L

Lsg Bart Base 4096 Multinews

Developed by ccdv
A BART-base model based on LSG technology, designed for long-text summarization tasks, supporting input sequences of up to 4096 tokens
Downloads 26
Release Time : 5/25/2022

Model Overview

This model employs a local-sparse-global attention mechanism to process long sequences and is fine-tuned on the multi_news dataset, making it suitable for multi-document summarization tasks

Model Features

Long Sequence Processing Capability
Supports input sequences of up to 4096 tokens, utilizing a local-sparse-global attention mechanism for efficient long-text processing
Multi-Document Summarization Optimization
Fine-tuned on the multi_news dataset, specifically optimized for multi-document summarization scenarios
Flexible Attention Configuration
Supports various sparse attention modes (pooling/strided/block-strided/normalized/LSH), allowing for a balance between performance and resource consumption as needed

Model Capabilities

Long-text summarization generation
Multi-document information integration
English text processing

Use Cases

News Summarization
Multi-Source News Summarization
Generates a unified summary from multiple related news articles
Achieves R1 47.10/R2 18.94/RL 25.22 on the multi_news test set
Document Organization
Long Document Summarization
Generates concise summaries for long texts such as technical documents and research reports
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
Š 2025AIbase