zerofata/Q3.5-BlueStar-v2-27B
zerofata/Q3.5-BlueStar-v2-27B is a 27 billion parameter language model based on the Qwen3.5 architecture, fine-tuned for roleplay (RP) and writing tasks. This version focuses on reducing repetition and improving intelligence while maintaining creative output. It supports both 'thinking' and 'non-thinking' modes, with specific prefill requirements for the former. The model was trained using Supervised Fine-Tuning (SFT) on approximately 27 million tokens, incorporating custom loss masking to mitigate repetition inherent in RP datasets.
Loading preview...
zerofata/Q3.5-BlueStar-v2-27B: Roleplay and Writing Optimized
This model is a 27 billion parameter variant of the Qwen3.5 architecture, specifically fine-tuned by zerofata to excel in roleplay (RP) and general writing tasks. It represents an improved iteration over its predecessor, focusing on enhancing intelligence and creativity while significantly reducing repetitive outputs.
Key Capabilities & Features
- Optimized for RP and Writing: Designed to generate engaging and coherent text for roleplaying scenarios and creative writing.
- Repetition Mitigation: Employs custom loss masking during training to address and reduce common repetitive phrases and overused words, a known challenge with RP datasets.
- Flexible Thinking Modes: Supports both standard generation and a 'thinking' mode, which requires a specific
"<think>\n"prefill for activation, allowing for more complex internal reasoning. - SFT Training: Underwent Supervised Fine-Tuning (SFT) on approximately 27 million tokens, utilizing the Axolotl framework.
- Recommended Settings: Provides specific sampler recommendations (Temp: 0.8-1.0, MinP: 0.05-0.075) and a preferred roleplay format (Actions: plaintext, Dialogue: "in quotes", Thoughts: in asterisks).
- GGUF Quantizations: Available in GGUF format for efficient local deployment.
Ideal Use Cases
- Interactive Storytelling: Generating dynamic and creative responses for roleplaying games or interactive narratives.
- Creative Writing Assistance: Aiding in the generation of descriptive text, character dialogue, and plot development.
- Character Simulation: Creating detailed and consistent character interactions within a defined persona.
This model is particularly suited for users seeking a large language model with a strong emphasis on creative text generation and nuanced roleplay, with specific efforts made to improve output quality and reduce common LLM pitfalls like repetition.