Number Cookbook: Number Understanding of Language Models and How to Improve It
Paper
•
2411.03766
•
Published
•
1
The dataset viewer is not available for this split.
Rows from parquet row groups are too big to be read: 9.15 GiB (max=286.10 MiB)
Error code: TooBigContentError
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
This repository contains the dataset used in the paper Number Cookbook: Number Understanding of Language Models and How to Improve It.
Code: https://github.com/GraphPKU/number_cookbook.
The dataset includes four number representations (Integer, Float, Fraction, Scientific Notation) and 17 tasks. The details about the benchmark can be found in the paper.
- Downloads last month
- 23
Size of the auto-converted Parquet files (First 5GB per split):
2.28 GB
Number of rows (First 5GB per split):
220
Estimated number of rows:
209