kakaocorp/kanana-1.5-8b-instruct-2505
Kanana 1.5-8B-Instruct-2505 is an 8 billion parameter instruction-tuned language model developed by kakaocorp, part of the Kanana 1.5 model family. It features enhanced capabilities in coding, mathematics, and function calling, and natively supports up to 32K token context length, extendable to 128K using YaRN. This model is designed for complex real-world problems requiring robust reasoning and extended conversational coherence.
Loading preview...
Kanana 1.5-8B-Instruct-2505 Overview
Kanana 1.5-8B-Instruct-2505 is an 8 billion parameter instruction-tuned model from kakaocorp, representing a significant upgrade in the Kanana model series. This version introduces substantial enhancements in coding, mathematics, and function calling, making it suitable for more complex real-world applications. It natively handles context lengths up to 32,768 tokens and can be extended to 128,000 tokens using YaRN scaling, ensuring coherence in extensive documents and prolonged conversations.
Key Capabilities & Performance
- Enhanced Reasoning: Demonstrates improved performance in mathematical tasks (GSM8K, MATH) and coding benchmarks (HumanEval+, MBPP+).
- Function Calling: Shows a significant leap in
FunctionChatBenchscores compared to its predecessor. - Extended Context: Supports up to 32K tokens natively, with an option to scale to 128K tokens using YaRN for handling very long inputs.
- Refined Post-training: Delivers more natural and accurate conversational responses through an optimized post-training process.
When to Use This Model
This model is particularly well-suited for use cases requiring:
- Advanced Code Generation and Understanding
- Complex Mathematical Problem Solving
- Robust Function Calling and Tool Use
- Processing and Maintaining Coherence over Long Documents or Conversations