LangChain
a framework for developing applications powered by language models
LangChain Parts (LL,Prom,Mem,Chai,Vec S,Doc)
LLMs, Prompts, Memory, Chains, Vector Stores, & Document Loaders
Chat Models
are specifically tuned for having conversations; Takes a list of chat messages as input and returns an output
LangChain String Prompt Template
created from a formatted python string and can have any # of variables; Typically used for generation models as its suited for one block of text
LangChain Chat Prompt Template
a list of messages; associated w/ content and additional parameters (role); Suited for chat models due to its multiple block nature
LangChain Chains
creates chains of components including LLMs and workflow to move through the “app”
LangChain Expression Language (LCEL)
a language for creating chains in LangChain
LangChain Memory
ability to store information about past interactions; It reads from memory after user input but before chain execution, its writes to memory after core logic but before output
RAG Pipeline Ingestion
we take in documents,where it gets split into chunks, embedding, and then put into a database
RAG Pipeline Retrieval
user inputs a query, gets through semantic searching, and the top k results are given back
RAG Pipeline Generation
top k results are given to the LLM for generation
RAG Ingestion Chunk Overlap
the number of overlapping characters between adjacent chunks
Oracle Vector Storage
a SQL data storage but one of the columns stores a vector object for each document (row) inside of a table
Vector Indexes
specialized data structures designed for similarity searches; Pretty much grouping
