KotEnjoyer/AIDungeonTextAdventuresMistral

VISIONConcurrency Cost:2Model Size:24BQuant:FP8Ctx Length:32kPublished:Jan 20, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

KotEnjoyer/AIDungeonTextAdventuresMistral is a 24 billion parameter Mistral-based model developed by KotEnjoyer, fine-tuned specifically for text adventure game generation. It was trained on the original AI Dungeon 2 dataset and is designed to recognize player actions prefixed with "> You" for interactive storytelling. This model excels at generating narrative responses in a second-person perspective, making it suitable for creating dynamic and engaging text-based games.

Loading preview...

Overview

KotEnjoyer/AIDungeonTextAdventuresMistral is a 24 billion parameter model, fine-tuned by KotEnjoyer from unsloth/mistral-small-3.1-24b-base-2503-bnb-4bit. It leverages the Mistral 3 architecture and was trained using Unsloth and Huggingface's TRL library, enabling 2x faster training. The model's primary distinction is its specialized fine-tuning on the original text_adventures.txt dataset, which was initially used for AI Dungeon 2.

Key Capabilities

  • Text Adventure Generation: Specifically optimized for creating interactive text-based adventures.
  • Player Action Recognition: Designed to interpret player inputs prefixed with "> You" as direct actions within the game narrative.
  • Second-Person Narrative: Generates responses in a second-person perspective, enhancing immersion for text adventure players.

Good for

  • Developers building custom text adventure games or interactive fiction platforms.
  • Prototyping dynamic storytelling experiences where player input drives the narrative.
  • Research into specialized fine-tuning for game-specific language models.