OpenAI Responses and Compaction APIs #4042
Conversation
github-actions
bot
assigned
codefromthecrypt,
karthikscale3,
lmolkova,
lzchen and
nirga
Complete testing, wrapping and coverage for the latest OpenAI responses API and Compaction API features in the python client.
There were a few assumptions being made about openai types in the tests and the openai wrappers that raised linting errors. Some of these errors were rightly raised, especially around streaming types, and have been patched.
iamemilio
force-pushed
the
responses_openai
branch
from
20e4e2b to
af4f76a
Compare
|
@xrmx I noticed you were making some improvements to the OpenAI v2 library. I was wondering if you could give me a little guidance on this PR. I recognize that the size is fairly large, so I am open to breaking it up into smaller changes if needed. I especially would appreciate guadance around the semantic conventions, and the OpenTelemetry GenAI input/output message schemas:
|
xrmx
left a comment
xrmx
left a comment
There was a problem hiding this comment.
Thanks for the PR. I would say to make the PR more reviewable open a preparatory PR with just the moving of the code outside of patch and whatever other change to the current code. Did you already get in touch with the genai people? If not please hop on the otel-genai-instrumentation channel in slack and share your work, SIG calls won't be run until next year.
| "Programming Language :: Python :: 3.13", | ||
| ] | ||
| dependencies = [ | ||
| "openai>=1.109.1", |
There was a problem hiding this comment.
Why? If it is the first version having Responses the instrumentation should continue to work with older versions.
There was a problem hiding this comment.
+1 this is not a dependency. There is a lot of junk in this PR, its a very rough draft and I am still getting my bearings, but its clearly doing far too much. I expect to cut out a majority of this.
| unwrap(openai.resources.embeddings.AsyncEmbeddings, "create") | ||
| def _uninstrument(self, **kwargs): | ||
| chat_mod = importlib.import_module("openai.resources.chat.completions") | ||
| unwrap(chat_mod.Completions, "create") |
There was a problem hiding this comment.
unwrap takes also strings as the wrap functions, you shouldn't need the importlib usage
| ) | ||
|
|
||
| # Re-export StreamWrapper for backwards compatibility | ||
| __all__ = ["StreamWrapper"] |
There was a problem hiding this comment.
instrumentations are not APIs, I don't think we need to care about this unless we have known users
If you want to implement the newer semantic conventions for genai events you have broadly two choices:
I suggest the first option so the PR adding responses would be hopefully smaller. |
|
I have not gotten in touch with them yet, but I will! Thank you for taking the time to review this and point me in the right direction. This work was largely exploratory, so I am more than happy to close this PR up and plan a more pragmatic approach to introducing these changes :). I will work on adding the opt in switch to gate usage of the new semantic conventions. |
7 participants
Complete testing, wrapping and coverage for the latest OpenAI responses API and Compaction API features in the python client.
NOTE This is a work in progress and I am a new contributor who would greatly appreciate any guidance maintainers think is needed.
Description
Please include a summary of the change and which issue is fixed. Please also include relevant motivation and context. List any dependencies that are required for this change.
Fixes #3436
Type of change
Please delete options that are not relevant.
How Has This Been Tested?
A number of tests have been added to cover this to make sure that it meets the following requirements:
Does This PR Require a Core Repo Change?
Checklist:
See contributing.md for styleguide, changelog guidelines, and more.