Specific metrics from geval (coherence, fluency) may also apply to Translation. (The caveat is the target language may not be well supported by the judge model, but this is something we could leave up to the user to understand - or we could highlight via UI or documentation.)
With this issue we'd at least give the option of running geval in translation workflows, which is now not possible. It involves:
Specific metrics from geval (coherence, fluency) may also apply to Translation. (The caveat is the target language may not be well supported by the judge model, but this is something we could leave up to the user to understand - or we could highlight via UI or documentation.)
With this issue we'd at least give the option of running geval in translation workflows, which is now not possible. It involves: