vitruv/vitruv_1

TEXT GENERATIONConcurrency Cost:1Model Size:15BQuant:FP8Ctx Length:8kTool Calling:SupportedPublished:Feb 1, 2024License:apache-2.0Architecture:Transformer Open Weights Cold

vitruv/vitruv_1 is a 15 billion parameter language model developed by Virtruv, built upon the 'beomi/OPEN-SOLAR-KO-10.7B' base. This model is specifically fine-tuned for mathematical tasks in Korean, leveraging a unique dataset including translated mathematical content. It is optimized to excel in Korean-language mathematical reasoning and problem-solving.

Loading preview...

vitruv/vitruv_1: Korean Math-Focused LLM

vitruv/vitruv_1 is a 15 billion parameter large language model developed by Virtruv, specifically engineered to enhance mathematical reasoning capabilities in Korean. It is built on the 'beomi/OPEN-SOLAR-KO-10.7B' base model.

Key Capabilities

  • Korean Mathematical Proficiency: The model has undergone specialized training with a focus on Korean mathematical content.
  • Targeted Fine-tuning: Utilizes a curated dataset including traintogpb/aihub-koen-translation-integrated-tiny-100k, kyujinpy/KOR-gugugu-platypus-set, and a custom-translated subset of GAIR/MathPile to improve its mathematical understanding and generation in Korean.

Good For

  • Applications requiring strong mathematical problem-solving in the Korean language.
  • Educational tools or platforms focused on Korean mathematics.
  • Research into domain-specific fine-tuning for non-English mathematical contexts.