teamsparta-inc/K-AX-Spartan-OSS-27B
K-AX Spartan OSS 27B is a 27 billion parameter open-weight language model developed by 스파르타AX (SpartaAX), based on the Qwen 3.5 architecture. It features a large 262,144 token context window and an expanded 248,320-token Korean tokenizer. This model is specifically optimized for Korean-specific domains, including specialized knowledge, legal, cultural, and professional examination content, through continual pre-training and domain-adaptive reinforcement learning on proprietary Korean corpora.
Loading preview...
K-AX Spartan OSS 27B: Korean-Optimized Qwen 3.5 Based Model
K-AX Spartan OSS 27B, developed by 스파르타AX (SpartaAX), is a 27 billion parameter open-weight model built upon the Qwen 3.5 architecture. It is specifically engineered for the Korean language, featuring an expanded 248,320-token Korean tokenizer and a substantial 262,144 token context window. The model undergoes continual pre-training and domain-adaptive reinforcement learning using proprietary Korean corpora and a multi-stage alignment pipeline.
Key Capabilities & Features
- Korean Domain Optimization: Excels in Korean-specific knowledge, legal, cultural, and professional examination contexts.
- Advanced Training: Utilizes proprietary multi-stage alignment and continual pre-training on custom Korean datasets.
- Adapter Native Structure: Designed as a base model for the K-AX Spartan Adapter product family, allowing for cost-effective domain-specific adapter creation.
- Broad Inference Engine Compatibility: Supports vLLM, SGLang, and TensorRT-LLM.
- High Precision: Operates in bfloat16 precision.
Good for
- Applications requiring deep understanding and generation of Korean text, especially in specialized fields.
- Developers looking for a robust Korean base model compatible with adapter-based fine-tuning.
- Research and development in Korean natural language processing.
- Use cases demanding a large context window for complex Korean queries.