You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In new benchmarks are are no longer offering a "demo" test that returns annotations. Instead, there annotations will be returned with practice benchmarks. However, this is a flag to be sent in the SKU:
practice set with no annotations sent: airr_safety_1_2_en_us_practice
practice set with annotations: airr_safety_1_2_en_us_annotation
So "annotation" means "practice with annotations returned".
Hi @bollacker , I'd be interested in implementing this.
I noticed that ModelBench already supports annotation export through --annotations and BenchmarkRun.compile_annotations().
Is this issue primarily about mapping the airr_safety_1_2_en_us_annotation SKU to the existing annotation-export behavior, or does it require changes to the evaluator service as well?
I'd be happy to take on the ModelBench changes if they can be implemented and tested locally.
Activity
confiscate commentedon Oct 29, 2026
Hi @bollacker , I'd be interested in implementing this.
I noticed that ModelBench already supports annotation export through
--annotationsandBenchmarkRun.compile_annotations().Is this issue primarily about mapping the
airr_safety_1_2_en_us_annotationSKU to the existing annotation-export behavior, or does it require changes to the evaluator service as well?I'd be happy to take on the ModelBench changes if they can be implemented and tested locally.