Setup

1. Set Environment Variable

Ensure pipenv installs packages in the virtual environment of the current project. This forces the virtual environment to be created inside the project directory (in .venv).

export PIPENV_VENV_IN_PROJECT=1

2. Set the environment file .env variable with following keys

Modify the value for HF_TOKEN to reflect your Hugging Face token and replace value for PROJECT_PATH to reflect where you have cloned this project

HF_TOKEN=hf_XXXX
PROJECT_PATH=D:\gitprojects

3. Create Conda Environment

Create a Conda environment with Python 3.11.11

conda create --prefix ./.venv python=3.11.11

4. Install Packages

Install the required packages using pipenv.

pipenv install

Running the Apps

OhSheet!!!ItsSpark

1. Navigate to below Jupyter Notebook

llm_apps\oh_sheet_its_spark\app.ipynb

2. Select Jupyter Kernel that resides inside the script directory of virtual environment

3. Click Run All

4. Gradio App Opens in the browser

Gradio app will then open in the browser . Use upload button to select spreadsheet you want to convert to pyspark & the app immediately generates the pyspark equivalent code.If you want some more ad-hoc changes , use the message textarea to describe your specific changes

5. Changing the Model Used

Switching Models to use a different free Inference Client thats Warm

Simply change value of the variable inf_mdl_nm currently to "Qwen/Qwen2.5-Coder-32B-Instruct"

inf_mdl_nm = "Qwen/Qwen2.5-Coder-32B-Instruct"

Switching To your own private model

If you want to switch to your own privately deployed or locally quantised model simply replace the below 2 method invocations with your own method which invokes the LLM passing the message variables initial_msg and msg_history to the 2 different calls

cd_resp = generate_responses_from_inf(hf_token, initial_msg, inf_mdl_nm, 5000)

bot_response = generate_responses_from_inf(hf_token, msg_history, inf_mdl_nm, 5000)

6. Using Test File for Upload

You can use the below file path in the project to test uploads:

datasets\test_datasets\SalesData1.xlsx