This repository contains the homework assignments and final project for the Applied Deep Learning (ADL) course at National Taiwan University (NTU) in the Fall of 2023.
The repository is organized into four main directories:
- hw1/
- hw2/
- hw3/
- Final_Project/
1. hw1
Task: Paragraph and Span Selection
This assignment focuses on implementing two distinct models:
- Paragraph Selection Model: Given a Chinese question and four Chinese paragraphs, the model identifies the paragraph most relevant to the question.
- Span Selection Model: After selecting the relevant paragraph, this model extracts the precise answer span from the chosen paragraph.
Key Concepts: Text classification, Span extraction, Chinese Natural Language Processing (NLP).
2. hw2
Task: Fine-tuning mT5 for Title Generation
In this assignment, we fine-tune a pre-trained Multilingual Text-to-Text Transfer Transformer (mT5) to generate appropriate titles for given paragraphs. The model performance is evaluated using the Rouge score, a standard metric for sequence generation tasks.
Key Concepts: mT5, Sequence generation, Rouge score, NLP in multiple languages.
3. hw3
Task: Instruction-tuning Taiwan LlaMa Model
This assignment involves instruction-tuning a Taiwan LlaMa Model for a language translation task. The model is trained to either:
- Translate plain Chinese into classical Chinese, or
- Translate classical Chinese into plain Chinese.
Performance is evaluated using the Perplexity metric.
Key Concepts: Instruction-tuning, Language translation, Classical Chinese NLP.