Hi there, I tried all models but I keep running into this error when running evals on the platform.
**
Unknown parameter: ‘data_source.sampling_params.text.verbosity’.**
Hey @yanist, that’s definitely a confusing one to run into, especially after trying multiple models. From what we’ve seen, that error usually points to a parameter mismatch somewhere in the eval config rather than the model itself.
data_source.sampling_params.text.verbosity doesn’t seem to be recognized on the current API surface, so there may be an older config/example getting passed through somewhere.
Is this still happening for you? If yes, which API surface are you using for the evals right now (Responses API, Chat Completions, SDK, dashboard evals, etc.)? That’ll help narrow down where the unsupported param is coming from.
-Mark G.
I’m seeing this error as well. I’m using Dashboard evals only — not the API/SDK directly.
Setup:
- New eval created from scratch in the OpenAI Platform Dashboard
- CSV upload as test data
- “Generate new responses” enabled
- Text columns only; no image columns
- No “Import as responses” selected
- Simple run with a developer/user prompt
- Semantic similarity grader
- I’ve tried recreating the eval and simplifying the CSV/prompt
The error appears when trying to run/review the eval:
Unknown parameter: data_source.sampling_params.text.verbosity
I did not intentionally set a verbosity parameter anywhere in the UI. It looks like the Dashboard may be injecting `data_source.sampling_params.text.verbosity` into the eval run config, but that path is not accepted by the current Evals API surface.
Hi Yanis! Have you found a workaround yet? I’m getting the same error now