Tether releases 191.4 billion token STEM dataset, betting on the 'explanatory power' of local small models.

BlockBeats news, September 23 - Stablecoin giant Tether's Tether AI Research today released QVAC Genesis III, a synthetic training dataset for the science, technology, engineering, and mathematics fields, with a scale of 191.43 billion tokens and covering 159.6 million documents.

It is reported that Tether aims to use this to enhance the reasoning and teaching capabilities of small AI models, enabling more AI assistants to run directly on laptops, phones, and local servers, reducing continuous reliance on large cloud models.

Genesis III continues the data approach of Genesis I and Genesis II, with content covering 19 curriculum-related fields including biology, chemistry, physics, mathematics, computer science, medicine, electrical engineering, and machine learning, targeting high school, university, and professional-level applications. Tether states that the dataset will focus on training models to explain problem-solving paths, identify erroneous reasoning, and provide corrections, rather than only outputting final answers.

According to test results disclosed by Tether, a 1.7 billion parameter model trained with Genesis III option-level data achieved an effective answer rate of 99.45% in relevant benchmark tests. Compared with a model trained on Cosmopedia-v2 with a comparable token scale, Genesis III improved scores on the ARC-Easy, ARC-Challenge, and MMLU STEM benchmarks by 28.57, 21.35, and 15.03 percentage points, respectively.