Eric Hartford open-sources QuixiMath: synthetic math dataset with step-by-step scratchpads
QuixiAI · x · 2026-09-07
Eric Hartford (QuixiAI) released Quixi Math, a synthetic math data generator, along with the QuixiMath-1B dataset on Hugging Face, aimed at training and evaluating LLMs on multi-step mathematical reasoning.
- Each example includes problem text, pipe-delimited solution steps, a canonical final answer, and curriculum metadata.
- The generator covers multiple domains (algebra, logic, probability, stats, sets) via modular problem generators.
- It supports both SFT-style trace learning and RL-style training; the repo ships design docs, training notes and tests.
Related event: Eric Hartford Open-Sources QuixiMath Synthetic Math Reasoning Dataset(2 posts)→
More from Research
- OR-Clarify benchmarks asking clarifying questions before optimization modeling — AIOR-Research · 2026-09-07
- How keyword spotting models power "Hey Siri": a detailed technical guide — cneuralnetwork · 2026-09-07
- Reward hacking stems from bad reward modeling and eval awareness, developer argues — secemp9 · 2026-09-07
- ECCV 2026 Oral Poppy: training-free polarization cues cut surface normal error by up to 26% — ssh4net · 2026-09-07
- AI OCR quietly corrupts protein sequences, and patent PDFs often contain the typos themselves — iskander · 2026-09-07
- New translation benchmark spans 109 languages; LLM verifier pass rate jumps 7.2% to 89.8% with explicit rules — LChoshen · 2026-09-07