Unleashing Reasoning Capability of LLMs via Scalable Question Synthesis from Scratch
Paper
•
2410.18693
•
Published
•
39
We introduce ScaleQuest, a scalable and novel data synthesis method. Project Page: https://scalequest.github.io/
Note Problem-Solving model fine-tuned based on Mistral-7B-v0.1 using ScaleQuest-Math dataset.
Note Problem-Solving Model fine-tuned based on Llama3-8B using ScaleQuest-Math dataset.
Note Problem-Solving Model fine-tuned based on DeepSeekMath-7B using ScaleQuest-Math dataset.
Note Problem-Solving Model fine-tuned based on Qwen2-Math-7B using ScaleQuest-Math dataset.
Note Question Generator Model based on DeepSeekMath-7B-RL.
Note Question Generator Model based on Qwen2-Math-7B-Instruct.