MuXodious/Wayfarer-2-12B-absolute-heresy

TEXT GENERATIONConcurrency Cost:1Model Size:12BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Jan 15, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

MuXodious/Wayfarer-2-12B-absolute-heresy is a 12 billion parameter fine-tuned Wayfarer-2 model, created by MuXodious using P-E-W's Heretic engine, with a 32768 token context length. This model is specifically designed for narrative generation in text-adventure and roleplay scenarios, emphasizing conflict, tension, and the possibility of character failure or death. It aims to counter the positivity bias in typical language models by generating pessimistic narratives with real stakes. Its primary strength lies in crafting vivid, engaging stories in a second-person present tense narrative style.

Loading preview...

Model Overview

MuXodious/Wayfarer-2-12B-absolute-heresy is a 12 billion parameter model, fine-tuned from the Wayfarer-2 architecture using P-E-W's Heretic engine. This specific version is classified with an "Absolute Heresy" index, indicating a low refusal rate (5/100) and KL Divergence (0.0193), though this classification is noted as arbitrary and not indicative of performance.

Key Capabilities

  • Narrative Generation: Excels at crafting vivid, engaging narratives, particularly for text-adventure and roleplay. It is designed to generate stories with conflict, tension, and the possibility of character failure or death.
  • Pessimistic Sentiment: Trained to produce narratives with a general emotional sentiment of pessimism, countering the positivity bias often found in other language models.
  • Second-Person Present Tense: Optimized for generating content in the second-person present tense ("You are") narrative style.
  • Instructional Capabilities: Retains instructional capabilities due to a small instruct core in its training.

Training Details

The model underwent Supervised Fine-Tuning (SFT) using a dataset comprising the Wayfarer 2 dataset, sentiment-balanced roleplay transcripts, and an instruct core. The text adventure data was generated by simulating playthroughs of AI Dungeon scenarios, with one language model acting as narrator and another as the user, running for up to 8k tokens or until character death.

Recommended Use Cases

  • Text Adventures: Ideal for creating dynamic and challenging text-based adventure games where player choices have significant consequences, including character death.
  • Roleplay Scenarios: Suitable for roleplaying applications that require a more realistic and less forgiving narrative experience.
  • Creative Writing: Can be used by writers looking for an AI assistant to generate darker, more conflict-driven story elements.

Limitations

  • Primarily trained on second-person present tense data; other perspectives may yield suboptimal results.
  • The Heresy Index is an arbitrary classification and does not directly indicate model performance.