pegasus-samsum

Aaditya-Nanda/pegasus-samsum

This model is a fine-tuned dialogue summarization model based on google/pegasus-xsum.

Model Details

• Base model: google/pegasus-xsum
• Task: dialogue summarization
• Dataset used for training: knkarthick/dialogsum
• Framework: Hugging Face Transformers

Intended Use

This model is intended to generate short abstractive summaries from multi-turn conversations.

Training Notes

• Training epochs: 3
• Train subset size: 2000
• Validation subset size: 200
• Max input length: 512
• Max target length: 128

Limitations

• This model was trained as a lightweight project run and may miss details in longer or more complex conversations.
• The repository name still uses samsum for compatibility with the deployed app, but the fine-tuning dataset is DialogSum.

Example

Input:

Riya: Can you send me the draft by 6 PM?
Karan: Yes, I will finish the conclusion and add the March sales numbers first.
Riya: Great, I will review it and prepare the slides.

Expected summary:

Karan will send the updated draft by 6 PM, and Riya will review it before preparing the slides.