|
|
|
|
|
|
by pksebben
846 days ago
|
|
|
Even if it is all open sourced, we still have the structural problem of training models large enough to do interesting stuff. Until we can train incrementally and distribute the workload scalably, it doesn't matter how open the models / methods for training are if you still need a bajilllion A100 hours to train the damn things. |
|
|