SpoomplesMaxx 27B 4500 SFT: Creative Writing & Roleplay Specialist

The aimeri/spoomplesmaxx-27b-4500 is a supervised fine-tuning (SFT) adapter for the Gemma 3 27B architecture, specifically building upon the aimeri/spoomplesmaxx-base-gemma3-27b-4500 checkpoint. This model is part of the larger SpoomplesMaxx project, a hobbyist ML research effort focused on enhancing creative writing and roleplay capabilities in open base models.

Key Capabilities

• System-prompt following: Adheres to instructions for character behavior and narrative structure.
• Persona consistency: Maintains persistent character traits and voices across multi-turn interactions.
• Structured narrative reasoning: Manages explicit scene, character, and continuity states for coherent storytelling.
• Multilingual creative writing: Supports generation in English and Brazilian Portuguese.

Training Details

This SFT adapter was trained using two primary datasets:

• spoomplesmaxx-olivia-sft: Approximately 78,000 rows of DanChat-2 formatted data, focusing on persona consistency and system-prompt following, featuring the "Olivia Costa" persona.
• spoomplesmaxx-rp-reasoning: Around 9,600 ShareGPT formatted entries for structured roleplay reasoning, teaching explicit narrative state tracking.

Intended Use Cases

This model is specifically designed for:

• Character roleplay and collaborative fiction.
• System-prompt-driven character following with multi-character mode-switching.
• Structured narrative generation with scene and continuity state tracking.

It is not intended for general-purpose instruction following, factual Q&A, or safety-critical applications, as it is optimized for creative writing and roleplay, and the base CPT model is uncensored. Future plans include a DPO alignment stage for further refinement.