sh0ck0r/Mero-Artemis-31B-v0.3.1-heretic

VISIONConcurrency Cost:2Model Size:31BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:May 28, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

sh0ck0r/Mero-Artemis-31B-v0.3.1-heretic is a 31 billion parameter decensored language model based on sophosympatheia/Mero-Artemis-31B-v0.3.1, created using the Heretic tool. This model is an experimental merge of BeaverAI/Artemis-31B-v1h-GGUF and zerofata/G4-MeroMero-31B, aiming to combine Artemis's personality with MeroMero's attention to detail. It features significantly reduced refusals compared to its original counterpart, making it suitable for applications requiring less restrictive content generation.

Loading preview...

Model Overview

sh0ck0r/Mero-Artemis-31B-v0.3.1-heretic is a 31 billion parameter language model that has been decensored from its base, sophosympatheia/Mero-Artemis-31B-v0.3.1, using the Heretic tool. This version is an experimental merge, combining the "personality and style" of Artemis with the "excellent attention to details" of MeroMero.

Key Characteristics

  • Decensored: Significantly reduces content refusals, with 42/100 refusals compared to 99/100 in the original model.
  • Experimental Merge: Created by merging BeaverAI/Artemis-31B-v1h-GGUF and zerofata/G4-MeroMero-31B.
  • Reproducible: The decensoring process is documented and reproducible, with details available in the reproduce directory.

Performance

This model demonstrates a KL divergence of 0.0793 when compared to the original sophosympatheia/Mero-Artemis-31B-v0.3.1. Its primary differentiator is the substantial reduction in content refusals, making it more permissive for various generative tasks.

Use Cases

This model is particularly suited for applications where a less restrictive content policy is desired, allowing for broader and more unconstrained text generation. Developers seeking a model that balances distinct personality with detailed output, while minimizing content filtering, may find this model beneficial.