http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
Framework for Creating Private LLM and performing Multi document Q&A on Safety Manuals
( Karthikeyan Jagadeesan ),( Prashant Ramappa ) 한국감성과학회 2023 한국감성과학회 국제학술대회(ICES) Vol.2023 No.-
Open-source large language models, including llama2, Zephyr, T5, and BERT, have emerged as transformative tools with vast potential across diverse industries. These models empower organizations to leverage the capabilities of natural language understanding and generation in unprecedented ways. As large language models (LLMs) continue to revolutionize sectors such as healthcare, education, and beyond, they are being deployed to automate various tasks and enhance customer service. The principal hurdles encountered by enterprises in their utilization of open-source Large Language Models (LLMs) pertain to the domains of Data Privacy and Security. LLMs necessitate the acquisition of extensive datasets for proficient training, frequently encompassing data of a sensitive and confidential nature. Safeguarding this data and ensuring adherence to data privacy regulations is paramount. Moreover, the intricacy lies in the substantial computational resources essential for the training and deployment of LLMs, which engender significant financial costs and mandate specialized infrastructure and expertise. This paper introduces a framework for hosting private Large Language Models (LLMs) by fine-tuning opensource models through the application of the PEFT quantization method within the Databricks platform. Furthermore, we have developed a question-answering system for safety manuals capable of providing answers to inquiries, even when the supporting evidence is dispersed across multiple documents, some of which may be quite extensive.