Make-AI-Clone-of-Yourself

Full Video Tutorial on Youtube - https://youtu.be/a2_ZvzE55cA

Motivation

I saw a reel on Instagram in which an AI enthusiast created an AI clone of himself to talk to his girlfriend (Certainly, I won't do that... xd) using RAG Retrieval Augmented Generation and the Chat GPT-3.5 turbo API. It kind of worked, but it had major privacy issues. Sending personal chats to Chat GPT could potentially result in those chats being used by OpenAI to train its model. This led me to think, what if I fine-tuned a pre-existing model like Llama or Mixtral on my personal WhatsApp chat history? It would be cool to have a model that can talk like me as I do on WhatsApp, primarily in Hinglish (Hindi + English).

Fine-tuning a large language model (LLM) requires a good understanding of LLMs in general. The major challenge with fine-tuning big models, such as a 7B parameter model, is that it requires a minimum of 32GB of GPU RAM, which is costly and not available in free-tier GPU compute services like Colab or Kaggle. So, I had to find a way around this limitation.

Another important aspect is that the fine-tuning results heavily depend on the quality and size of the dataset used. Converting raw WhatsApp chat data into a usable dataset is challenging but worth pursuing.

Let's see how it looks in reality and how it's being carried out.

Google Colab Notebook

Here is link of Google Colab Notebook

• This Notebook Contains -
  • Data Collection From Whatsapp
  • Data Prepration
  • Data Filtering
  • Model Training
  • Inference
  • Saving Finetuned Model
  • GGUF Conversion
  • Downloading Saved Model

Chatting With Fine-Tuned Model Using Ollama (You can also use LM Studio & GPT4ALL)

Downloading Ollama

Go to the downloads page of Ollama and download and install it according to your OS.

Loading the Model into Ollama

1. Open your file manager.
2. Navigate to the directory where you have downloaded the fine-tuned model, generally the Downloads folder.
3. Right-click anywhere on the screen and choose "Open terminal here." If you are using Windows, you can directly type cmd into the address bar and hit enter.
4. Type ollama --version into the terminal to check if you have installed Ollama successfully.
5. Make sure the model unsloth.Q8_0.gguf and Modelfile are in the current directory.
6. Open Modelfile using any text editor.
7. Edit the first line to ensure it looks like FROM ./unsloth.Q8_0.gguf, and save it.
8. Type this command: ollama create my_model -f Modelfile. This will add the model into Ollama.
9. Now type ollama run my_model.
10. You can now chat with your model.

Using Model to Automate WhatsApp

Here comes the final part. I am using this wonderful tool WPP_Whatsapp to automate WhatsApp. By using this, we can use our model to respond to any incoming messages on WhatsApp. We can define specific people to talk to. Here is the step-by-step guide:

1. Clone this repo using: git clone https://github.com/Eviltr0N/Make-AI-Clone-of-Yourself.git

2. Go to the cloned repo: cd Make-AI-Clone-of-Yourself/

3. Install all the required packages by: pip install -r requirements.txt

4. First run: python3 chat.py

   Then type something like "Hello" and hit enter. If it works, it means everything is set up correctly.

5. Exit by pressing Ctrl+C.

6. Now run: python3 ai_to_whatsapp.py

7. It will take a bit of time at first to download the Chromium browser. As it finishes, a browser window will appear, and you have to scan the QR code using WhatsApp to link your WhatsApp account.

8. Switch back to the terminal. You have to provide the phone number of the other person you want to respond to with this AI model.

9. The phone number must include the country code without the + symbol, such as 916969696969, then press enter.

10. As soon as that person sends any message, it will be printed on the terminal, and the AI model will respond to it.

Keep in Mind

• If you want to change the temperature and top_k of the model (in simpler terms, temperature means creativity of the model), then:

  1. Open the ai_to_whatsapp.py file using any text editor.

  2. Go to the 9th line where you will find: my_llm = LLM("my_model", 0.3, 50, 128)

  3. Change 0.3 to a higher value such as 0.9 to increase the temperature (creativity) of the model.

  4. You can also change top_k from 50 to a higher value like 95.

• Temperature can vary from 0.1 to 1 and top_k from 1 to 100. The higher the temperature, the more creative and unpredictable the model becomes.

• Keep playing with the values of temperature and top_k until you are satisfied with the model's responses.

What's Next

• Adding multimodal capabilities so this can understand and react to images too, as it currently supports only text messages.

• Adding some sort of agent pipeline above this model so another model, such as LLaMA 3.1, can see the message and the response of this model and judge if it is suitable or not, and can modify the model response accordingly.