Arxiv URL: https://arxiv.org/abs/2305.05176
Authors: Lingjiao Chen, Matei Zaharia, James Zou
The paper discusses the cost associated with querying large language models (LLMs) and proposes FrugalGPT, a framework that uses LLM APIs to process natural language queries within a budget constraint. The framework uses prompt adaptation, LLM approximation, and LLM cascade to reduce the inference cost while improving accuracy.
Key Insights & Learnings:
- Using LLMs on large collections of queries and text can be expensive.
- Prompt adaptation, LLM approximation, and LLM cascade are three strategies that users can exploit to reduce the inference cost associated with using LLMs.
- FrugalGPT can match the performance of the best individual LLM with up to 98% cost reduction or improve the accuracy over the best individual LLM by 4% with the same cost.
- Prompt engineering and model ensemble are related works that have emerged to enhance LLMs' performance and reduce costs.
- FrugalGPT is a flexible framework that uses LLM APIs to process natural language queries within a budget constraint and can reduce the inference cost by 98% while exceeding the performance of the best individual LLM.
Terms Mentioned: large language models, LLMs, FrugalGPT, prompt adaptation, LLM approximation, LLM cascade, natural language queries, inference cost, prompt engineering, model ensemble
Technologies / Libraries Mentioned: GPT-4, ChatGPT, J1-Jumbo, OpenAI, AI21, CoHere, Textsynth