4 ms·
> we don't really have a way for the FOSS community to pool together that much money There must be open source projects with enough money to pool into such a
by seydor 4y ago
> we don't really have a way for the FOSS community to pool together that much money
There must be open source projects with enough money to pool into such a project. I wonder whether wikimedia or apache are considering anything.
- sceadu 4y agoMaybe we can repurpose the SETI@home infrastructure :)
- kmeisthax 4y agoBOINC might be usable but the existing distributed training setups assume all nodes have very high speed I/O so they can trade gradients and model updates around quickly. The kind of setup that's feasible for BOINC is "here's a dataset shard, here's the last epoch, send me back gradients and I'll average them with the other ones I get to make the next epoch". This is quite a bit different from, say, the single-node case which is entirely serial and model updates happen every step rather than epoch.
- totony 4y agoOr big cloud platform could give some compute for free, give back some of the profit they get from oss.