Saxo/Linkbricks-Horizon-AI-Llama-3.3-Japanese-70B-Advanced

TEXT GENERATIONPricing:Input $3.5 / Cached $0.7 / Output $8.3Concurrent Unit Cost:4Model Size:70BQuant:FP8Context Size:8kTool Calling:SupportedPublished:Dec 17, 2024License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Saxo/Linkbricks-Horizon-AI-Llama-3.3-Japanese-70B-Advanced is a 70 billion parameter Japanese-enhanced language model developed by Yunsung Ji (Saxo) at Linkbricks. This model, based on Llama-3.3, was trained using CPT->SFT->DPO on 30 million Japanese news and wiki corpora, alongside cross-lingual data for Japanese, Korean, Chinese, and English. It features an extended 128k context window and is specifically optimized for high-dimensional analysis of customer reviews and social posts, coding, writing, mathematics, and complex logical reasoning.

Loading preview...

Saxo/Linkbricks-Horizon-AI-Llama-3.3-Japanese-70B-Advanced Overview

This 70 billion parameter model, developed by Yunsung Ji (Saxo) at Linkbricks, is a Japanese-enhanced language model built upon the Llama-3.3 base. It underwent a comprehensive CPT->SFT->DPO training process utilizing 8 H100-80G GPUs. A key focus of its training involved 30 million Japanese news and wiki corpora, augmented with cross-lingual data spanning Japanese, Korean, Chinese, and English to bolster its multilingual capabilities and logical reasoning.

Key Capabilities

  • Enhanced Japanese Processing: Specialized training on extensive Japanese datasets for superior performance in the language.
  • Multilingual Cross-Learning: Incorporates data from Japanese, Korean, Chinese, and English for improved understanding and generation across these languages.
  • Extended Context Window: Features a 128k context window, enabling the processing of longer and more complex inputs.
  • Advanced Analytical Skills: Strengthened for high-dimensional analysis of customer reviews and social media posts.
  • Coding and Writing Proficiency: Optimized for code generation and various writing tasks.
  • Mathematical and Logical Reasoning: Designed to handle complex mathematical problems and intricate logical judgments.
  • Function and Tool Calling: Supports function calling and tool calling, enhancing its utility in integrated systems.

Good For

  • Applications requiring deep analysis of Japanese text, such as news summarization or content generation.
  • Cross-lingual tasks involving Japanese, Korean, Chinese, and English.
  • Use cases demanding long-context understanding and generation.
  • Developing intelligent agents for customer service, social media monitoring, and content creation.
  • Tasks that involve complex problem-solving in mathematics and logic.

Finetunes
1 models