I think master/helm in separate repositories makes complete sense. Especially since the Helm public

Hi Everyone, I created a project homepage at https://mlbench.github.io/ It's a Jekyll blog hosted

We can use CIFAR-10 to develop framework and perform some first results on benchmarking. It is fast

This metric is a lot easier to implement and evaluate. We train for a fixed amount of time, then stop

Makes sense, thanks On Mon, Aug 13, 2018, 03:48 Martin Jaggi <m.jaggi@gmail.com> wrote: On Mon,

New thread to discuss the work being done right now for the first release. I've added several

Do we use master node for computation? I don't think we should use it for computation, to not

also prefer master/workers naming On Tue, Jul 10, 2018 at 5:31 PM Ralf Grubenmann <ralf.grubenmann