Open damonbayer opened 1 month ago
Agree
This looks nice.
I don't know much about this but here is a big list of tools: https://github.com/kelvins/awesome-mlops
IMO I think the key is we use something that has some useful set of features and isn't that hard to get started with not that we make the right choice the first time
Also considering https://metaflow.org/
Have done a tiny bit of reading on this recently. Broadly, we should be considering "MLOps" tools. A lot of these seem to be proprietary and/or aimed at more traditional machine learning, but maybe some would be useful to us or offer lessons. The main one I keep seeing is https://mlflow.org/
Random example using mlflow with numpyro: https://gitlab.com/tk0345/numpyro-mlflow-example/-/blob/main/numpyro_mlflow_example.ipynb