this community obviously cares about running models locally. but i've been wondering if the bigger problem is training, not inference local inference is cool but the models still get trained in datacenters by big labs. is there a path where training also gets distributed or is that fundamentally too
