This dataset is an augmented version of the GSM8K dataset. It extends the original GSM8K training set to 385k samples by prompting GPT-4. The dataset was originally proposed in paper "From Explicit CoT to Implicit CoT: Learning to Internalize CoT Step by Step".
This specific version is used in the paper Parallel Continuous Chain-of-Thought with Jacobi Iteration.
Code:
https://github.com/whynlp/pccotLoad the dataset using the datasets library:
from… See the full description on the dataset page:
https://huggingface.co/datasets/whynlp/gsm8k-aug-nl.
