Saxo/Linkbricks-Horizon-AI-Japanese-Base-32B

TEXT GENERATIONPricing:Input $2.72 / Output $4.8Concurrent Unit Cost:2Model Size:32.8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Dec 28, 2024License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Saxo/Linkbricks-Horizon-AI-Japanese-Base-32B is a 32.8 billion parameter language model developed by Linkbricks, based on the Qwen/QwQ-32B-Preview architecture. It has undergone continued pre-training, supervised fine-tuning, and DPO specifically to enhance its Japanese language capabilities, utilizing 20 million Japanese news and wiki corpora. This model excels in high-dimensional analysis of customer reviews and social posts, coding, writing, mathematics, and logical reasoning, supporting a 128k context window and function calling.

Loading preview...

Overview

Saxo/Linkbricks-Horizon-AI-Japanese-Base-32B is a 32.8 billion parameter language model developed by Yunsung Ji (Saxo), a data scientist at Linkbricks, an AI and big data analytics company. This model is built upon the Qwen/QwQ-32B-Preview base model and has undergone extensive continued pre-training (CPT), supervised fine-tuning (SFT), and Direct Preference Optimization (DPO) with a focus on Japanese language enhancement. Approximately 25% of its parameters were fine-tuned using 8 H100-80G GPUs.

Key Capabilities

  • Japanese Language Enhancement: Trained on 20 million Japanese news and wiki corpora, making it highly proficient in Japanese.
  • Multilingual Cross-Learning: Incorporates Japanese, Korean, Chinese, and English cross-training data for various tasks.
  • Advanced Analytical Skills: Enhanced for high-dimensional analysis of customer reviews and social posts.
  • Coding and Writing: Strong capabilities in code generation and creative writing.
  • Mathematical and Logical Reasoning: Designed to handle complex mathematical problems and logical decision-making.
  • Extended Context Window: Supports a 128k context window, allowing for processing longer inputs.
  • Function Calling: Includes support for function and tool calling.

Good for

  • Applications requiring strong Japanese language understanding and generation.
  • Analyzing customer feedback and social media data in a high-dimensional context.
  • Tasks involving coding, creative writing, and content generation.
  • Solving complex mathematical and logical reasoning problems.
  • Use cases benefiting from a large context window for detailed analysis.

Finetunes
1 models