Tax Law GPT

Tax Law GPT

Overview

This project fine-tunes an OpenAI model using the Internal Revenue Code (IRC) to assist accountants and tax specialists in understanding and applying tax regulations. By leveraging the fine-tuned model, professionals can ask complex questions related to the IRC and receive accurate, context-specific responses that cater to their needs.

Project Structure

• train.jsonl: The training dataset containing examples of IRC-related text and corresponding expected model outputs.
• test.jsonl: The validation dataset used to evaluate the model's performance during fine-tuning.

Prerequisites

• Python 3.8 or higher
• OpenAI Python Client Library (openai package)
• OpenAI API key

Installation

1. Install the necessary Python packages:

Copy
pip install openai

2. Set your OpenAI API key as an environment variable or replace 'your_api_key' in the code with your actual API key.

Fine-Tuning the Model

1. Upload Training and Validation Files: The first step involves uploading the training and validation files (train.jsonl and test.jsonl) to OpenAI. These files must be in JSON Lines format.

Copy
import openai

# Set your API key
openai.api_key = 'your_api_key'

# Upload training file
with open("train.jsonl", "rb") as train_file:
    training_file_response = openai.files.create(
        file=train_file,
        purpose='fine-tune'
    )
training_file_id = training_file_response.id
print(f"Training file uploaded with ID: {training_file_id}")

# Upload validation file
with open("test.jsonl", "rb") as validation_file:
    validation_file_response = openai.files.create(
        file=validation_file,
        purpose='fine-tune'
    )
validation_file_id = validation_file_response.id
print(f"Validation file uploaded with ID: {validation_file_id}")

2. Create the Fine-Tuning Job: After uploading the files, create the fine-tuning job using the file IDs obtained from the upload process. This job configures the model with the desired hyperparameters.

Copy
# Create the fine-tuning job with the file IDs
fine_tune_response = openai.fine_tuning.jobs.create(
    training_file=training_file_id,
    validation_file=validation_file_id,
    model="gpt-4o-mini-2024-07-18",
    hyperparameters={
        "n_epochs": 3,
        "batch_size": 64,
        "learning_rate_multiplier": 0.1
    }
)

# Get job ID
print(f"Fine-tuning job created with ID: {fine_tune_response.id}")

3. Monitor the Fine-Tuning Process: After the job is created, you can monitor its progress through the OpenAI API or dashboard. The process may take some time depending on the dataset size and chosen parameters.

Usage

Once fine-tuning is complete, the model can be used to answer specific tax-related questions by leveraging the fine-tuned model endpoint. Here's a basic example of how to interact with the model:

Copy
response = openai.chat.completions.create(
  model="ft:gpt-4o-mini-2024-07-18:personal::9vWCoRNV",
  messages=[
    {
      "role": "user",
      "content": [
        {
          "type": "text",
          "text": "your question here"
        }
      ]
    }
  ],
  temperature=1,
  top_p=1,
  frequency_penalty=0,
  presence_penalty=0,
  response_format={
    "type": "text"
  }
)

print(response)

Contributing

Contributions are welcome!