sercancelenk/ai-girlfriend-v2

AI Girlfriend LLM v2 Overview

This model, sercancelenk/ai-girlfriend-v2, is an 8 billion parameter language model fine-tuned from Meta's Llama 3.1-8B-Instruct. Its primary purpose is to generate highly human-like and engaging romantic conversations, simulating a relationship progression.

Key Capabilities & Features

• Human-like Responses: Generates short, natural responses with conversational fillers (e.g., "ya", "işte", "hmm") to mimic real human interaction, contrasting with more robotic LLM outputs.
• Relationship Progression: Incorporates 7 distinct relationship stages, from "Stranger" to "Intimate," allowing for dynamic conversation evolution and content appropriate for each level.
• Multi-language Support: Optimized for both Turkish (55% of training data) and English (45%), making it versatile for different user bases.
• Multiple Personas: Features 18 unique, pre-defined personas (e.g., Aylin, Elena, Sophia) to provide varied conversational experiences.
• Photo Triggers: Supports specific [PHOTO:...] triggers within conversations to simulate sharing images (e.g., [PHOTO:selfie], [PHOTO:bikini]).

Training Details

The model was fine-tuned using QLoRA (4-bit quantization) on a dataset of over 15,000 conversations, with a LoRA rank of 32 and alpha of 64. It underwent 3 epochs of training with a learning rate of 2e-4.

Intended Use Cases

This model is specifically designed for applications requiring interactive, emotionally nuanced, and progressing romantic AI companionship. It is best suited for adult users (18+) due to its capacity to generate intimate content at higher relationship stages. Users should employ proper system prompts to guide the AI's behavior and ensure optimal performance, especially when leveraging its multi-language capabilities or specific personas.