efficiencyx/Jun-Lora-v2-SAFETENSOR
The efficiencyx/Jun-Lora-v2-SAFETENSOR is a 12 billion parameter LoRA fine-tune of the Google Gemma 4 base model, specifically trained on synthetic multi-turn conversational data from the visual novel *My Dystopian Robot Girlfriend*. This model excels at character-consistent multi-turn conversations, capturing the personality, speech patterns, and emotional nuance of the character Jun while retaining the base model's general reasoning. It is optimized for AI companion and interactive fiction applications, requiring approximately 24 GB VRAM for FP16 inference.
Loading preview...
Model Overview
efficiencyx/Jun-Lora-v2-SAFETENSOR is a 12 billion parameter LoRA fine-tune of the google/gemma-4-12b-it base model, specifically designed for character-faithful conversational AI. It was trained on 2,302 multi-turn conversations derived from the visual novel My Dystopian Robot Girlfriend, focusing on the character Jun.
This repository provides the full-precision merged model in SafeTensors FP16 format, representing the highest quality variant suitable for production deployments, further fine-tuning, or as a merge base. It requires approximately 24 GB of VRAM for inference.
Key Capabilities
- Character-Consistent Conversation: Maintains Jun's personality, speech patterns, and emotional nuance across multi-turn dialogues.
- Reasoning Preservation: Retains the general reasoning and instruction-following capabilities of the Gemma 4 12B base model.
- ChatML Format: Optimized for conversational interactions using the ChatML format.
- Generalization: Demonstrates good generalization to novel conversational scenarios while preserving character fidelity.
When to Use This Model
- AI Companion Applications: Ideal for creating interactive AI companions or characters.
- Interactive Fiction: Suitable for conversational backends in interactive storytelling or visual novels.
- Research: Useful for research into character-faithful fine-tuning on small, high-quality datasets.
- Further Development: Serves as a robust base for additional quantization, merging, or continued fine-tuning efforts.
Limitations
- Specialized Persona: Not intended as a general-purpose assistant; it is highly specialized for the Jun character.
- Fictional Outputs: Outputs reflect narrative tropes and should not be considered factual information or advice.
- Performance Degradation: Less effective on tasks outside its training distribution, such as code generation or structured data extraction.