metadata
license: llama2
datasets:
- meta-math/MetaMathQA
- gsm8k
language:
- en
pipeline_tag: text-generation
This model is a reproduction of Meta-Math. We follow the training instructions described in the original repo as much as possible with minimal modifications.
Prompt
Unlike the original Meta-math model, we do not use any instructions.
"""
Problem: {query_problem}\nSolution: {gt_solution}
"""
The {gt_solution} is provided in the training stage, whereas it is empty string in the evaluation stage.