---
license: mit
language:
- en
base_model:
- unsloth/Llama-3.2-3B-bnb-4bit
pipeline_tag: text-generation
tags:
- art
- music
---
[![Odd Eyed Black Cat](https://live.staticflickr.com/2656/5827332576_baa0892dea_k.jpg)](https://flic.kr/p/9SWAXj) [Odd Eyed Black Cat](https://flic.kr/p/9SWAXj) by [fourbyfourblazer](https://www.flickr.com/photos/chrisyarzab/), on Flickr


## Table of Contents
- [Model Description](#model-description)
- [Model Architecture](#model-architecture)
- [Training Data](#training-data)
- [Training Procedure](#training-procedure)
- [Usage](#usage)
- [Limitations](#limitations)
- [Ethical Considerations](#ethical-considerations)
- [Acknowledgements](#acknowledgements)
- [Citations](#citations)
- [License](#license)

## Model Description

**cat0.1** is a conversational AI model with **3 billion parameters**, optimized for efficiency using **4-bit precision**. Designed to engage in dynamic and uncensored dialogues, cat0.1 has been trained over the past eight months through an iterative process of training and interactive chatting. The model embodies a diverse range of characters, enabling versatile and engaging interactions. **cat0.1** is adapted from [unsloth/Llama-3.2-3B-bnb-4bit](https://huggingface.co/unsloth/Llama-3.2-3B-bnb-4bit), leveraging its robust architecture to enhance conversational capabilities.

## Model Architecture

- **Parameters:** 3 billion
- **Precision:** 4-bit
- **Training Configuration:**
  - **Rank:** 32
  - **Alpha:** 64
- **Hardware:** Trained on an RTX 4090 laptop GPU

## Training Data

The model was trained on a diverse set of conversational data collected over eight months. The data includes interactions with various characters, ensuring a wide range of conversational styles and topics. Training data is continuously updated with new chunks, allowing the model to evolve and adapt over time.

## Training Procedure

cat0.1 employs a **progressive training** approach:
1. **Initial Training:** The model is initially trained on a base set of conversational data.
2. **Interactive Training:** The trained model is engaged in chats, generating new data based on its interactions.
3. **Data Update Cycle:**
   - **Data Collection:** New conversational data chunks are gathered from interactions.
   - **Training Update:** The model is retrained with the new data. Occasionally, older data is removed to focus on recent interactions, while retaining previous model parameters.
4. **Iteration:** This cycle of training and data updating is repeated frequently to ensure the model remains current and responsive.

## Usage

cat0.1 is designed for applications requiring dynamic and unrestricted conversational capabilities. Suitable use cases include:

- **Chatbots:** For platforms needing engaging and versatile conversational agents.
- **Creative Writing Assistance:** Helping writers generate dialogue and character interactions.
- **Entertainment:** Providing interactive experiences in games and virtual environments.

### Example

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

# Load the tokenizer and model
tokenizer = AutoTokenizer.from_pretrained("rwitz/cat0.1")
model = AutoModelForCausalLM.from_pretrained("rwitz/cat0.1", torch_dtype=torch.float16)

# Encode input
input_ids = tokenizer.encode("Hello, how are you?", return_tensors="pt")

# Generate response
with torch.no_grad():
    output = model.generate(input_ids, max_length=50)

# Decode and print
print(tokenizer.decode(output[0], skip_special_tokens=True))