FT-LLM 2026 Reasoning SFT

A supervised fine-tuning (SFT) dataset of mathematical chain-of-thought traces used in Phase 2 of the COMPASS project to enhance the reasoning capability of a Japanese VLM's LLM backbone. Reasoning traces were distilled from a Qwen3-30B teacher over a broad pool of public math-reasoning datasets (GSM8K, MATH, SVAMP, AQuA-RAT, MathInstruct, MGSM-ja, Orca Math, NuminaMath-CoT, OpenR1-Math, Open Math Reasoning, and others). Each sample follows a structured XML format with <Problem>, <Thinking>, and <Answer> tags, making the reasoning steps and final answer easy to parse.

Part of the Compass collection.

License

Released under the Apache License 2.0.

Note on source materials and Japanese copyright law: Under Article 30-4 of the Japanese Copyright Act, the use of copyrighted works for the purpose of information analysis — including machine learning training — is a permitted use that does not require authorization from, or trigger license conditions of, the copyright holders. This dataset was produced in Japan on that basis, and the resulting artifacts are redistributed under Apache-2.0. Downstream users are responsible for complying with any applicable terms of the original source datasets and the teacher model's usage policy in their own jurisdiction.