eu9ene/llm_sum

Exploration of summarization with Large Language Models as a part of Stanford cs224u course project

Summarization with LLMs

Exploration of styled summarization with Large Language Models as a part of Stanford cs224u course project.

Datasets:

• CNN/DailyMail
• Newsroom
• Xsum

Samples used for evaluation

Explored models:

• Eleuther 1.3B
• OpenAI Davinci
• OpenAI Curie
• BRIO

Generated summaries

Metrics:

• ROUGE1/ROUGE2/ROUGEL/ROUGELsum
• BERT Score
• Readability
• Coherence
• Sentiment (positve rate)

Evaluation results

Associated paper: Summarizing in style: Exploring summarization with Large Language Models