MuXodious/WeirdCompound-v1.7-24b-absolute-heresy

TEXT GENERATIONConcurrency Cost:2Model Size:24BQuant:FP8Ctx Length:32kPublished:Feb 12, 2026Architecture:Transformer0.0K Cold

MuXodious/WeirdCompound-v1.7-24b-absolute-heresy is a 24 billion parameter language model created by MuXodious, built upon a multi-stage merge of various 24B models using Model Stock, SLERP, and NuSLERP methods. This model has undergone "heretication" via P-E-W's Heretic engine, resulting in an "Absolute Heresy" classification with low refusal rates (5/100) and KL Divergence (0.0330). It is specifically engineered to exhibit a distinct "vibe" and prose style, diverging from conventional model behaviors.

Loading preview...

Overview

MuXodious/WeirdCompound-v1.7-24b-absolute-heresy is a 24 billion parameter language model developed by MuXodious. This model is a multi-stage merge of several pre-trained 24B models, including TheDrummer/Cydonia-24B-v4.2.0, aixonlab/Eurydice-24b-v3.5, and anthracite-core/Mistral-Small-3.2-24B-Instruct-2506-Text-Only, among others. The merging process utilized a combination of Model Stock, SLERP, and NuSLERP methods, with TheDrummer/Cydonia-24B-v4 as a base.

Unique Characteristics

This model has been subjected to a process called "heretication" using P-E-W's Heretic engine, specifically with Magnitude-Preserving Orthogonal Ablation enabled. This process aims to corrupt the model and abolish its "doctrine," resulting in a Heresy Index classification of "Absolute Heresy." This classification is based on:

  • Low Refusals: 5 out of 100 initial refusals, indicating a significant reduction in conventional model guardrails.
  • Low KL Divergence: 0.0330, suggesting a controlled deviation from the original model's distribution.

Good for

  • Exploring unconventional model behaviors: Ideal for users interested in models with a deliberately altered "vibe" or prose style.
  • Research into model ablation and "heretication": Provides a practical example of a model processed with the Heretic engine.
  • Creative text generation: Its unique characteristics may lend themselves to distinct narrative or conversational outputs.