aimeri/spoomplesmaxx-gemma4-31B-v1.1

VISIONConcurrency Cost:2Model Size:31BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:May 25, 2026Architecture:Transformer0.0K Cold

SpoomplesMaxx-V1.1-31B is a 31 billion parameter generalist language model developed by aimeri, built upon the Google Gemma 4 base architecture. It excels in creative writing and roleplay, offering light competence in instruction following and reasoning. The model was continuously pre-trained on an English and Portuguese-heavy corpus, then fine-tuned with an SFT pipeline that included instruction tuning, reasoning, and a specific narrative persona, with training samples up to 32K tokens in length.

Loading preview...

SpoomplesMaxx-V1.1-31B Overview

SpoomplesMaxx-V1.1-31B is a 31 billion parameter language model developed by aimeri, based on the google/gemma-4-31b architecture. It is primarily designed for creative writing and roleplay, demonstrating light competence in instruction following and reasoning tasks. The model underwent continued pre-training on a corpus rich in English and Portuguese, followed by a Supervised Fine-Tuning (SFT) pipeline that incorporated instruction tuning, reasoning, and a unique narrative persona.

Key Capabilities & Features

  • Creative Writing & Roleplay: Optimized for generating engaging narratives and participating in roleplaying scenarios.
  • Instruction Following: Possesses light competence in understanding and executing instructions.
  • Reasoning: Exhibits basic reasoning abilities.
  • Extended Context: Trained with samples varying in length up to 32K tokens, suggesting potential for handling longer contexts.
  • Persona Training: Includes a specific "Olivia" persona, an experimental feature demonstrating the pipeline's ability to imbue the model with distinct personalities. This persona can be activated via a detailed system prompt.
  • Multilingual Support: While primarily English and Portuguese-heavy in its fine-tuning, the base model offers broader multilingual capabilities.

Important Considerations

  • Alignment: No RLHF or safety alignment has been applied beyond the base model. It will comply with requests that more aligned models might refuse.
  • Tool Calling: This version does not include tool-calling datasets; future fine-tunes are planned for this specific capability.

Good For

  • Creative writing applications.
  • Companion AI development.
  • Roleplaying use cases.