nlpxucan / WizardLM

LLMs build upon Evol Insturct: WizardLM, WizardCoder, WizardMath
9.19k stars 712 forks source link

Where I can download the math_instruction_data.json? #158

Closed w32zhong closed 1 year ago

w32zhong commented 1 year ago

Without the data math_instruction_data.json, how can I reproduce your evaluation?

flyinghpluo commented 1 year ago

Thank you for your interest in our work, our training data is currently not publicly available. If you want to evaluate the GSM8k or MATH benchmarks, please refer to the "Evalution" section of the WizardMath README.

w32zhong commented 1 year ago

Thank you for your response.

I want to evaluate the model, but given that the data is not available, I hope we can know at least on which datasets it is trained, do we need to train it on MATH training set? is it zero-shot? For it to be used in a research evaluation, I think these are important information.

flyinghpluo commented 1 year ago

Due to company policies,Our training data is currently not publicly available. You can learn about in detail how our training data are constructed through our paper https://arxiv.org/abs/2308.09583

w32zhong commented 1 year ago

Thanks for your notice and sharing your paper. I will close this issue.

EmiyaArcher233 commented 1 year ago

Hello, when can we get math_instruction_data.json? We want to reproduce your evaluation, but we don't have the necessary data. Could you release the dataset?