AI and LLMs · Senior · Question 54 of 75
How can you make LLM eval results reproducible?
- 1Use the latest model alias so results stay current
- 2Run evals only once and never change the dataset
- 3Pin model and judge versions; store prompts and data
Show the answer
Correct answer: Pin model and judge versions; store prompts and data
Explanation
Without versioned inputs, a score change can't be traced to a cause.