With PyTorch now supporting production, thoughts on great projects for serving, model mgmt, etc…? We’ve tested a few projects like MLFlow/Kubeflow (with varying success) but would love to hear your thoughts on whether you’ve found them to support your needs or if there are other options to consider.
I’d really like to see something like tf serving for PyTorch. In terms of framework, automatic request batching seems one of the most important features where I would not know how to easily get that.
Thanks Thomas, we have someone looking into implementing Autobatching and will keep you looped as to the progress.
On serving, are you using a cloud platform today or another project? Concretely what features do you most critically need (tf serving has a lot of things that I suspect go unused)?