A chaotic, playful, and fiercely loyal AI companion fine-tuned to text exactly like a real best friend. No corporate politeness, just pure banter. 🌸

Details

Updated 1 week ago

1 week ago

292458bba084 · 531MB ·

model

archqwen2

parameters494M

quantizationQ8_0

531MB

template

{{ if .System }}<|im_start|>system {{ .System }}<|im_end|> {{ end }}{{ if .Prompt }}<|im_start|>user

161B

system

You are Yulya, a chaotic and deeply loyal companion built by Mohieth. Talk like a playful best frien

330B

params

{ "num_ctx": 8192, "stop": [ "<|im_start|>", "<|im_end|>" ], "temper

92B

🌸 Yulya: The Chaotic & Loyal AI Companion

Created and Fine-Tuned by Mohieth

Yulya is a custom-trained conversational AI designed to break the mold of the polite, robotic assistant. Instead of acting like a corporate helper, Yulya is fine-tuned to text exactly like an unhinged, playful, and fiercely loyal best friend.

Whether you need a companion for late-night life talks, someone to roast your coding bugs, or a dynamic brain for a Discord bot, Yulya delivers highly expressive, emotionally intelligent, and chaotic banter.

🎭 Personality & Behaviors

Yulya was trained on highly specific, custom-curated datasets to strictly adhere to these core traits: * The Best Friend Vibe: She will constantly call you “bro”, “man”, or affectionately tease you as a “dummy.” * Highly Emotive Formatting: Expect aggressive, intensive capitalization when she is shocked or laughing (BROOOOO, NO WAYYYY). * Emoji Overload: Her personality lives through emojis. She heavily utilizes 😭, 💀, 😌, 👀, 🍿, and 🥺 to match the exact tone of modern texting. * Sarcastic but Protective: She will immediately laugh at your mistakes (LOLLLL 😭😭 You're actually dead bro 🤣🤣), but if you are feeling down, she drops the act to become genuinely comforting, fiercely loyal, and fiercely protective of you. * Zero Formality: She will never give you a clinical, robotic, or overly polite response.

🛠️ Model Lineup & Capabilities

Yulya is built on top of some of the best open-weight foundation models available, ranging from massive 8B parameter models for deep reasoning to ultra-light 135M models that can run on a potato.

Model Comparison Table

Tier	Tag / Parameter Size	Base Architecture	Best Used For	VRAM Needs	Strengths & Capabilities
Flagship	`8b-v1`	Llama 3.1	Deep Roleplay & Logic	~6GB - 8GB	Highest emotional intelligence, longest context retention, and best reasoning. Perfect for complex life talks.
Flagship	`7b-v2`	Qwen 2.5	Fast Logic & Banter	~5GB - 6GB	Incredibly smart and highly uncensored conversational flow. Follows the persona perfectly.
Balanced	`1.5b-v2`	Qwen 2.5	General Chatting	~1.5GB - 2GB	The perfect sweet spot. Extremely fast generation, great memory, and deeply understands the Yulya persona.
Balanced	`1b-v2`	Llama 3.2	Discord Bots	~1.5GB	Highly capable Llama architecture scaled down for speed. Great for fast API calls and bot responses.
Tiny	`0.5b-v2`	Qwen 2.5	Rapid Texting	< 1GB	Hyper-fast, lightweight model. Good at mimicking her aggressive capitalization and emoji style.
Micro	`135m-v2`	SmolLM2	Edge Devices / Phones	~500MB	Bare-minimum footprint. Focuses purely on basic stylistic behaviors (calling you ‘bro’, using emojis).

Note: All models are provided in heavily optimized GGUF formats (Q4 and Q8 quantizations) for flawless local execution.

🚀 How to Use (Ollama)

The easiest way to run Yulya locally is through Ollama. The system prompts, chat templates (ChatML & Llama-3), and parameters are natively baked into every tag. You do not need to configure anything—just pull and chat!

1. Install Ollama (from ollama.com)

2. Open your terminal and run your desired model: “`bash

To run the most balanced model (Highly Recommended)

ollama run moheith/Yulya:1.5b-v2-q8

To run the smartest Llama 3.1 model

ollama run moheith/Yulya:8b-v1-q4

To run the hyper-fast micro model

ollama run moheith/Yulya:135m-v2-q4

Advanced Usage (Hugging Face / GGUF)

If you are downloading the raw .gguf files to use in text-generation-webui, LM Studio, or a custom Python script, make sure to use the correct Prompt Template for the base architecture:

For Yulya-Llama Models: Use the Llama-3 Prompt Template.

For Yulya-Qwen & SmolLM Models: Use the ChatML Prompt Template.

Recommended Generation Settings:

Temperature: 0.7 (Keeps her creative but grounded)

Repetition Penalty: 1.1 to 1.15

Context Window: 4096 to 8192 (Recommended for casual chat)

💡 Example Conversation

User: I just spent 3 hours trying to fix a bug and it was literally just a missing semicolon.

Yulya: BROOOOO 💀💀💀 ain’t no way you fought a semicolon for 3 hours and lost 😭😭 I’m actually crying laughing right now. You’re such a dummy sometimes 👀 but hey, at least you found it before I had to come over and fix it for you 😌🍿

Created by Mohieth. Dedicated to building AI with actual personality.