mlem
mlem copied to clipboard
iterative
🐶 A tool to package, serve, and deploy any ML model on any platform. Archived to be resurrected one day🤞
We've been considering that change in mlem.yaml introduces new model version. But if model description or tags ("image-classification", etc) change, this is not a new version. - commit (doesn't change)...
As discussed with @mnrozhkov and @tapadipti, it would be great if MLEM would somehow simplify model application within Airflow DAGs. Two questions to start with: 1. We can create something...
``` mlem build rf docker --conf server.type=fastapi --conf image.name=rf-image-test ⏳️ Loading model from .mlem/model/rf.mlem 💼 Adding model files... 🛠 Generating dockerfile... 💼 Adding sources... 💼 Generating requirements file... 🛠 Building...
Now it's `Union[str, AbstractFileSystem, None]`. But underlying implementation (in `MlemObject._parse_dump_args`) has it as `Optional[AbstractFileSystem]`. Need to either include `str` there (and corresponding logic) or remove `str` from `save` signature
We need to add support for the huggingface transformers models and datasets.
If we add this support in May, @alex000kim could use it to showcase MLEM in his tutorial.
Right now you have to manually set importing type when you use `import` or `apply` except for pickle imports, because all other implementations (pandas) are a part of ext and...
I think it would be helpful to have `mlem.api.ls()` return something easier to read/parse, especially because: * By default, mlem is now saving everything in `.mlem`, making it harder for...
In get started, the example shows `y_pred = apply("rf", "test_x.csv", method="predict_proba")` and below says to use `import` if your dataset is not saved as a mlem object. First, I tried...
Labels are a part of `lgb.Dataset`, but not a part of `inner` DatasetType. So we need to add separate field `lables: DatasetType` to DatasetType and corresponding field to Reader/Writer and...