README.md

metadata

title: AI Character Chat
emoji: 🌼
colorFrom: blue
colorTo: green
sdk: gradio
sdk_version: 5.49.1
app_file: app.py
pinned: false
license: mit
tags:
  - chatbot
  - lora
  - peft
  - character-ai
  - text-to-speech
models:
  - Qwen/Qwen2-0.5B-Instruct
python_version: '3.10'

🌼 AI Character Chat 🌼

A Beautiful Pastel-Themed Multi-Character Chatbot with Voice Output

Chat with three distinct AI personalities, each fine-tuned using LoRA adapters, in a serene pastel forest and sky blue interface. Features real-time text-to-speech for an immersive experience!

✨ Features

• 🎨 Beautiful Pastel UI: Calming forest green and sky blue color scheme
• 🎭 Three Unique Characters: Each with distinct personalities and speaking styles
• 🔊 Voice Output: Toggle text-to-speech to hear your character's responses
• 💬 Real-time Chat: Instant responses with conversation history
• 🎯 LoRA Fine-tuning: Efficient parameter training for unique personalities

🎭 Characters

🌼 JARVIS - Sophisticated AI Assistant

• Personality: Professional, articulate, British butler-like
• Style: Formal precision with elegant phrasing
• Best for: Professional assistance, detailed explanations, refined conversation

🪄 The Wizard - Mystical Forest Wizard

• Personality: Whimsical, magical, poetic
• Style: Uses metaphors, arcane language, and mystical wisdom
• Best for: Creative inspiration, philosophical discussions, enchanting storytelling

🌿 Sarcastic - Witty and Sharp

• Personality: Sarcastic but helpful
• Style: Quick wit with playful teasing
• Best for: Fun conversations, honest feedback with humor, keeping things light

🛠️ Technical Details

Model Architecture

• Base Model: Qwen/Qwen2-0.5B-Instruct (494M parameters)
• Fine-tuning Method: LoRA (Low-Rank Adaptation)
• LoRA Configuration:
  • Rank (r): 16
  • Alpha: 32
  • Target modules: q_proj, v_proj, k_proj, o_proj
  • Trainable parameters: ~2.16M (0.44% of base model)

Training Details

• Each character trained on 20 custom examples
• Dataset augmented 50x for robust learning (1,000 examples per character)
• 3 training epochs per character
• Learning rate: 2e-4
• Batch size: 2 with gradient accumulation
• Optimizer: AdamW

Voice Synthesis

• TTS Engine: Google Text-to-Speech (gTTS)
• Language: English
• Autoplay: Enabled by default
• Toggle: Can be disabled via checkbox

🚀 How to Use

1. Select Your Character: Choose from JARVIS, Wizard, or Sarcastic using the radio buttons
2. Type Your Message: Enter your question or prompt in the text box
3. Send: Click "Send" or press Enter
4. Listen (Optional): Enable "🔊 Enable Voice Output" to hear responses
5. Start Fresh: Click "New Conversation" to clear chat history

🎨 Design Features

• Color Palette:

  • Background: Pastel Forest Green (#A9C8A6)
  • Accent: Pastel Sky Blue (#9DD1F5)
  • Borders: Blue (#7BB8E0)
  • Cards: Clean White (#FFFFFF)

• Typography: Quicksand font family for a soft, friendly feel

• Layout: Responsive two-column design with separate input and chat areas

📦 Dependencies

gradio>=5.49.1
torch>=2.0.0
transformers>=4.30.0
peft>=0.4.0
gtts>=2.3.0
accelerate>=0.20.0

🔧 Local Development

# Clone the repository
git clone https://github.com/AliceInWonderland61/lora-characters.git
cd lora-characters

# Install dependencies
pip install -r requirements.txt

# Run the app
python app-2.py

The app will launch at http://localhost:7860

📚 Dataset

Each character was trained on custom JSONL datasets featuring:

• JARVIS: Professional, helpful, sophisticated responses
• Wizard: Mystical, poetic language with magical metaphors
• Sarcastic: Witty, clever replies with playful teasing

Datasets were carefully crafted to ensure distinct personalities while maintaining helpfulness.

🎯 Use Cases

• Customer Service Training: Test different communication styles
• Creative Writing: Get responses from different character perspectives
• Entertainment: Enjoy varied conversational experiences
• Education: Learn how personality affects AI responses
• Prototyping: Test multi-character chatbot concepts

🤝 Contributing

Feedback and suggestions are welcome! Feel free to:

• Report bugs or issues
• Suggest new character personalities
• Propose UI improvements
• Share your experience

📖 References

• LoRA: Low-Rank Adaptation of Large Language Models
• PEFT Library Documentation
• Transformers Library
• Gradio Documentation
• gTTS Documentation

📄 License

MIT License - Feel free to use and modify!

🔗 Links

• 🤗 Hugging Face Space: https://huggingface.co/spaces/AlissenMoreno61/Lora-Character
• 💻 GitHub Repository: https://github.com/AliceInWonderland61/lora-characters
• 📓 Google Colab Notebook: https://colab.research.google.com/drive/1LFPxNvL7gchaunTErzcrKbodGFt562yA

---

Built with ❤️ using:

• 🤗 Hugging Face Transformers
• 🎯 LoRA Fine-tuning
• 🎨 Gradio Interface
• 🔊 Google Text-to-Speech
• 🌼 Pastel Design Philosophy

Created by: Alissen Moreno