GenAI: Observability Starter for Langchain LLMChain-Based Solutions

Generative AI Horizontal LangChain

This accelerator shows how users can quickly and seamlessly enable LLMOPs or Observability in their existing LLMChain-based generative AI solutions without the need of code refactoring.

Request a Demo

This accelerator shows how you can easily enable observability in your LLMChain-based AI Solutions with the DataRobot LLMOps feature tools.

It outlines an example of a byte-sized solution in its current state and then uses DataRobot tools to enable observability almost instantly for the solution.

DataRobot provides tools to enable the observability of external generative models. All the hallmarks of DataRobot MLOps are now available for LLMOps.


Install the prerequisite libraries

This notebook uses Langchain’s LLMChain to illustrate a simple generative AI solution. Langchain is a framework for developing applications powered by language models easily. It simplifies application development by providing a simple and comprehensive API for a wide variety of Large Language Modeling inference tasks. Langchain’s Chain is an end-to-end wrapper around multiple individual components.


!pip install datarobotx[llm] datarobot-mlops datarobot-mlops-connected-client langchain transformers nltk py-readability-metrics

Current state

The following cells outline the current state of a simple function utilizing the Azure-hosted GPT LLM. This accelerator uses the LLMChain object from the Langchain library to build a text generation example. An LLMChain is the type of chain intended for use in LLM inference. It takes multiple input variables, uses the PromptTemplate to format them into a prompt. It then passes that to the model. Finally, it uses the OutputParser (if provided) to parse the output of the LLM into a final format.

In[ ]:

import os
import warnings

import datarobot as dr
from langchain import PromptTemplate
from langchain.chains import LLMChain
from langchain.chat_models import AzureChatOpenAI


prompt = PromptTemplate(
    template="Explain the {concept} provided in the conversation to me like I am 5 year old.",

chain = LLMChain(
Out[ ]:

'Agile is like playing with building blocks. Instead of building one big tower all at once, we build smaller towers step by step. We can change our plans as we go and make improvements along the way. It helps us work together and finish our tower faster.'

Observability with DataRobot

To enable observability on the above LLMChain object, you first need to create a deployment in DataRobot. This can be done from the GUI or the API based on your preference.
If the chain object is used as is in your application for inference, DataRobot’s extended library DRX provides users the functionality to monitor the LLMChain object without the need for code rewrites such as wrapping the chain inside a new function.

Connect to DataRobot

In [ ]:

# Initialize the DataRobot Client if you are not running this code on the DataRobot platform.
# import datarobot as dr
# dr.Client(endpoint=ENDPOINT,token=TOKEN)

from utilities import create_external_llm_deployment

deployment_id, model_id = create_external_llm_deployment("Deployment Guarded Chain")
Out[ ]:

[nltk_data] Downloading package punkt to /home/notebooks/nltk_data...
[nltk_data]   Unzipping tokenizers/

Downloading (…)lve/main/config.json:   0%|       | 0.00/811 [00:00<?, ?B/s]

Downloading pytorch_model.bin:   0%|       | 0.00/438M [00:00<?, ?B/s]

Downloading (…)okenizer_config.json:   0%|       | 0.00/174 [00:00<?, ?B/s]

Downloading (…)solve/main/vocab.txt:   0%|       | 0.00/232k [00:00<?, ?B/s]

Downloading (…)cial_tokens_map.json:   0%|       | 0.00/112 [00:00<?, ?B/s]


Initiate monitoring configuration

The cells below declare and initialize monitoring configuration. The MonitoringConfig object lets DataRobot capture the input and output of the LLMChain object and record the values on the above deployments. The GuardChain object expects the LLMChain to be monitored and the MonitoringConfig objects as inputs.

With just a few lines of code shown below, you can enable observability in your LLMChain-based solutions.

In[ ]:

import json

from datarobotx.llm.chains.guard import GuardChain, MonitoringConfig

monitor_config1 = MonitoringConfig(
    deployment_id=deployment_id, model_id=model_id, target="text"

chain = GuardChain(inner_chain=chain, monitor=monitor_config1)
Out[ ]:

GuardChain(memory=None, callbacks=None, callback_manager=None, verbose=False, tags=None, metadata=None, inner_chain=LLMChain(memory=None, callbacks=None, callback_manager=None, verbose=False, tags=None, metadata=None, prompt=PromptTemplate(input_variables=['concept'], output_parser=None, partial_variables={}, template='Explain the {concept} provided in the conversation to me like I am 5 year old.', template_format='f-string', validate_template=True), llm=AzureChatOpenAI(cache=None, verbose=True, callbacks=None, callback_manager=None, tags=None, metadata=None, client=<class 'openai.api_resources.chat_completion.ChatCompletion'>, model_name='gpt-35-turbo-16k', temperature=0.0, model_kwargs={}, openai_api_key='5d88cbce4ba84373816840e3abc1d898', openai_api_base='', openai_organization='datarobot-openai-prototyping', openai_proxy='', request_timeout=None, max_retries=6, streaming=False, n=1, max_tokens=None, tiktoken_model_name=None, deployment_name='gpt-35-turbo-16k', model_version='', openai_api_type='azure', openai_api_version='2023-03-15-preview'), output_key='text', output_parser=StrOutputParser(), return_final_only=True, llm_kwargs={}), monitor=MonitoringConfig(deployment_id='6512ef9195d7566749fddd81', model_id='6512ef917b60d72d7df59985', inputs_parser=<function GuardChain._apply_langchain_parsers.<locals>.<lambda> at 0x7fda94736a60>, output_parser=<function GuardChain._apply_langchain_parsers.<locals>.<lambda> at 0x7fda94736af0>, target='text'), guardrails=[])
In[ ]:"agile")
Out[ ]:

'Agile is like playing with building blocks. Instead of building one big tower all at once, we build smaller towers step by step. We can change our plans as we go and make improvements along the way. It helps us work together and finish our project faster.'


Using the DataRobot tools for LLMOps, you can implement observability for your LLMChain-based applications easily without the need for significant code rewrites.

Get Started with Free Trial

Experience new features and capabilities previously only available in our full AI Platform product.

Get Started with This Observability Workflow

Explore more LangChain AI Accelerators

Explore more AI Accelerators