🤖 pistoBot

Create an AI that chats like you.

🥜 In a nutshell

1. Get your whatsapp and telegram data
2. Parse it using this
3. Train one of the available models (gpt2 suggested)
4. Chat with the model

👀 Example of chat

An example of GPT-2 model trained on my whatsapp and telegram messages.
See the website for more examples.

Chat:

✏️ come sei messo col pistobot?
🤖 ahaha male
✏️ chatta meglio di te? 😂
🤖 si
✏️ non che ci volesse molto...
🤖 ma tu che dici
✏️ io dico che potevi impegnarti di più

✏️ ⟶ message that I have wrote
🤖 ⟶ message generated from the model

👉 I want one

• Create and chat with your model using this colab notebook
• For more info visit the project website

---

⚠️ Disclaimer

This project is only a personal playground build during the week-ends of Covid-19 quarantine.
Used mainly to:

• Use interesting packages (like gpt-2-simple, tensorflow 2.0 etc.)
• Gain experience

Due to this nature, this repository has probably:

• ML naif approaches
• For sure: not so good english text (sorry 😢)

---

📝 Note

• Thanks to Salvinator: Under Covid 19 quarantine I found this project, that had inspired me to start this repository.
• Why pistoBot name?
  • 🔮 The answer could be handed down only by voice

Run Docker

1. Go to the telegram bot directory

cd telegramBot

2. Make sure you have the model files config.json and pytorch_model.bin in the model directory

3. Docker build

docker build -t zubot:latest .

4. Docker run

docker run -e BOT_TOKEN='add bot token' --name zubot zubot:latest

send messages to your telegram bot and enjoy