AI說書 - 從0開始 - 221 | GPT 4 & RAG 實作

我想要一天分享一點「LLM從底層堆疊的技術」，並且每篇文章長度控制在三分鐘以內，讓大家不會壓力太大，但是又能夠每天成長一點。

回顧目前手上有的素材：

• 依賴的安裝：AI說書 - 從0開始 - 218 | OpenAI GPT 4 & RAG
• 關鍵字偵測：AI說書 - 從0開始 - 219 | GPT 4 & RAG 關鍵字偵測函數
• 文章檢索器：AI說書 - 從0開始 - 220 | GPT 4 & RAG 文章檢索器

現在可以來實作 RAG 函數了：

from openai import OpenAI
import ipywidgets as widgets 
from IPython.display import display

client = OpenAI()
AImodel = "gpt-4o"

def openai_chat(input_text, document_excerpt, web_article_summary):
	response = client.chat.completions.create(model = AImodel,
																								 messages = [{"role": "system",
																								 								"content": f"The following is an excerpt from a document about climate change: {document_excerpt}"}, 
																								 							 {"role": "system",
																								 							  "content": f"The following is a summary of a web article on renewable energy: {web_article_summary}"},
																								 							 {"role": "user",
																								 								"content": input_text}],
																								 temperature = 0.1,
																								 max_tokens = 150,
																								 top_p = 0.9,
																								 frequency_penalty = 0.5, # Reduces repetition of the same text. Higher values discourage repetition
																								 presence_penalty = 0.5) # Reduces repetition of similar topics. Higher values encourage new topics
	return response.choices[0].message.content