Chaining
-
When we interact with llm basic usage is
- prompt
- llm
-
we can chain these things
prompt | llm
- I’m interacting with llm but i want response in json structure
prompt | llm | jsonstrcture
- The above notations are referred as LCEL
PROMPT
- PROMPT is an input to the model
-
The underlying LLM deals with different types of prompts
- SYSTEM PROMPT => define the role
- USER PROMPT => this is where we ask a question
-
When we pass this prompt we get a response
MESSAGES
- When we interact with llm we will have different types of prompts or responses
- System Prompt
- User Prompt
- Response from llm
- To generalize this langchain does this. we have 4 types of messages
- SystemMessage => (System Prompt)
- HumanMessage => User Prompt
- AIMessage => llm response
- ToolMessage
Experiment
- Create a new folder
hello_llms
mkdir hello_llms
cd hello_llms
uv init .
- Lets add a langchain package
#uv pip install langchain
uv add langchain
- We need to interact with some llm
- lets create an api key Refer Here
- Install one more package
uv add python-dotenv
- Create a new file called as .env with
GEMINI_API_KEY='paste your api key here'
-
Now since we are using gemini Refer Here
-
We need to install one more package
uv add langchain-google-genai
- Now lets write simple code to interact with model
-
Refer Here for the changes
-
Now lets try chaining
