Dark Mode
Logical Reasoning Improvement Dataset
Education & Learning Analytics
Tags and Keywords
Trusted By
"No reviews yet"
Free
About
This curated collection of data is specifically designed to enhance logical reasoning skills in various LLM models. It acts as a training and evaluation resource that supports models in solving a wide array of logical reasoning questions, thereby improving overall model performance. The dataset ensures high quality by eliminating redundant or overly similar questions through an advanced filtering technique.
Columns
The dataset is organised into four primary fields:
- input: The initial statement, text, or question that requires logical reasoning to solve.
- output: The correct answer or intended solution for the logical reasoning question.
- instruction: Additional specific guidelines or requirements necessary for solving the particular reasoning problem.
- data_source: Specifies the original source or origin from which the logical reasoning question was derived.
Distribution
The core data is contained within the file named
train.csv. The file size is 30.63 MB and contains four columns of text data. While the output, instruction, and data_source columns have 24.9k valid entries, the input column currently has approximately 5,155 valid entries, with a significant percentage of missing values. The content draws heavily from sources such as MATH/PRM-800K and reclor.Usage
- Training and evaluating LLM models, such as Platypus2, to sharpen their performance on logical reasoning tasks.
- Serving as a standard benchmark for researchers testing and comparing different logical reasoning algorithms and techniques.
- Creating educational materials or dedicated platforms aimed at improving user ability in logical thinking.
Coverage
The data coverage focuses purely on the domain of abstract logical reasoning questions and does not contain specific geographic, time range, or demographic information. The scope is defined by the origins of the questions, including various external logical question sets.
License (State the license URL.)
CC0 1.0 Universal (CC0 1.0) - Public Domain
Who Can Use It
- Artificial Intelligence Developers: Utilising the data to finetune or pre-train new generations of LLMs for better logical performance.
- Data Scientists: Employing the questions as a benchmark to rigorously test the efficiency and accuracy of new reasoning algorithms.
- Educational Content Creators: Leveraging the filtered questions to build reliable practice resources for students or learners.
Dataset Name Suggestions
- Logical Reasoning Improvement Dataset
- LLM Logical Reasoning Skill Enhancer
- Platypus2 Training Question Bank
- Filtered Logical Reasoning Data
Attributes
Original Data Source: Logical Reasoning Improvement Dataset
QUALITYLoading...