IBM Technology
6 minutes
Grading Model Output with a Model-Based Grader
Use a second model call as the judge when the correct answer is open-ended, writing a rubric the grader applies consistently, and knowing where LLM-as-judge is unreliable.