Stable version of phoenix

[rajabhupati]

Do you have a stable version flag for releases? We observed inconsistencies while running through versions?

[dosubot[bot]]

Hey @rajabhupati! 🚀 Long time no see in the digital corridors of code! How's everything on your end?

Yes, the Arize Phoenix repository does follow a specific versioning system, which is Semantic Versioning. This is indicated by the version format "3.17.1", which aligns with the Semantic Versioning pattern of MAJOR.MINOR.PATCH. This format is typically used to indicate stable releases.

In terms of distinguishing stable versions from unstable or beta versions, the repository uses tags to indicate the stability of a release. This is evident from the GitHub Actions workflow file, specifically in the build job with the condition if: startsWith(github.ref, 'refs/tags/'). This condition means that the publishing to PyPI process is triggered only when a tag is pushed to the repository. Tags in Git are often used to mark specific points in a repository's history as important, typically for releases. Therefore, the use of tags in this context suggests they are used to indicate stable releases that are ready for distribution.

However, I wasn't able to find specific information about the release cycle of the Arize Phoenix repository or how the stable versions are distinguished from the unstable or beta versions within the repository. If you could provide more details about the inconsistencies you observed, it might help in providing a more accurate response.

You can refer to the following sources for more information:

• src/phoenix/version.py
• CHANGELOG.md
• .github/workflows/publish-to-pypi.yml

I hope this helps! If you have any more questions or need further clarification, feel free to ask.

Sources

• src/phoenix/version.py
• CHANGELOG.md
• .github/workflows/publish-to-pypi.yml

About Dosu

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.

[mikeldking]

@rajabhupati sorry that you were experiencing inconsistencies. Can you describe the stability issues you've been dealing with? In general, we follow, semantic, versioning. However, if inconsistencies occur, it can be due to external factors such as LLM orchestration layers and providers. Let us know.

[gourav1298]

Hi, we used the image versions till 3.12.0 and they were not running on EKS deployment and were always giving some type of issue. We were only able to use the image 3.4.1 which is running perfectly fine right now but its doesn't show to token count and other things that it was showing earlier.

[axiomofjoy]

@gourav1298 Thanks for the report. If you hit any further issues with the Docker image, please let us know. Can you also tell us, which instrumentation are you using that you don't see token counts for?

[arizedatngo]

@mikeldking speaking to @rajabhupati right now

Docs / Code should be aligned and stable. @mikeldking will speak later today on this with you

[mikeldking]

@rajabhupati I think I get what you mean by stable in terms of docs. Right now we only have mechanisms by which to have our docs be sync'd with our releases so it always represents the latest but you can always download the docs for a previous version via our releases. https://github.com/Arize-ai/phoenix/releases

[mikeldking]

@gourav1298 sorry you are facing issues with token counts - though I would be pretty surprised if you had token counts and then they did not show up in a newer version. Can you give us more details on your setup (LLM, instrumentation, etc.)?

[gourav1298]

Hi @mikeldking & @axiomofjoy as I stated earlier I was using the version 3.4.1, in which the token count is not showing and we are mainly using Langchain Instrumenter right now, also we tried for newer versions till 3.12.0 but that was giving the following error "Cannot query field 'streamingLastUpdatedAt' on type 'Query'". The current deployment is running in EKS and exposed via a route53 endpoint backed by a load balancer.

Also just wanted to have an idea whether arize-phoenix has a storage backend that we can declare for storing all the traces.

[mikeldking]

@gourav1298

"Cannot query field 'streamingLastUpdatedAt' on type 'Query'". This is just a caching miss on our end from what I can tell. We have this fixed in later versions. Can you try resetting your browser cache for phoenix? https://support.google.com/accounts/answer/32050?hl=en&co=GENIE.Platform%3DDesktop

Also just wanted to have an idea whether arize-phoenix has a storage backend that we can declare for storing all the traces. Yes, it's coming and is top priority for us. We need a bit more time however so right now we recommend downloading and storing your data periodically.

[gourav1298]

hi @mikeldking, Can you specify which is the latest version that I can deploy for my use?

[mikeldking]

@gourav1298 all git tags correspond to a release (both python and containers): https://github.com/Arize-ai/phoenix/releases https://hub.docker.com/r/arizephoenix/phoenix/tags

[gourav1298]

Hi @mikeldking, the latest image is running but the token count is still not showing.

[mikeldking]

Hey @gourav1298 this is probably un-related to the version of phoenix but rather that not all LLM providers expose tokens when streaming OR in general. We are in the process of building a token counting registry to address this ( https://github.com/Arize-ai/phoenix/issues/2519)

Can you tell us a bit more about the LLM you are using? We may be able to unblock you depending on your deployment.

[gourav1298]

We are using databricks serving endpoint.

[mikeldking]

We are using databricks serving endpoint.

Gotcha. We will have to investigate. Thanks for the info.