chimbiwide/Gemma4NPC-E2B
chimbiwide/Gemma4NPC-E2B is a 5.1 billion parameter language model based on the Gemma4-E2B architecture, fine-tuned by chimbiwide. This model is specifically optimized for game-focused Non-Player Character (NPC) roleplaying, trained on the RolePlay-NPC-Quest dataset. It demonstrates improved instruction-following capabilities compared to previous Gemma iterations, making it suitable for generating authentic and conversational NPC interactions.
Loading preview...
Model Overview
chimbiwide/Gemma4NPC-E2B is a 5.1 billion parameter model built upon the Gemma4-E2B base, specifically fine-tuned for game-focused Non-Player Character (NPC) roleplaying. This model leverages the RolePlay-NPC-Quest dataset to enhance its ability to generate realistic and engaging NPC interactions.
Key Capabilities
- Game-focused NPC Roleplaying: Designed to act as an NPC within game environments, providing conversational and authentic responses.
- Improved Instruction Following: Demonstrates a significant step up in instruction-following capabilities compared to earlier Gemma3/3n models.
- Contextual Interaction: Capable of referencing character background, current surroundings, and reacting to player input.
Training Details
The model was trained as an r=32, alpha=64 LoRA adapter over 2 epochs using the RolePlay-NPC-Quest dataset. Training parameters included a learning rate of 1e-4, an effective batch size of 32, a cosine learning rate scheduler with a 500-step warmup, and a gradient clipping of 1.0.
Recommended Usage
For optimal performance, users should employ specific inference settings and system prompts. Recommended settings include temp = 1.0, top_p = 0.95, and top_k = 64. The model is designed to respond to a "Greetings" prompt as its first user input, mimicking in-game NPC behavior.