sangwon1472/gemma4-e2b-mud

VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:5.1BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Apr 9, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The sangwon1472/gemma4-e2b-mud model is a 5.1 billion parameter Gemma 4 E2B-based language model developed by sangwon1472, specifically fine-tuned for Korean spacefaring text MUD (Multi-User Dungeon) style interactions. It excels at generating short command responses, NPC dialogue, room descriptions, lore explanations, and navigation hints, focusing on the narrative layer of game experiences. This model is optimized for maintaining in-world responses rather than general assistant-style answers, supporting a 32768 token context length.

Loading preview...

Overview

gemma4-e2b-mud is a 5.1 billion parameter model derived from google/gemma-4-E2B-it, specifically adapted for Korean text MUD (Multi-User Dungeon) scenarios. It focuses on generating narrative elements like NPC dialogue, atmospheric room descriptions, and lore explanations, rather than game engine logic. The model is distributed as a Transformers checkpoint, a GGUF file for local inference (e.g., LM Studio), and a comprehensive Colab starter package for reproduction and further experimentation.

Key Capabilities

  • In-world responses: Designed to stay within the game's context, avoiding meta-commentary or general knowledge spills.
  • Narrative generation: Excels at creating short, clear NPC dialogues, vivid room descriptions, and concise lore based on in-game proper nouns.
  • Korean language focus: Primarily developed for Korean text-based game environments.
  • Comprehensive package: Includes the model, GGUF for easy local deployment, and Colab notebooks with example datasets for fine-tuning.

Good for

  • Text MUD development: Ideal for prototyping narrative layers, NPC interactions, and world-building in Korean text-based games.
  • Atmosphere and lore: Generating short rumors, lore, and signal responses that enhance game immersion.
  • Local experimentation: The provided GGUF and Colab starter package facilitate quick local testing and re-training.

Not for

  • Game logic: Not intended for handling game state, quest completion, combat calculations, or reward systems.
  • General assistant tasks: Avoids providing encyclopedic knowledge or acting as a general-purpose chatbot.
  • High-stakes information: Not suitable for legal, medical, or financial advice.

Finetunes
1 models