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🤞
My 2cs feedback on UX that I got while following the tutorial. Looking at how packaging works now, I think we can make subcommands to make it more self-explainable. E.g....
investigation lead me nowhere so far, so just disabling pylint for now in #193
For unsupported model types we can add a "mock" `model_type`. This way we'll enable users to use different mlem commands that dont need to load actual model like `clone`. RN,...
While writing tests for `fastapi`, we needed to create payload for both `numpy` and `pandas` since the fixtures were parametrised. The expected column types for columns of the dataframe were...
https://github.com/iterative/mlem/blob/8d2be329b8f97a993a5835eb6d97d61e4d6c585f/README.md?plain=1#L30 Sorry, I haven't been able to really play with this yet and give any kind of real feedback, But I noticed the linked line and I'm really interested in...
This was pointed out by Vladimir: would be good to automatically assign tags for models. E.g., if we can infer from input and output that this is an image segmentation...
One important thing about model registry is showing which metrics were changed and how after in the specific model version. Although you can produce metrics in DVC Pipeline, DVC doesn't...
To move faster while adding flake8 and pylint, we've to add some exceptions, which should be reviewed.
We treat all the requirements as exact versions using `==`. Sometimes this leads to conflicts since originally it could be `
Save/load files with DVC. Question: should this be part of core or should we make it a plugin to facilitate creation of other storage extensions?