For this project, you will select a dataset of your choice to explore an LLM or AI project.
- Your project should involve applying a pretrained model for text data
- Your task must be one or more of the following:
- translation
- summarization
- sentiment analysis
- entity recognition
- Your text must be labeled or annotated data
- it is recommended to use a dataset that was previously used for benchmarking or research
- Explore Hugging Face datasets, Kaggle, etc.
- You must evaluate the performance of your model with appropriate metrics for your task
- (examples may include: ROUGE or BLEU )
- Use Hugging Face tooling
- Answer the following questions in a brief writeup:
- Why is model appropriate for your task and dataset?
- What are the limitations and biases of your model?
- How did the model perform on your task?
- How would you improve the model in the future?
Steps:
- Identify your task and dataset
- Select an appropriate model
- Apply model to your data and evaluate
- Iterate as needed
- Write up final conclusions
- Notebook with working code and markdown answering the questions above
- Brief Presentation to classmates
- Consider your audiance
- Be sure to highlight importand decisions and findings
- Make recommendations for next steps
This project involves using pretrained models which can be computationally intensive. Important Note for Students Without GPU Access:
If you don't have access to a GPU, consider the following strategies:
- Use smaller model variants (e.g., distilled or tiny versions) that require less computational power
- Process smaller batches of data
- Monitor memory usage carefully to avoid crashes
- Reduce the number of inference examples if necessary
- Utilize Hugging Face's model quantization techniques to reduce memory requirements
- Consider free cloud options like Google Colab (free tier includes limited GPU access)