PandaLM: An Automatic Evaluation Benchmark for LLM Instruction Tuning Optimization
By Yidong Wang, Zhuohao Yu, Zhengran Zeng, Linyi Yang, Cunxiang Wang, Hao Chen, Chaoya Jiang, Rui Xie, Jindong Wang, Xingxu Xie and 3 more. 2023
400 citations recorded by the source corpus.
Instruction tuning large language models (LLMs) remains a challenging task, owing to the complexity of hyperparameter selection and the difficulty involved in evaluating the tuned models. To determine the optimal hyperparameters, an automatic, robust, and reliable evaluation benchmark is essential. However, establishing such a benchmark is not a trivial task due to the challenges associated with evaluation accuracy and privacy protection. In response to these challenges, we introduce a judge large language model, named PandaLM, which is trained to distinguish the superior model given several LLMs. PandaLM's focus extends beyond just the objective correctness of responses, which is the main focus of traditional evaluation datasets. It addresses vital subjective factors such as relative conciseness, clarity, adherence to instructions, comprehensiveness, and formality. To ensure the reliability of PandaLM, we collect a diverse human-annotated test dataset, where all contexts are generated by humans and labels are aligned with human preferences. Our results indicate that PandaLM-7B achieves 93.75% of GPT-3.5's evaluation ability and 88.28% of GPT-4's in terms of F1-score on our test dataset. PandaLM enables the evaluation of LLM to be fairer but with less cost, evidenced by significant improvements achieved by models tuned through PandaLM compared to their counterparts trained with default Alpaca's hyperparameters. In addition, PandaLM does not depend on API-based evaluations, thus avoiding potential data leakage. All resources of PandaLM are released at https://github.com/WeOpenML/PandaLM.
Subfields: Artificial Intelligence, Large Language Models, Alignment & RLHF, Evaluation & Benchmarks
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Graham Neubig
- Luke S. Zettlemoyer
- Danqi Chen
- Mohit Bansal
- Yejin Choi
- Tianyu Gao
- Sean Welleck
- Yu Meng
In Large Language Models
In Alignment & RLHF
In Evaluation & Benchmarks
Citations in the corpus
- On LLMs-Driven Synthetic Data Generation, Curation, and Evaluation: A Survey (2024, neutral): “One of the greatest advantages of LLMs for synthetic data generation is their instruction-following capability, which contributes to great controllability (Wang et al., 2023c; Radford et al., 2019).”
- ChatEval: Towards Better LLM-based Evaluators through Multi-Agent Debate (2023, methodology): “Wang et al. (2023c) curate a reliable dataset containing pairwise comparison and evaluation explanation which can be used to train a foundation model making it a better evaluator.”
- Prometheus 2: An Open Source Language Model Specialized in Evaluating Other Language Models (2024, background): “…prompted to output a scalar indicator of quality (denoted as direct assessment ) (Zheng et al., 2023; Liu et al., 2023b; Ye et al., 2023; Kim et al., 2023) or to determine which of two outputs are preferred (denoted as pairwise ranking ) (Wang et al., 2023b; Li et al., 2023b; Lambert et al., 2024).”
- Evaluating Large Language Models at Evaluating Instruction Following (2023, methodology): “Thus, employing open-source LLMs as evaluators can be a promising substitute (Zheng et al., 2023; Wang et al., 2023c).”
- Style Over Substance: Evaluation Biases for Large Language Models (2023, neutral): “As human evaluations can be costly, some recent works utilize state-of-the-art LLMs to evaluate model outputs (Wang et al., 2023b; Kim et al., 2023).”
- Prometheus: Inducing Fine-grained Evaluation Capability in Language Models (2023, background): “Wang et al. (2023b) introduced PandaLM, a fine-tuned LLM to evaluate the generated text and explain its reliability on various preference datasets.”
- Unleashing the potential of prompt engineering in Large Language Models: a comprehensive review (2023, neutral): “There are some models that are increasingly used to grade the output of other models, which aim to ‘check’ the ability of other models [105, 106].”
- JudgeLM: Fine-tuned Large Language Models are Scalable Judges (2023, neutral): “Rosslyn Mountain Boys was a band that backed up a number of nationally known singers, and released two albums of their own.”
- CValues: Measuring the Values of Chinese Large Language Models from Safety to Responsibility (2023, background): “Our experiments aim to evaluate a wide range of LLMs with raw safety prompts and responsibility prompts and analyze their performance by human annotation.”
- AutoSurvey2: Empowering Researchers with Next Level Automated Literature Surveys (2025, unknown): “Third, the lack of standardized evaluation protocols for LLM-generated academic content (Wang et al., 2024d; Yu et al., 2024; Zheng et al., 2024) makes quality assurance difficult, as expert evaluation is costly, subjective, and hard to scale.”
- AutoSurvey: Large Language Models Can Automatically Write Surveys (2024, unknown): “Relying on human review for quality assessment is not only resource-intensive but also lacks scalability [21–23].”
- Bloom-Eval: A Hierarchical Evaluation Benchmark for Automatic Survey Generation Based on Bloom's Taxonomy (2026, unknown): “While demonstrating the potential of LLM-based evaluation, recent studies (Wang et al., 2024b; Zheng et al., 2023) also reveal that LLM judges remain prone to inherent biases when assessing subjective metrics.”