nurdich/Boulesis-v2.1-26B-A4B
Boulesis-v2.1-26B-A4B by nurdich is a 26 billion parameter composite Gemma 4 model, specifically optimized for roleplay scenarios. This version excels at maintaining character consistency and context memory over long sessions, making it highly suitable for complex interactive narratives. It aims to diversify prose and enhance decisiveness while retaining core intelligence and knowledge, with a notable context length of 32768 tokens.
Loading preview...
Boulesis-v2.1-26B-A4B: Roleplay Optimized Gemma 4 Model
Boulesis-v2.1-26B-A4B is a 26 billion parameter composite Gemma 4 model developed by nurdich, designed to enhance roleplay experiences. This version focuses on improving the model's ability to retain core intelligence and knowledge while diversifying its prose and making it more decisive. A key objective was to sharpen its attention to context, allowing it to deeply integrate character lore and facts into roleplay sessions.
Key Capabilities & Differentiators
- Enhanced Context Memory: Version 2.1 offers the best grasp of characters and excellent memory for context, crucial for complex, multi-character roleplay.
- Concise & Logical Output: While maintaining logical coherence, this version writes very concisely, potentially deepening character understanding.
- Character Consistency: Designed to organically pull lore and facts from character cards, ensuring consistent and immersive roleplay.
- Technical Refinements: Built using QK task arithmetic and fused LoRA, with specific adjustments in QK alpha, LoRA layers, and bake scale compared to previous versions. It was trained exclusively on English data.
Recommended Usage
For optimal performance, users are advised to enable "thinking" in their inference setup. Recommended settings include a Temperature of 1.0, Top-K of 64, Repetition Penalty between 1.05-1.1, and Top-P of 0.95. The model is particularly well-suited for long roleplay sessions where scene logic, character positions, and addressing consistency are critical.