Made onxxrutime an optional dependency

[mangin]

Description of changes

Now onxxrutime is used to calculate default embedding. But onxxrutime doesn't support all operation systems (for example it doesn't work with alpine at all)

I think to replace onxxruntime to something else it's a big task, but at least we can allow to run chromadb without default embedding if onxxrutime is not present.

Summarize the changes made by this PR.

• Improvements & Bug fixes
  • Make onxxrutime optional
• New functionality
  • Added initial support of alpine

[github-actions]

Reviewer Checklist

Please leverage this checklist to ensure your code review is thorough before approving

Testing, Bugs, Errors, Logs, Documentation

• Can you think of any use case in which the code does not behave as intended? Have they been tested?
• Can you think of any inputs or external events that could break the code? Is user input validated and safe? Have they been tested?
• If appropriate, are there adequate property based tests?
• If appropriate, are there adequate unit tests?
• Should any logging, debugging, tracing information be added or removed?
• Are error messages user-friendly?
• Have all documentation changes needed been made?
• Have all non-obvious changes been commented?

System Compatibility

• Are there any potential impacts on other parts of the system or backward compatibility?
• Does this change intersect with any items on our roadmap, and if so, is there a plan for fitting them together?

Quality

• Is this code of a unexpectedly high quality (Readability, Modularity, Intuitiveness)

[jeffchuber]

@tazarov can you take a look here?

[JoshKarpel]

Just chiming in in favor of this change - we're running into a problem where we'd like to install onnxruntime-gpu, which provides an alternate implementation of the onnxruntime package, but we can't because the chromadb packages requires onnxruntime unconditionally in its requirements.

[tazarov]

@mangin, you have a point about alpine. Arguably though, there is a build for onnx that runs on musl distros, although the pypi package isn't there, which makes it a tad more difficult to run onnx and therefore Chroma on alpine.

That said, this change can be challenging to merge as it will break DX. Consider this: today you install chroma and doing something like:

collection.add(ids=[...], documents=[...])

just works.

With an optional onnxruntime, users will get an error as optional deps are not installed by default. Which creates a friction for the starting experience.

We're also in the process of transitioning away onnx-based default EF, so that would be a natural point for merging the proposed solution in this PR.

Either way, I think this needs a bit deeper consideration. So give me couple of days to do some experiments.

[JoshKarpel]

That said, this change can be challenging to merge as it will break DX. Consider this: today you install chroma and doing something like:

collection.add(ids=[...], documents=[...])

just works.

With an optional onnxruntime, users will get an error as optional deps are not installed by default. Which creates a friction for the starting experience.

Could you instead recommend pip install 'chromadb[onnx]' in your various guides? That would still give the desired UX, and then advanced users can drop the optional requirement when needed.