`build`: more formats to export your model

aguschin commented 2 years ago

MLEM could be a powerful tool if you need to distribute your model with different channels and use it in different circumstances (or easily switch between those). This could be a part of Value Props for us. So I think MLEM should be able to convert/build/export MLEM model to any widely used format and back, including:

[x] Python package
[x] Docker image
[ ] MLflow format
[ ] Seldon-compatible format
[ ] Pickle
[ ] .pt
[ ] .h5
[ ] apache TVM
[ ] onnx
[ ] Nvidia tensorrt
[ ] Explore integration with Ivy to export to ML frameworks other than the model was trained with

This list is going to be updated. Please feel free to post a comment or upvote the existing ones if you need something we don't support yet :)

daavoo commented 2 years ago

Would love https://onnx.ai/

daavoo commented 2 years ago

And https://developer.nvidia.com/tensorrt :)

judahrand commented 1 year ago

BentoML compatible format? Alternatively, it might be good to contribute a MLEM runner to BentoML?

aguschin commented 1 year ago

last comment is related to https://github.com/iterative/mlem/issues/265

iterative / mlem

`build`: more formats to export your model #208