arXiv：cs.CL· Shenbin Qian, David Samuel, Jind\v{r}ich Helcl, Andrey Kutuzov, Stephan Oepen·· 10 小时前
如何评估你不知如何评估的场景？为大语言模型基座评测选择可靠设置
How to Evaluate When You Don't Know How to Evaluate? Choosing Reliable Setups for Base Language Model Benchmarking
AI 导读
针对 LLM 评测设计决策缺乏指导的问题，研究考察了任务形式化、上下文示例数量等 8 个方面，用单调性、非随机性和中间 checkpoint 排序一致性三项信号准则作为规范元度量进行比较。研究在 26 个 2B 参数单语模型、80 项任务、4 种语言上评估，并检验了三个 ≥7B 的更大 LLM，发现多提示词等选择始终有效，而任务形式化等则取决于模型规模。
正文
Abstract:Evaluation of large language models (LLMs) remains largely pre-scientific: it involves many design decisions, with little guidance on how to make them. These choices can substantially affect measured performance and, consequently, decisions about subsequent pretraining. To identify effective evaluation setups, we investigate 8 aspects, such as task formulation or the number of in-context examples. We compare the setups using three widely-used signal criteria, repurposed as normative meta-measures: monotonicity, non-randomness, and ranking consistency of the evaluation scores across intermediate checkpoints. In an early pretraining scenario, we evaluate these setups on 26 monolingual 2B-parameter models trained on different types of data, using 80 tasks in four languages. We also examine the setups on three larger LLMs ($\geq$7B) in the same four languages and explore how the preferred setups change with model scale. Our results yield concrete recommendations: some choices, such as the use of multiple prompts, are consistently useful, whereas others, such as task formulation, depend on model scale. More generally, our framework serves as a recipe for the choice of evaluation setups in new scenarios.
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.19185 [cs.CL] |
| (or arXiv:2610.19185v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.19185 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Shenbin Qian [view email]
[v1]
Tue, 13 Oct 2026 20:50:28 UTC (247 KB)
来源：arXiv：cs.CL · arxiv.org