Frontend: Chatbot

Overview

This project involves creating a frontend website interface that interacts with a fine-tuned Mistral model hosted on Ollama. The interface sends POST requests to the model to generate specialized responses to instructions related to the Artificial Intelligence and Data Act (AIDA).

Unfinished work

The Home, Contact and About page are not finished.

Code: Dataset Creation using Local Language Models

This project involves creating datasets using local language models. It provides a way to generate structured data based on input text, with a focus on summarizing content and generating instructions and responses.

Overview

The project consists of several Python scripts that work together to generate datasets by interacting with local language models. The key scripts are:

1. connect_model.py: Handles the interaction with the local language model API.
2. data_set_creation.py: Contains functions to format, save, and manage datasets.
3. process_text.py: Split the text into smaller paragraphs.
4. main.py: The main driver script that coordinates the dataset creation process.

Files and Functions

connect_model.py

This module is responsible for connecting to a local language model and sending prompts to it.

• request_model(model, prompt):
  • Sends a prompt to a specified language model and retrieves the response.
  • Parameters:
    • model: The name of the model to use.
    • prompt: The text prompt to send to the model.
  • Returns: The generated response from the model or an error message.

data_set_creation.py

This module contains various utility functions for dataset creation and management.

• format_data(model_response):

  • Adds an 'input' key with an empty string as the value to each dataset entry.
  • Parameters:
    • model_response: A JSON string representing a list of dataset dictionaries.
  • Returns: A list of dictionaries with an additional 'input' key.

• read_data(save_path):

  • Reads JSON data from a file.
  • Parameters:
    • save_path: The path to the JSON file.
  • Returns: The data from the file.

• save_data(dataset, file_path, save_path):

  • Saves the dataset by extending the existing data in a JSON file.
  • Parameters:
    • dataset: A list of dictionaries to be saved.
    • file_path: The path to save the updated dataset.
    • save_path: The path to the existing data file.

• confirm_data_format(dataset):

  • Ensures that the dataset string is properly formatted as a JSON array.
  • Parameters:
    • dataset: The dataset string that needs formatting.

• create_input(model, dataset, content):

  • Generates 'input' key values for each dataset entry based on provided content.
  • Parameters:
    • model: The model to use for generating inputs.
    • dataset: A JSON string representing a list of dataset dictionaries.
    • content: The content used for generating 'input' values.
  • Returns: The updated dataset with generated 'input' values.

process_text.py

This module contains functions for processing and breaking down text into manageable components.

• break_down_text():

  • Breaks down the text into a single string and stores its length in a list.
  • Parameters: None.
  • Returns: A list where the first element is the full text as a string and the second element is the length of the text.

• text_to_list():

  • Splits the text into a list of paragraphs.
  • Parameters: None.
  • Returns: A list of paragraphs as strings with leading and trailing whitespace removed.

main.py

The main driver script that coordinates the entire dataset creation process.

• main():
  • The primary function that controls the flow of the program.
  • It allows the user to choose between different models (mistral or gemma2) and generates datasets by summarizing paragraphs of text.
  • It divides the text into paragraphs, generates instructions and responses using the selected model, and saves the datasets to a JSON file.

Data

• aida_data_gemma2.json: Gemma2 generated dataset.
• aida_data_mistral.json: Mistral generated dataset.
• aida_data_gpt.json: ChatGPT generated dataset.
• aida_data_merged.json: Merged gemma2, Mistral and ChatGPT datasets into one. Further processed the dataset with AI to remove redundancy. Added "The Artificial Intelligence Data Act (AIDA):" to any instructions that did not contain "AIDA" in the string.
• text_to_summarize.txt: Text containing the web document of the Artificial Intelligence Data Act (AIDA).

The datasets generated by the 3 models might contain inaccurate information. If there was more time, I would have liked to validate each of the labelled data to confirm its accuracy and be more selective about which data is used to fine-tune the model.

Usage

1. venv Setup: Ensure virtual environment is setup.

2. Model Setup: Ensure the local language model API is running and accessible at the specified URL in connect_model.py.

3. Run the Script: Execute main.py to start the dataset creation process. You can choose between the mistral or gemma2 models by setting the mistral variable in the main() function.

4. Output: The generated datasets will be saved in JSON format in the specified file paths.

Requirements

• Python 3.x
• requests library for making HTTP requests.

Notes

• Ensure that the local language model API is configured correctly and accessible.
• The text to be summarized should be available in the specified file path (../data/text_to_summarize.txt).

License

This project is licensed under the MIT License. See the LICENSE file for details.