8000 GitHub - UNITES-Lab/AgentSymbiotic
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

UNITES-Lab/AgentSymbiotic

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 

Repository files navigation

Paper License: MIT

Symbolic Mixture-of-Experts Architecture

This repository contains the implementation of AgentSymbiotic, an advanced web agent framework. It introduces a symbiotic learning framework that enables large language models (LLMs) and small language models (SLLMs) to enhance each other iteratively, achieving superior performance in web-based automation tasks. Our code is built upon AgentOccam.

🚀 Key Features

  • Mutual Improvement Between LLMs and SLLMs
    • Large LLMs generate high-quality trajectory data.
    • Distilled small LLMs explore diverse strategies, enriching the dataset.
  • Enhanced Knowledge Distillation
    • Speculative Data Synthesis: Mitigates off-policy bias and improves SLLM adaptation.
    • Multi-Task Learning: Enhances reasoning capabilities of small LLMs.
  • Hybrid Mode for Privacy Preservation
    • Automatically switches to local SLLMs when handling sensitive data.
  • State-of-the-Art Performance
    • Achieved 52% success rate with large LLMs and 49% with distilled small LLMs on the WEBARENA benchmark.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

0