AI記憶體機制與技術文章生成策略

from typing import List, Dict, Any
from langchain.memory import ConversationSummaryBufferMemory
from langchain_core.messages import SystemMessage

class OnlyStoreAIMemory(ConversationSummaryBufferMemory):
    """
    一個自定義的記憶體類別，可以根據需求覆寫 save_context 方法，
    以實現特定的記憶儲存邏輯。
    """
    def save_context(self, inputs: Dict[str, Any], 
                     messages: List[SystemMessage]) -> None:
        # 在此處可以實作自定義的記憶儲存邏輯，
        # 例如，只儲存 AI 的回應，或過濾掉特定內容。
        # 此處的 pass 表示它目前繼承父類別的行為。
        super().save_context(inputs, messages)
        pass

def generate_blog_post(self) -> List[str]:
    """
    生成部落格文章。
    
    :return: 部落格文章段落列表。
    """
    blog_post = []
    print("Generating the blog post...\n---")
    
    for subheading in self.outline.sub_headings:
        k = 5  # 初始化k值
        
        while k >= 0:
            try:
                # 查詢相關檔案
                relevant_documents = (self.chroma_db.as_retriever() 
                                   .invoke(subheading.title, k=k))
                
                # 生成文章段落
                section_prompt = f"""
               ...prompt_excluded_for_brevity...
                
                Section text:
                """
                result = self.blog_post_chain.predict(section_prompt)
                
                # 新增生成的段落到部落格文章中
                blog_post.append(result)
                
                break
            except Exception as e:
                # 處理異常情況
                pass
    
    return blog_post

def generate_blog_post(subheadings, questions_and_answers):
    blog_post = ""
    for subheading in subheadings:
        relevant_documents = fetch_relevant_documents(subheading)
        attempts = 5
        while attempts > 0 and not relevant_documents:
            attempts -= 1
            try:
                relevant_documents = fetch_relevant_documents(subheading)
            except Exception as e:
                print(f"An error occurred: {e}")
        
        if not relevant_documents:
            print('''All attempts to fetch relevant documents have 
            failed. Using an empty string for relevant_documents.
            
            ''')
            relevant_documents = ""
        
        section_prompt = f"""You are currently writing the section: {subheading.title}

---
Here are the relevant documents for this section: {relevant_documents}.
If the relevant documents are not useful, you can ignore them.
You must never copy the relevant documents as this is plagiarism.

---

Here are the relevant insights that we gathered from our interview questions 
and answers: {questions_and_answers}.

You must include these insights where possible as they are important and will 
help our content rank better.

---
You must follow the following principles:

- You must write the section: {subheading.title}
- Render the output in.md format
- Include relevant formats such as bullet points, numbered lists, etc.

"""
        section_content = generate_section_content(section_prompt)
        blog_post += section_content
    
    print("Finished generating the blog post!\n---")
    return blog_post