Matt Rickard

Images not cleaned up properly

3

I'm not sure if this actually affects functionality, but I think these images should be cleaned up even if the machine fails to start somehow. This is from our jenkins...

Update examples

2

The examples are out of date according the new API /kind bug /kind documentation

Preprocessor context generation respects dockerignore

2

Currently we build the context manually and do not respect the dockerignore file.

Publish prediction notebook example for Kubeflow

1

https://github.com/kubeflow/fairing/tree/master/examples/prediction but in notebook form

Chain preprocessors

1

It might be interesting to chain multiple pre-processors together e.g. ConvertNotebookPreprocessor -> FunctionPreProcessor ->

* Vertices should be `type Vertex interface{}` instead of string * Sort should return `[][]Vertex` instead of `[]string`, grouping tasks that can be done in parallel together. This would support...

Add OpenLM LLM multi-provider

1

OpenLM is a zero-dependency OpenAI-compatible LLM provider that can call different inference endpoints directly via HTTP. It implements the OpenAI Completion class so that it can be used as a...

Better Error Handling for /v1/completion/typed

1

Better error handling. Especially for TypeScript definition bugs. It's hard (impossible) to know how to fix something right now if it's broken (e.g., interface instead of type, missing syntax, etc.)...

Namespace support

1

Namespaces aren't currently supported. We'll have to extract the namespace flag and make sure that it is piped through to the command to get the latest resource, as well as...

Other commands should be able to be implemented. The trickier part here is that each command has slightly different syntax and some work on all resources (label, delete, patch), while...