VIDraft/JGOS-31B-Think
VIDraft/JGOS-31B-Think is a 31 billion parameter large language model developed by VIDraft, based on the Gemma4 architecture. This model is specifically optimized for reasoning tasks in Korean, featuring native step-by-step 'think' reasoning capabilities. With a context length of 32768 tokens, it is designed to excel in complex analytical and logical problem-solving scenarios within the Korean language domain.
Loading preview...
Overview
VIDraft/JGOS-31B-Think is a 31 billion parameter large language model developed by VIDraft, built upon the Gemma4 architecture. Its primary focus is on enhancing reasoning capabilities specifically for the Korean language, incorporating native step-by-step 'think' processes.
Key Capabilities
- Korean Reasoning: Optimized for complex analytical and logical problem-solving in Korean.
- Step-by-Step Thinking: Features native support for detailed, sequential reasoning.
- Gemma4 Base: Inherits the foundational strengths and licensing of the Gemma model family.
- High Context Length: Supports a substantial context window of 32768 tokens, beneficial for intricate reasoning tasks.
Deployment and Usage
This model is available for Docker deployment, with a pre-configured image (vidraft/jgos-31b-think:01.03) that includes vLLM 0.22.0. It offers an OpenAI-compatible API on port 8000. For evaluation, it's memory-tuned for single GPU setups, requiring approximately 59GB of VRAM for bf16 weights (suitable for GPUs with 64GB or more) and can be adapted for smaller GPUs using FP8. The model's license is inherited from its Gemma base.