The AML3304 project involves creating a Python-based generative AI language model.
Students can work in teams of up to four and have options like using Google Collab Notebook, HuggingFace Spaces Lab Space, or local premises with Visual Studio Code for project development.
The project's deliverable is a Python generative text AI model, trained on a topic of interest, with a focus on using transformers and embeddings.
The project emphasizes practical skills in machine learning, particularly in creating and applying embeddings using Hugging Face's Spaces API and the transformers library.
For a grading rubric and solution outline, consider the following:
Code Quality and Organization (20%): Evaluate the readability, organization, and documentation of the code. Check for proper use of functions, classes, and modules.
Implementation of AI Model (30%): Assess the effective use of the transformers library, correct implementation of the chosen model, and how well the embeddings and model training are executed.
Creativity and Originality (20%): Consider the uniqueness of the chosen topic and the innovative use of AI in the project.
Functionality and Performance (20%): Test the AI model for its performance in generating coherent and contextually appropriate responses. Check for accuracy, response time, and error handling.
Presentation and Documentation (10%): Evaluate the clarity and completeness of the project documentation, including explanations of the code, model choice, and decision-making process. The project's presentation, whether in a report or a video, should be clear and professional.
For the solution, start with a foundational language model from Hugging Face, like GPT or BERT.
Customize the model for the specific use case by fine-tuning it on a relevant dataset. Implement and test the model in a Python environment, ensuring it can generate coherent text based on the input provided. Document the process, detailing the model choice, training process, challenges faced, and solutions implemented.
Want to print your doc? This is not the way.
Try clicking the ⋯ next to your doc name or using a keyboard shortcut (
