Fine-Tuning a model to build an AI assistant

AI Juris is an artificial intelligence-based legal assistant application. It uses fine-tuned open-source large language models (LLMs) to provide accurate and efficient legal support. This project aims to democratize access to legal tools and facilitate consultation processes in various areas of law by leveraging modern AI technologies.

Project Structure

• Base Model: The project utilizes the pre-trained flan-t5-base model, a sequence-to-sequence transformation model developed by Google.
• Data Source: Data was collected from the StackExchange website using web scraping techniques. The collected data is structured in a question-and-answer format, which is ideal for model training.
• Training Refinement: The Seq2SeqTrainingArguments package is used to configure training parameters. Training is conducted with Seq2SeqTrainer to fine-tune the model with custom data.
• Evaluation: The Rouge metric, one of the most common metrics for text generation tasks, is used to assess model performance.