- Hw1 -- Intent Classification & Slot Tagging
- RNN-based model
- Score: 20/20
- Hw2 -- BERT Multiple Choice & Question Answering
- Chinese-macbert
- Score: 21/21
- Hw3 -- mT5 Chinese News Summarization
- mT5-small
- Tw-rouge
- Score: 20/22
- CrossFit
- BART-based, T5-base, mT5 (Hugging Face)
- MAML, Multitask Learning (PLM Fine-tuning Baseline)
- Prefix Prompt (Google Prefix Prompt)
- Adapter (Houlsby)