Skip to content

Include evals report averages in span attributes #3053

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 5 commits into from
Oct 21, 2025
Merged

Include evals report averages in span attributes #3053

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
7fdab58
Include report averages
cetra3 Oct 1, 2025
53b4af4
Adjust test
cetra3 Oct 1, 2025
1c1c804
Merge remote-tracking branch 'origin/main' into include_report_averages
cetra3 Oct 20, 2025
52af873
Adjust to be under `logfire.` prefix
cetra3 Oct 20, 2025
2002a26
Merge branch 'main' into include_report_averages
DouweM Oct 21, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 2 additions & 0 deletions pydantic_evals/pydantic_evals/dataset.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -324,6 +324,8 @@ async def _handle_case(case: Case[InputsT, OutputT, MetadataT], report_case_name
trace_id=trace_id,
)
if (averages := report.averages()) is not None and averages.assertions is not None:
experiment_metadata = {'n_cases': len(self.cases), 'averages': averages}
Copy link
Collaborator

@DouweM DouweM Oct 1, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

From the test below, it looks like we already had n_cases in here, do we need to repeat it? Or could we drop it there and have it just under the metadata?

eval_span.set_attribute('experiment.metadata', experiment_metadata)
Copy link
Collaborator

@DouweM DouweM Oct 1, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can you add a logfire. prefix here since it's not general OTel but specific to our platform?

Copy link
Contributor Author

@cetra3 cetra3 Oct 2, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'd argue it's specific to pydantic evals, rather than logfire itself?

We use the experiment.metadata field to ingest information about the test run. Are you saying we should change this to logfire.experiment.metadata ? or are you saying underneath experiment.metadata we should have a logfire field.

Copy link
Collaborator

@DouweM DouweM Oct 3, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@cetra3 Yeah I was suggesting (based on @alexmojaki suggestion in Slack) to change it to logfire.experiment.metadata. But I see your (implied) point that even though right now only Logfire will use these attributes, they could be useful to users sending pydantic_evals data to different platform as well.

@alexmojaki What do you think about a pydantic_evals prefix?

Copy link
Contributor

@alexmojaki alexmojaki Oct 3, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Everything here is specific to pydantic evals. This is particularly specific not just to logfire but to fusionfire popped attributes. n_cases is being repeated here so that it can be displayed cheaply in the frontend. We may add or remove things here freely depending on what the frontend needs, it's not intended to be queried. I think logfire. makes sense for that.

eval_span.set_attribute('assertion_pass_rate', averages.assertions)
return report

Expand Down
28 changes: 27 additions & 1 deletion tests/evals/test_dataset.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1496,10 +1496,36 @@ async def mock_async_task(inputs: TaskInput) -> TaskOutput:
'assertion_pass_rate': 1.0,
'logfire.msg_template': 'evaluate {name}',
'logfire.msg': 'evaluate mock_async_task',
'experiment.metadata': {
'n_cases': 2,
'averages': {
Copy link
Collaborator

@DouweM DouweM Oct 1, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If we're going to be using this on the frontend, I don't know if it's wise to pass along the entire ReportCaseAggregate object or just the subset we need

'name': 'Averages',
'scores': {'confidence': 1.0},
'labels': {},
'metrics': {},
'assertions': 1.0,
'task_duration': 1.0,
'total_duration': 9.0,
},
},
'logfire.span_type': 'span',
'logfire.json_schema': {
'type': 'object',
'properties': {'name': {}, 'n_cases': {}, 'assertion_pass_rate': {}},
'properties': {
'name': {},
'n_cases': {},
'experiment.metadata': {
'type': 'object',
'properties': {
'averages': {
Copy link
Collaborator

@DouweM DouweM Oct 1, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'd expect n_cases to also be listed here

Copy link
Contributor

@alexmojaki alexmojaki Oct 21, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

That's normal, schemas for primitives only appear at the top level.

'type': 'object',
'title': 'ReportCaseAggregate',
'x-python-datatype': 'PydanticModel',
}
},
},
'assertion_pass_rate': {},
},
},
},
),
Expand Down