Scholarlink

The paper aims to test the viability of a small number of open-source Large Language Models for the role of a live learning assistant integrated into an adaptive learning system. Several models were selected for testing which are based on variants of Meta’s Llama and were fine-tuned by Open-Assistant with high-quality human feedback data. The models were tested against a previous-generation LLM (a variant of GPT2) and a current generation LLM (GPT3.5-Turbo). The quality of the generated text was analyzed, as well as the general performance and running cost. The open-source models were found to be acceptable and had the added advantages of increased control (over a commercial solution) and being available for an on-prem installation, with the notable caveat that they had significantly higher running costs.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Viability of Open Source Assistants for Adaptive Learning Systems

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Bobocea Andrei,
  • Corina Marina Mirea

摘要

The paper aims to test the viability of a small number of open-source Large Language Models for the role of a live learning assistant integrated into an adaptive learning system. Several models were selected for testing which are based on variants of Meta’s Llama and were fine-tuned by Open-Assistant with high-quality human feedback data. The models were tested against a previous-generation LLM (a variant of GPT2) and a current generation LLM (GPT3.5-Turbo). The quality of the generated text was analyzed, as well as the general performance and running cost. The open-source models were found to be acceptable and had the added advantages of increased control (over a commercial solution) and being available for an on-prem installation, with the notable caveat that they had significantly higher running costs.