MistralAI Questionnaire

This project provides a toolkit for generating questionnaire from documents: [txt, docx, pdf] to .csv dataset format.

Requirements

Before starting, you need to install the following libraries: .. code-block:: python

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu124

• langchain
• langchain_community
• langchain-huggingface
• playwright
• html2text
• sentence_transformers
• faiss-cpu
• pandas
• peft==0.4.0
• trl==0.4.7
• pypdf
• bitsandbytes
• accelerate

Description

ModelManager

This class is responsible for loading mistralai model and generating QA.

Constructor
^^^^^^^^^^^

.. code-block:: python

   __init__(self, model_name)

- **model_name**: The path or name of the pre-trained model.


Methods
^^^^^^^

- **setup_tokenizer()**: Loads and configures the tokenizer for the model.
- **setup_bitsandbytes_parameters()**: Configures parameters for bit quantization (BitsAndBytes).
- **from_pretrained()**: Loads the model with pre-trained weights and quantization configuration.
- **print_model_parameters(examples)**: Prints the number of trainable and total parameters of the model.
- **__call__(self, *args, **kwargs)**: The main method for running the generate tasks.

Usage
-----

To start generating QA, you should create an instance of the ``ModelManager`` class and call its ``__call__`` method, passing the necessary arguments.

.. code-block:: python
   from questionnaire_mistral.models import ModelManager
   model = ModelManager(model_name="path_to_model")
   model(document=document, task=task, document_content=document_content, task_count=task_count)

License
-------

The project is distributed under the MIT License.