LATIDIA · Investigación

QVAC Genesis III: un corpus STEM sintético abierto a gran escala y de alta calidad para una capacitación previa eficiente del modelo de lenguaje

arXiv: 2609.19513v1 Tipo de anuncio: nuevo Resumen: Los datos previos a la capacitación de alta calidad son un cuello de botella crítico para los modelos de lenguaje educativos y específicos de STEM dirigidos a la IA perimetral y la implementación en el dispositivo donde los presupuestos de tokens son

La noticia

arXiv:2609.19513v1 Announce Type: new Abstract: High-quality pre-training data is a critical bottleneck for educational and STEM-specific language models targeting edge AI and on-device deployment where token budgets are tightly constrained. While major organizations train ever-larger models on private corpora, the open ecosystem lacks STEM-focused synthetic datasets that deliver high per-token learning value efficiently for small models. To address this gap, we introduce QVAC Genesis III, a 191.43B-token, STEM-focused multi-domain synthetic corpus covering 19 domains across several difficulty levels and different educational styles. QVAC Genesis III is built via a dual generation strategy that performs targeted teacher distillation using a weak edge-scale student model as signal: the student's failures are converted