Created Japanese quantitative reasoning dataset with reasoning processes "JaMathCoT" and performed SFT on Sarashina2.2
To improve the Japanese quantitative reasoning ability of LLMs, a synthetic dataset "JaMathCoT" containing problems and reasoning processes was created, and SFT of Sarashina2.2 was attempted.
Chain-of-Thought (CoT) is a method to improve the reasoning ability of Large Language Models (LLMs) by outputting not only the final answer to a problem but also the intermediate reasoning process leading to the answer in natural language.
To improve the Japanese quantitative reasoning ability of LLMs using CoT, a synthetic dataset "JaMathCoT" containing Japanese quantitative reasoning problems and reasoning processes was created. Additionally, SFT (Supervised Fine-Tuning) of Sarashina2.2 was attempted using the created JaMathCoT and existing quantitative reasoning datasets.
This item is a summary. The full, checkable version is at the original source:
Read the original at Zenn (Hugging Face) ↗