metadata
language: en
license: mit
datasets:
- huuuyeah/meetingbank
tags:
- meeting-summarization
- summarization
library: transformers
pipeline_tag: summarization
library_name: transformers
Meeting Summarizer - Fine-tuned on MeetingBank
This model was fine-tuned on the MeetingBank dataset to generate concise summaries of meeting transcripts.
Performance
| Metric | Validation | Test |
|---|---|---|
| ROUGE-1 | 49.99 | 49.26 |
| ROUGE-2 | 37.42 | 37.07 |
| ROUGE-L | 45.45 | 45.01 |
| ROUGE-Lsum | 46.86 | 46.30 |
Usage
from transformers import pipeline
summarizer = pipeline("summarization", model="ahmeddeldalyyy/meeting-summarizer-meetingbank")
meeting_transcript = """
[Your meeting transcript here]
"""
summary = summarizer(meeting_transcript, max_length=142, min_length=56, num_beams=4, length_penalty=2.0)
print(summary[0]['summary_text'])
Model Details
- Base model: BART
- Fine-tuned on: MeetingBank dataset (city council meetings)
- Parameters:
- max_input_length: 1,024 tokens
- min_output_length: 56 tokens
- max_output_length: 142 tokens
- beam width: 4
- length penalty: 2.0
- learning rate: 2.5e-6
- trained for 2 epochs