๐ก Purpose of project:
With the vast amount of data available on the internet, it has become increasingly challenging for users to find relevant information quickly and efficiently. Traditional search engines rely on keywords and algorithms to rank search results, which can be limiting and often fail to provide accurate and relevant information. To address this issue, weโd like to develop a query engine that leverages the power of large language models to provide more accurate and efficient search results.
๐ Objectives:
The primary objective of this project is to develop a query engine that utilizes large language models to understand the intent behind a user's search query and provide more accurate and relevant search results (including references). The following are the specific objectives of this project.
- LLMs should be run locally. (The maximum usage of VRAM should be less than 80 GB)
- Read/handle various file formats (ppt, excel, word, pdf, and text).
- Need to extract the exact contents or units of a table contained in the document.
- Developed model should be able to handle English (or both Korean and English) doucumnets.
- Need to return a reference list of its contents after searching.
๐ ์ฐธ๊ณ ๋ด์ฉ
- ํด๋น ๊ณผ์ ํ๊ฐ๋ ๋ ผ๋ฌธ ํ์์ pdf, word, ๊ทธ๋ฆฌ๊ณ ์ฃผ์ด์ง ์นํ์ด์ง์์ ์ง๋ฌธ์ ๋ํ ์ฌ๋ฐ๋ฅธ ์ ๋ณด๋ฅผ ์ ๊ณตํ๋ ์ง๋ฅผ ํ๋จํฉ๋๋ค.
- (Download Llama 2 model) [https://ai.meta.com/llama/] - 13B or 70B model is preferred
- (Open platform for LLM based chatbot) [https://github.com/lm-sys/FastChat]
- (Interface between LLMs and data) [https://github.com/run-llama/llama_index]
์๋๋ ๋ํํ ์ธํฐํ์ด์ค์ ์์์ ๋๋ค. ์ฌ์ฉ์๊ฐ ์ง๋ฌธ์ ์ ๋ ฅํ๋ฉด, target data ํด๋์์ ํด๋น ๋ด์ฉ / ์์น๋ฅผ ๋ํํ์ผ๋ก ๋ฐํํฉ๋๋ค.