The Necessity of Setting Temperature in LLM-as-a-Judge
This addresses a practical problem for researchers and engineers using LLM-as-a-Judge by highlighting the need for principled temperature configuration, though it is incremental as it builds on prior empirical conventions.
The paper investigates how temperature settings affect the performance of LLM-as-a-Judge in evaluating text quality and factual correctness, finding that lower temperatures do not always yield optimal results and that effects are task-dependent, with causal analysis providing engineering insights.
LLM-as-a-Judge has emerged as an effective and low-cost paradigm for evaluating text quality and factual correctness. Prior studies have shown substantial agreement between LLM judges and human experts, even on tasks that are difficult to assess automatically. In practice, researchers commonly employ fixed temperature configurations during the evaluation process-with values of 0.1 and 1.0 being the most prevalent choices-a convention that is largely empirical rather than principled. However, recent researches suggest that LLM performance exhibits non-trivial sensitivity to temperature settings, that lower temperatures do not universally yield optimal outcomes, and that such effects are highly task-dependent. This raises a critical research question: does temperature influence judge performance in LLM centric evaluation? To address this, we systematically investigate the relationship between temperature and judge performance through a series of controlled experiments, and further adopt a causal inference framework within our empirical statistical analysis to rigorously examine the direct causal effect of temperature on judge behavior, offering actionable engineering insights for the design of LLM-centric evaluation pipelines.