Pranavz/Hearthfire-24B-heretic
Pranavz/Hearthfire-24B-heretic is a 24 billion parameter decensored version of LatitudeGames' Hearthfire-24B, created using Heretic v1.2.0. This model is fine-tuned from Mistral Small 3.2 Instruct for narrative longform writing, specifically embracing atmosphere and introspection over high-stakes action. It excels at generating immersive, slower-paced narratives in second-person present tense, prioritizing 'vibes over velocity' in storytelling.
Loading preview...
Overview
Pranavz/Hearthfire-24B-heretic is a 24 billion parameter language model derived from LatitudeGames/Hearthfire-24B, specifically modified using Heretic v1.2.0 to be a decensored version. It is fine-tuned from Mistral Small 3.2 Instruct, focusing on narrative longform writing with a unique emphasis on atmosphere, introspection, and a slower pace, rather than constant plot progression.
Key Characteristics
- Decensored Behavior: Significantly reduced refusals (2/100) compared to the original model (96/100), achieved through specific 'abliteration' parameters.
- Narrative Style: Tuned to appreciate 'vibes over velocity,' comfortable with silence, and avoids forcing dramatic events.
- Training Data: Supervised Fine-Tuning (SFT) on thousands of 8-16K context longform writing examples, primarily in second-person present tense with some third-person data.
- Continue-Heavy Structure: Training data allows the model to drive the plot independently, with extended narrative blocks and sparse player actions.
- Cooperative Tone: Retains the base Mistral model's warmth, inclined to be cooperative and atmospheric rather than hostile.
- Agency: Designed to write in the user's stead, acting and speaking for characters to maintain narrative flow.
Recommended Use Cases
- Immersive Storytelling: Ideal for generating rich, atmospheric narratives where pacing is deliberate and introspection is valued.
- Creative Writing: Suited for scenarios requiring detailed scene descriptions and character development without constant external conflict.
- Roleplay: Excellent for roleplay experiences where the model can contribute significantly to the narrative and character actions, rather than just reacting to user input.
Limitations
- Pacing: Deliberately slower-paced; may not react to player actions with immediate dramatic consequences.
- Tone: More cooperative and atmospheric; less inclined towards grit or hostile interactions compared to other models.