Model Overview

Saxo/Linkbricks-Horizon-AI-Avengers-V1-32B is a 32.8 billion parameter multilingual enhanced language model developed by Linkbricks Horizon-AI. Built upon the Qwen/Qwen2.5-32B-Instruct base model, it underwent extensive fine-tuning (CPT->SFT->DPO->ORPO) using 8 H100-80G GPUs, with approximately 35% of its parameters adjusted.

Key Capabilities

• Multilingual Enhancement: Trained on 80 million multilingual news and wiki corpora, with a strong focus on cross-learning data for Japanese, Korean, Chinese, and English. This enables robust performance across these languages.
• Advanced Analysis: Enhanced for high-dimensional analysis of customer reviews and social posts, providing deeper insights into unstructured text data.
• Cognitive Tasks: Demonstrates strengthened capabilities in coding, writing, mathematics, and complex logical reasoning, making it suitable for tasks requiring analytical and problem-solving skills.
• Function Calling: Supports Function Call and Tool Calling, allowing for integration with external tools and APIs.

Training Details

The model utilizes the base model's tokenizer without word expansion and was trained using advanced techniques including Deepspeed Stage=3, rslora, and BAdam Layer Mode. It is designed to handle complex logical problems and cross-language understanding effectively.