[darshanime]

Hi, thanks for sharing. There are quite different tools; afaiu, the one you shared does not have any means of cross referencing other data. Also I could see only basic knobs to control the data generation -- ints b/w max/min, weighted distribution from a set of possible options etc.

datagen on the other hand allows you to access the data of any model, any field, any row to create new data; much like a DAG. This is a very powerful abstraction.

Of course, not having to write "code" in json is great too!

[ProofHouse]

Is there a good way this could be used for model distillation? Hmmm