Skip to content

LangChain Pipeline Development

The ChatPracticus library seamlessly integrates Practicus AI’s hosted private LLM models into the LangChain framework. This allows you to easily interact with language models that are privately hosted and secured within the Practicus AI environment.

To get started with ChatPracticus, you will need the following parameters:

  • endpoint_url: The API endpoint for the LLM model host.
  • api_token: A secret key required to authenticate with the model host API.
  • model_id: The identifier of the target LLM model.

Once these parameters are set, you can define a chat instance using ChatPracticus and invoke it with any prompt of your choice.

import practicuscore as prt

region = prt.get_region()

model_name = None
model_prefix = None
host = None # e.g. 'company.practicus.io'

If you don't know your prefixes and models you can check them out by using the SDK like down below:

my_model_list = region.model_list
display(my_model_list.to_pandas())

my_model_prefixes = region.model_prefix_list
display(my_model_prefixes.to_pandas())

assert model_name, "Please select an LLM model."
assert model_prefix, "Please select the prefix LLM deployed."
assert host, "Please enter your host"

Now we can define our API url and it's token.

api_url = f"https://{host}/{model_prefix}/{model_name}/"
token = prt.models.get_session_token(api_url=api_url)

Below is an example of how to create and use a ChatPracticus instance in your code.

from langchain_core.messages import HumanMessage, SystemMessage
from langchain_practicus import ChatPracticus

def test_langchain_practicus(api_url, token, inputs):
    chat = ChatPracticus(
        endpoint_url=api_url,
        api_token=token,
        model_id="some models will ignore this",
        stream = True
    )

    response = chat.invoke(input=inputs)

    print("\n\nReceived response:\n", response)
    print("\n\nReceived Content:\n", response.content)

Async calls also work using the below (but not on jupyter)

import asyncio
asyncio.run(llm.ainvoke([sys_input, human_input1, human_input2]))
print(response)

After defining your token and API URL, you can easily incorporate prompts into your workflow. By using the langchain library, you can structure your messages as follows:

  • System Messages: Use SystemMessage to provide overarching instructions or context that guide the LLM’s behavior.
  • Human Messages: Wrap user prompts and queries with HumanMessage to represent the user’s input.

This structured approach ensures that the model receives clear, role-specific instructions, enhancing the quality and relevance of its responses.

human_input1 = HumanMessage("Capital of United Kingdom?")
human_input2 = HumanMessage("And things to do there?")
system_message = SystemMessage("Less 50 words.")

inputs = [human_input1, human_input2, system_message]

test_langchain_practicus(api_url, token, ['who is einstein'])

Received json response:

{
    content="Albert Einstein was a theoretical physicist born on March 14, 1879, in Ulm, in the Kingdom of Württemberg in the German Empire. He is best known for developing the theory of relativity, which revolutionized the understanding of space, time, and energy. His most famous equation, E=mc², expresses the equivalence of mass and energy.\n\nEinstein's work laid the foundation for much of modern physics and he received the Nobel Prize in Physics in 1921 for his explanation of the photoelectric effect, which was pivotal in the development of quantum theory. Beyond his scientific contributions, Einstein was also known for his philosophical views, advocacy for civil rights, and his involvement in political and humanitarian causes. He passed away on April 18, 1955, in Princeton, New Jersey, USA." 

    ... with additional metadata 
}

Previous: Building Visual Apps | Next: Streaming