S

Summllama3 70B

Developed by DISLab
SummLlama3-70B is a text summarization model initialized from Llama3-70B-Instruct, optimized through large-scale summarization feedback via DPO training, excelling in fidelity, completeness, and conciseness.
Downloads 15
Release Time : 10/11/2024

Model Overview

An advanced model designed for multi-domain text summarization, supporting both dialogue and non-dialogue texts across seven domains including news and healthcare, generating summaries that better align with human preferences.

Model Features

Multi-domain adaptation
Covers seven domains including news, lifestyle, and healthcare, supporting summarization for both dialogue and non-dialogue texts.
DPO optimization
Direct Preference Optimization based on 100,000+ LLM-generated summary feedbacks, significantly improving summary quality.
Three-dimensional advantages
Outperforms baseline models in fidelity (0.950), completeness (0.632), and conciseness (0.754).

Model Capabilities

Long-text summarization
Dialogue record summarization
Cross-domain summarization
Key information extraction

Use Cases

Content aggregation
News briefing generation
Extracts core events from multiple news articles to generate daily briefings
Automated evaluation achieves 0.95 fidelity, reducing manual editing time by 80%
Meeting minutes
Automatic meeting minutes generation
Condenses multi-party dialogue records into structured action-item summaries
Improves completeness by 6% over Llama3-70B on test sets
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
© 2025AIbase