Overview

  • Large Language Models (LLMs) and Vision-and-Language Models (VLMs) are evaluated across a wide array of benchmarks, which test their abilities in language understanding, reasoning, coding, and multimedia understanding (in case of VLMs).
  • These benchmarks are crucial for the development of AI models as they provide standardized challenges that help identify both strengths and weaknesses, driving improvements in future iterations.
  • This primer offers an overview of these benchmarks, attributes of their datasets, and relevant papers.

Large Language Models (LLMs)

General Benchmarks

Language Understanding and Reasoning

Contextual Comprehension

Knowledge and Reasoning

Specialized Knowledge and Skills

  • MMLU (Massive Multitask Language Understanding): Assesses model performance across a broad range of subjects and task formats to test general knowledge. Introduced in “Massive Multitask Language Understanding: A Benchmark for Measuring General Linguistic Intelligence”.
    • Dataset Attributes: Covers 57 tasks across subjects like humanities, STEM, and social sciences, requiring broad and specialized knowledge.
  • MMLU-Pro (Massive Multitask Language Understanding Pro): A robust and challenging dataset designed to rigorously benchmark large language models’ capabilities. With 12K complex questions across various disciplines, it enhances evaluation complexity and model robustness by increasing options from 4 to 10, making random guessing less effective. Unlike the original MMLU’s knowledge-driven questions, MMLU-Pro focuses on more difficult, reasoning-based problems, where chain-of-thought (CoT) results can be 20% higher than perplexity (PPL). This increased difficulty results in more consistent model performance, as seen with Llama-2-7B’s variance of within 1%, compared to 4-5% in the original MMLU. Introduced in Hugging Face: MMLU-Pro.
    • Dataset Attributes: 12K questions with 10 options each. Sources include Original MMLU, STEM websites, TheoremQA, and SciBench. Covers disciplines such as Math, Physics, Chemistry, Law, Engineering, Health, Psychology, Economics, Business, Biology, Philosophy, Computer Science, and History. Focus on reasoning, increased problem difficulty, and manual expert review by a panel of over ten experts.
  • GSM8K (Grade School Math 8K): A benchmark for evaluating the reasoning capabilities of models through grade school level math problems. Found in “Can Language Models do Grade School Math?”.
    • Dataset Attributes: Consists of arithmetic and word problems typical of elementary school mathematics, emphasizing logical and numerical reasoning.
  • HumanEval: Tests models on generating code snippets to solve programming tasks, evaluating coding abilities. Proposed in “Evaluating Large Language Models Trained on Code”.
    • Dataset Attributes: Programming problems requiring synthesis of function bodies, testing understanding of code logic and syntax.
  • Physical Interaction Question Answering (PIQA): Evaluates understanding of physical properties through problem-solving scenarios. Introduced in “PIQA: Reasoning about Physical Commonsense in Natural Language”.
    • Dataset Attributes: Focuses on questions that require reasoning about everyday physical interactions, pushing models to understand and predict physical outcomes.
  • Social Interaction Question Answering (SIQA): Tests the ability of models to navigate social situations through multiple-choice questions. Described in “Social IQa: Commonsense Reasoning about Social Interactions”.
    • Dataset Attributes: Challenges models with scenarios involving human interactions, requiring understanding of social norms and behaviors.

Mathematical and Scientific Reasoning

  • MATH: A comprehensive set of mathematical problems designed to challenge models on various levels of mathematics. Proposed in “Measuring Mathematical Problem Solving With the MATH Dataset”.
    • Dataset Attributes: Contains complex, multi-step mathematical problems from various branches of mathematics, requiring advanced reasoning and problem-solving skills.

Medical Benchmarks

  • In the medical/biomedical field, benchmarks play a critical role in evaluating the ability of AI models to handle domain-specific tasks such as clinical decision support, medical image analysis, and processing of biomedical literature. Here’s an expanded overview of common benchmarks in these areas, including additional benchmarks and the attributes of their datasets, along with references to the original papers where these benchmarks were proposed: Here’s the grouping of the medical benchmarks based on the specific tasks and challenges they address within the medical domain:

Clinical Decision Support and Patient Outcomes

  • MIMIC-III (Medical Information Mart for Intensive Care): A widely used dataset comprising de-identified health data associated with over forty thousand patients who stayed in critical care units. This dataset is used for tasks such as predicting patient outcomes, extracting clinical information, and generating clinical notes.
    • Dataset Attributes: Includes notes, lab test results, vital signs, and more, requiring understanding of medical terminology and clinical narratives.
    • Reference: “The MIMIC-III Clinical Database”

Biomedical Question Answering

  • BioASQ: A challenge for testing biomedical semantic indexing and question answering capabilities. The tasks include factoid, list-based, yes/no, and summary questions based on biomedical research articles.
  • MedQA (USMLE): A question answering benchmark based on the United States Medical Licensing Examination, which assesses a model’s ability to reason with medical knowledge under exam conditions.
  • MultiMedQA: A benchmark collection that integrates multiple datasets for evaluating question answering across various medical fields, including consumer health, clinical medicine, and genetics.
  • PubMedQA: A dataset for natural language question answering using abstracts from PubMed as the context, focusing on yes/no questions.
  • MedMCQA: A medical multiple-choice question answering benchmark that evaluates comprehensive understanding and application of medical concepts.

Biomedical Language Understanding

  • BLUE (Biomedical Language Understanding Evaluation): A benchmark consisting of several diverse biomedical NLP tasks such as named entity recognition, relation extraction, and sentence similarity in the biomedical domain.

Code LLM Benchmarks

  • In the domain of code synthesis and understanding, benchmarks play a pivotal role in assessing the performance of Code LLMs. These benchmarks challenge models on various aspects such as code generation, understanding, and debugging. Here’s a detailed overview of common benchmarks used for evaluating code LLMs, including the attributes of their datasets and references to the original papers where these benchmarks were proposed:

Code Generation and Synthesis

  • HumanEval: This benchmark is designed to test the ability of language models to generate code. It consists of a set of Python programming problems that require writing function definitions from scratch.
  • MBPP (Mostly Basic Python Problems): A benchmark consisting of simple Python coding problems intended to evaluate the capabilities of code generation models in solving basic programming tasks.

Code Debugging and Error Detection

  • DS-1000 (DeepSource Python Bugs Dataset): This dataset is used to evaluate the ability of models to detect bugs in Python code. It includes a diverse set of real-world bugs.
    • Dataset Attributes: Comprises 1000 annotated Python functions with detailed bug annotations, testing models on their ability to identify and understand common coding errors.
    • Reference: [No direct paper; part of DeepSource tooling offerings, information can be typically found on the DeepSource website or technical blogs associated with the tool.]

Comprehensive Code Understanding and Multi-language Evaluation

Algorithmic Problem Solving

  • APP (Algorithms Problems from Programming): Focuses on algorithmic problem-solving skills by presenting problems typically found in programming competitions.
    • Dataset Attributes: Challenges models with complex algorithmic questions that require not only the correct code but also efficiency and optimization.
    • Reference: [Specific paper might not exist; generally sourced from programming contest sites and similar venues.]

Vision-Language Models (VLMs)

General Benchmarks

  • VLMs are pivotal in AI research as they combine visual data with linguistic elements, offering insights into how machines can interpret and generate human-like responses based on visual inputs. This section delves into key benchmarks that test these hybrid capabilities:

Visual Question Answering

  • Visual Question Answering (VQA) and VQAv2: Requires models to answer questions about images, testing both visual comprehension and language processing. Described in “VQA: Visual Question Answering” and its subsequent updates.
    • Dataset Attributes: Combines real and abstract images with questions that require understanding of object properties, spatial relationships, and activities.

Image Captioning

  • MSCOCO Captions: Models generate captions for images, focusing on accuracy and relevance of the visual descriptions. Introduced in “Microsoft COCO: Common Objects in Context”.
    • Dataset Attributes: Real-world images with annotations requiring descriptive and detailed captions that cover a broad range of everyday scenes and objects.

Visual Reasoning

  • NLVR2 (Natural Language for Visual Reasoning for Real): Evaluates reasoning about the relationship between textual descriptions and image pairs. Proposed in “A Corpus for Reasoning About Natural Language Grounded in Photographs”.
    • Dataset Attributes: Pairs of photographs with text statements that models must verify, focusing on logical reasoning across visually disparate images.
  • MMMU (MultiModal MultiTask Understanding): Tests models’ ability to understand and generate responses based on both visual and textual stimuli. Introduced in the paper [Title not available].
    • Dataset Attributes: Involves tasks like visual question answering, image captioning, and visual reasoning, testing both visual and textual understanding.

Video Understanding

  • Perception Test: A benchmark designed to evaluate models on understanding and interpreting video content. Detailed in “Perception Test: Benchmark for Autonomous Vehicle Perception”.
    • Dataset Attributes: Video sequences requiring models to interpret dynamic scenes, focusing on object detection, movement prediction, and scene classification.

Medical VLM Benchmarks

  • Medical VLMs are essential in merging AI’s visual and linguistic analysis for healthcare applications. They are pivotal for developing systems that can interpret complex medical imagery alongside textual data, enhancing diagnostic accuracy and treatment efficiency. This section explores major benchmarks testing these interdisciplinary skills:

Medical Image Annotation and Retrieval

  • ImageCLEFmed: Part of the ImageCLEF challenge, this benchmark tests image-based information retrieval, automatic annotation, and visual question answering using medical images.

Disease Classification and Detection

  • CheXpert: A large dataset of chest radiographs for identifying and classifying key thoracic pathologies. This benchmark is often used for tasks that involve reading and interpreting X-ray images.
  • Diabetic Retinopathy Detection: Focused on the classification of retinal images to diagnose diabetic retinopathy, a common cause of vision loss.
    • Dataset Attributes: Features high-resolution retinal images, where models need to detect subtle indicators of disease progression, requiring high levels of visual detail recognition.
    • Reference: [Typically linked with the Kaggle competition for diabetic retinopathy detection, more formal academic references may vary.]

Common Challenges Across Benchmarks

  • Generalization: Assessing how well models can generalize from the training data to unseen problems.
  • Robustness: Evaluating the robustness of models against edge cases and unusual inputs.
  • Execution Correctness: Beyond generating syntactically correct code, the emphasis is also on whether the code runs correctly and solves the problem as intended.
  • Bias and Fairness: Ensuring that models do not inherit or perpetuate biases that could impact patient care outcomes, especially given the diversity of patient demographics.
  • Data Privacy and Security: Addressing concerns related to the handling and processing of sensitive health data in compliance with regulations such as HIPAA.
  • Domain Specificity: Handling the high complexity of medical and biomedical terminologies and imaging, which requires not only technical accuracy but also clinical relevancy.

Citation

If you found our work useful, please cite it as:

@article{Chadha2020DistilledLLMVLMBenchmarks,
  title   = {LLM/VLM Benchmarks},
  author  = {Chadha, Aman},
  journal = {Distilled AI},
  year    = {2020},
  note    = {\url{https://aman.ai}}
}