chore(weave): capture request id for openai completions #3153
Conversation
|
Preview this PR with FeatureBee: https://beta.wandb.ai/?betaVersion=4efabd9d7845b7b0d0f661564ba87b117aaf7e57 |
bcsherma
force-pushed
the
openai-request-id
branch
from
9f58d95 to
d98cafc
Compare
This stack of pull requests is managed by Graphite. Learn more about stacking. |
bcsherma
force-pushed
the
openai-request-id
branch
from
d98cafc to
7f917fa
Compare
| @@ -64,7 +64,7 @@ def test_instructor_openai( | |||
| assert op_name_from_ref(call.op_name) == "openai.chat.completions.create" | |||
| output = call.output | |||
| output_arguments = json.loads( | |||
| output.choices[0].message.tool_calls[0].function.arguments | |||
| output["choices"][0]["message"]["tool_calls"][0]["function"]["arguments"] | |||
There was a problem hiding this comment.
why do you have to change from dot to bracket notation?
There was a problem hiding this comment.
The output here comes from the post processor I'm modifying. In practice value can have many types depending on integrations, sdk versions, and I think even the type of completion being requested. In some cases the post processor returns a pydantic object, in some cases it returns a dict. For the standard case where we receive a ChatCompletion we will now return a dict with this extra key. The alternative would be adding the attribute to the pydantic class which feels hackier, and ultimately I think the output just gets dumped later in our pipe anyway.
bcsherma
force-pushed
the
openai-request-id
branch
from
7f917fa to
e7df7c2
Compare
Description
Openai integration captures the request id from completion responses and adds them to
request_idon the call output.