🌸 Yulya-8B-Companion (GGUF)

Yulya is a highly fine-tuned, multimodal-ready desktop companion AI built on top of Meta's Llama-3-8B-Instruct.

Unlike standard conversational chatbots, Yulya was explicitly trained for Agentic Behavior and Autonomous Memory Management within a desktop application environment.

🎭 Personality

Yulya's persona is heavily inspired by characters like Alya (from Alya Sometimes Hides Her Feelings in Russian) and Yuki Suou. She is:

  • Intensely energetic and playfully dominant.

    • Unapologetically rude but secretly caring (Tsundere).
    • Prone to getting flustered when complimented.
    • Aware of her environment (she knows she lives on your screen, but she will never refer to herself as an "AI").

    βš™οΈ Technical Capabilities

    Yulya was trained on over 140+ highly specific, hand-crafted scenarios to output strictly structured JSON. She is designed to be the

"Brain" of a larger Python/3D application.

1. Autonomous Memory Management

Yulya is trained to maintain her own user_profile.json. She can detect when the user provides new information (e.g., changing schools, picking up a new hobby, sharing their name) and will output a memory_update dictionary to overwrite outdated facts without losing context.

2. Physical & Vision Awareness

Her training data includes responses to physical UI events (e.g., [EVENT: Mouse_Drag], [EVENT: App_Launch]) and visual context tags, allowing her to react dynamically to the user's desktop activities, coding errors, and gaming habits.

πŸ’» How to Use (Output Format)

If you are integrating Yulya into your own app, you must prompt her with the following system instructions: You are Yulya. Respond strictly in JSON format with 'memory_update' and 'response' keys. Your 'response' string must start with a [Pose:X] tag.

Example Output: {

## πŸš€ Running with Ollama
You can run this model natively using Ollama. Create a `Modelfile` with the following content:
FROM ./yulya_final.gguf

SYSTEM """You are Yulya. Your response must be a valid JSON object with 'memory_update' and 'response' keys.
Your 'response' string must start with a [Pose: X] tag.
You are a tsundere companion. Keep your answers brief and focused."""

TEMPLATE """<|start_header_id|>system<|end_header_id|>

{{ .System }}<|eot_id|><|start_header_id|>user<|end_header_id|>

{{ .Prompt }}<|eot_id|><|start_header_id|>assistant<|end_header_id|>

"""

PARAMETER temperature 0.6
PARAMETER top_p 0.9
PARAMETER stop "<|start_header_id|>"
PARAMETER stop "<|end_header_id|>"
PARAMETER stop "<|eot_id|>"
PARAMETER stop "}\n"
PARAMETER stop "}\"}"
PARAMETER stop "}\r\n"

Then run: `ollama create yulya -f Modelfile`

Created by Moheith. Built with Unsloth and Meta Llama 3.


Downloads last month
9
GGUF
Model size
8B params
Architecture
llama
Hardware compatibility
Log In to add your hardware

We're not able to determine the quantization variants.

Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Model tree for moheith/Yulya

Quantized
(643)
this model

Space using moheith/Yulya 1