You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We present WebGLM, a web-enhanced question-answering system based on theGeneral Language Model (GLM). Its goal is to augment a pre-trained largelanguage model (LLM) with web search and retrieval capabilities while beingefficient for real-world deployments. To achieve this, we develop WebGLM withstrategies for the LLM-augmented retriever, bootstrapped generator, and humanpreference-aware scorer. Specifically, we identify and address the limitationsof WebGPT (OpenAI), through which WebGLM is enabled with accuracy, efficiency,and cost-effectiveness advantages. In addition, we propose systematic criteriafor evaluating web-enhanced QA systems. We conduct multi-dimensional humanevaluation and quantitative ablation studies, which suggest the outperformanceof the proposed WebGLM designs over existing systems. WebGLM with the10-billion-parameter GLM (10B) is shown to perform better than thesimilar-sized WebGPT (13B) and even comparably to WebGPT (175B) in humanevaluation. The code, demo, and data are at\url{https://github.com/THUDM/WebGLM}.
AkihikoWatanabe
changed the title
あ
WebGLM: Towards An Efficient Web-Enhanced Question Answering System with
Human Preferences, Xiao Liu+, N/A, arXiv'23
Jun 16, 2023
URL
Affiliations
Abstract
Translation (by gpt-3.5-turbo)
Summary (by gpt-3.5-turbo)
The text was updated successfully, but these errors were encountered: