Memory is a crucial element in building chatbots that can maintain a coherent conversation with users. In this video, we will explore different memory types that can be used with ChatGPT, and how they can be implemented using LangChain memory. We will discuss how different memory types work, how they can be used in a chain, and what kind of information they can return.

We will start with the simplest memory type, "buffer" memory, which keeps a buffer of all prior messages. Then we will dive into ConversationSummaryBufferMemory, which combines a buffer of recent interactions with a summary of past interactions. We will also explore ConversationBufferWindowMemory, which keeps a sliding window of the most recent interactions, and VectorStoreRetrieverMemory, which stores memories in a vector database (e.g. ChromaDB) and queries the top-K most "salient" docs every time it is called.

Prompt Engineering Tutorial: [ Ссылка ]
Prompt Engineering GitHub Repository: [ Ссылка ]

Discord: [ Ссылка ]
Prepare for the Machine Learning interview: [ Ссылка ]
Subscribe: [ Ссылка ]

00:00 - Intro
00:36 - Google Colab Setup
01:14 - What is Memory?
03:45 - Conversation Chain
06:18 - Change the Prompt
09:41 - Save/load Messages
11:18 - Window Memory
12:58 - Summary Memory
15:30 - Vectorstore Memory
19:20 - Simple Chatbot with Memory
21:57 - Conclusion

#chatgpt #langchain #gpt4 #promptengineering #python #artificialintelligence

свернуть