A couple years ago I would have had a bigger interest in this. Pytorch added elastic training as part of version 1.9 when they incorporated torch elastic into i

A couple years ago I would have had a bigger interest in this. Pytorch added elastic training as part of version 1.9 when they incorporated torch elastic into it. Tensorflow added an easier way to do fault tolerant training in 2.4 with backup and restore callback for keras models. For non-keras models it'll take some extra code but they do include some examples for this too. Also there's at least two open source free solutions for elastic training I know of. RaySGD, https://docs.ray.io/en/master/raysgd/raysgd.html https://docs.ray.io/en/master/raysgd/raysgd.html and elastic horovod, https://horovod.readthedocs.io/en/stable/elastic_include.html https://horovod.readthedocs.io/en/stable/elastic_include.htm... For me to consider this I'd need a comparison table with native framework solutions and these solutions along with what does it add for me.