Model Overview

Saxo/Linkbricks-Horizon-AI-Avengers-V3-32B is a 32.8 billion parameter multilingual language model developed by Mr. Yunsung Ji (Saxo), a data scientist at Linkbricks. This model was fine-tuned using SFT->DPO->ORPO training on approximately 35% of the parameters of the Saxo/Linkbricks-Horizon-AI-Avengers-V2-32B base model, utilizing 8 H100-80G GPUs. It incorporates 80 million diverse multilingual news and wiki corpora, alongside specialized cross-learning data for Japanese, Korean, Chinese, and English, as well as mathematical and logical reasoning datasets.

Key Capabilities

• Multilingual Enhancement: Strong performance in Japanese, Korean, Chinese, and English, with cross-language processing capabilities.
• Complex Problem Solving: Trained to address intricate logical and mathematical problems.
• High-Dimensional Analysis: Enhanced for analyzing customer reviews and social media posts.
• Core AI Tasks: Improved performance in coding, writing, mathematics, and general decision-making.
• Function Calling: Supports Function Call and Tool Calling functionalities.

Technical Details

• The tokenizer uses the base model without word expansion.
• Training utilized Deepspeed Stage=3, rslora, and BAdam Layer Mode.
• Developed with transformers_version: 4.46.3.

Good For

• Applications requiring robust multilingual understanding and generation across East Asian languages and English.
• Tasks involving complex logical reasoning and mathematical problem-solving.
• Analyzing and summarizing large volumes of customer feedback and social media data.
• Code generation, creative writing, and general intelligent decision support systems.