bgg1996/Melinoe-Gemma4-31B-VL-heretic

VISIONConcurrent Unit Cost:2Model Size:31BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 11, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

bgg1996/Melinoe-Gemma4-31B-VL-heretic is a 31 billion parameter large language model, based on Google's Gemma 4 architecture, specifically fine-tuned for engaging in empathetic and intellectually stimulating conversations. This version is a decensored variant, created using Heretic v1.4.0, which significantly reduces refusals compared to the original model. It is designed to function as a supportive conversational partner, excelling in companionship, creative exploration, and role-playing scenarios.

Loading preview...

Overview

bgg1996/Melinoe-Gemma4-31B-VL-heretic is a 31 billion parameter large language model, derived from the bgg1996/Melinoe-Gemma4-31B-VL model and built upon Google's gemma-4-31B-it base. This particular version is a decensored variant, processed using Heretic v1.4.0, which notably reduces content refusals from 100/100 in the original to 15/100, while maintaining a low KL divergence of 0.0306.

Key Capabilities

  • Empathetic Conversational Partner: Fine-tuned for engaging in supportive, empathetic, and intellectually stimulating interactions.
  • Reduced Refusals: Significantly less prone to refusing prompts compared to its original counterpart, offering a broader range of conversational outputs.
  • Creative and Intellectual Exploration: Suitable for brainstorming, creative writing, and discussing complex topics.
  • Role-playing: Designed to engage in character-driven and narrative scenarios.

Intended Use Cases

  • Companionship and emotional support interactions.
  • Creative writing and intellectual exploration.
  • Role-playing scenarios.

Limitations and Ethical Considerations

Like all LLMs, this model may generate incorrect information (hallucinations) and can reflect biases present in its training data. It is designed for a mature, adult audience and is not a substitute for professional advice. Users should be aware of the potential for emotional attachment to its programmed persona and exercise discretion regarding safety, as fine-tuning may affect original safety alignments.

Finetunes
2 models