AI/ML, Cloud Computing, Data Analytics

5 Mins Read

Language Model Development and Optimization with LangSmith

Voiced by Amazon Polly


LangSmith is a unified platform that helps create practical advanced language models (LLMs). It allows you to find and fix issues test, and assess the performance of chains and intelligent agents made using any LLM framework. It works smoothly with LangChain, a popular open-source framework for building LLM projects. LangSmith is created by LangChain, the company responsible for the open-source LangChain framework.

It allows developers to work with language models more easily. It provides tools for debugging, testing, and monitoring language model applications. Think of it like a control center where developers can see how different components of their applications work together, test different prompts and language models, and ensure their applications are running smoothly. It’s like having a toolkit that simplifies the process of building and improving applications that use language models. With LangSmith, developers can create high-quality language model applications with less effort and more confidence.

How LangSmith is different from LangChain?

LangSmith and LangChain are related but serve different purposes. LangChain is a framework for developing applications powered by language models. It provides a set of modular components and abstractions for working with language models, such as models, prompts, chains, agents, memory, and callbacks. LangChain allows developers to build and customize their language model applications. LangSmith, on the other hand, is a platform built on top of LangChain. It provides tools for debugging, testing, and monitoring language model applications. LangSmith helps developers with tasks like visualizing components, evaluating prompts and language models, capturing usage traces, and generating insights. It simplifies working with language models and improves the development experience. In summary, LangChain is the framework for building language model applications, while LangSmith is the platform that enhances the development process by providing additional tools and capabilities.

Pioneers in Cloud Consulting & Migration Services

  • Reduced infrastructural costs
  • Accelerated application deployment
Get Started

Where can LangSmith fit in LLM architecture?

LangSmith can fit into the Language Model (LLM) architecture in several ways:

  1. Fine-tuning LLMs: LangSmith provides resources and examples for fine-tuning LLMs on real usage data. This allows users to train their models to perform better on specific tasks or domains.
  2. Evaluation and auditing: LangSmith offers tools for evaluating and auditing the performance of LLM workflows. Users can assess the correctness and effectiveness of their models, ensuring they meet the desired quality standards.
  3. Debugging and testing: LangSmith is a developer platform enabling users to debug, test, and monitor chains built on any LLM framework. It helps identify and resolve issues in the LLM architecture during the development process.
  4. Integration with LangChain: LangSmith seamlessly integrates with LangChain, a framework for developing applications powered by language models. It provides a platform for developers to inspect, test, evaluate, and monitor chains built on any LLM framework.


We may find LangSmith handy when we want to:

  • Find and fix issues in a new chain or agent.
  • Visualize how components like chains, LLMs, retrievers, etc. connect and are being used.
  • Test different prompts and LLMs for a single component.
  • Execute a particular sequence multiple times on a data set to ensure it consistently meets quality standards.
  • Record usage traces and use LLMs to generate insights.

Key Features of LangSmith

Let’s look at each component of LangSmith individually:

  1. Tracing:
  • Log runs: Building and working with LLM applications can become complex, particularly when dealing with agents or chains that involve multiple layers of LLMs and other components. LangSmith simplifies this by allowing you to log runs of your LLM applications, enabling you to examine the inputs and outputs of each component in the chain. This feature is especially helpful for debugging your application and gaining insights into specific component behavior.
  • Organize the work: Your runs are stored in projects. The primary runs, called traces, are stored in the default project if you don’t specify one. You can also see all your runs without any nesting. You can make as many projects as you need to keep things organized. For example, you might have a project for each of your LLM application environments or create projects to separate runs on different days. It’s also handy for specific experiments or debugging sessions.
  • Visualize the runs: Whenever you run a LangChain component with tracing turned on or use the LangSmith SDK to save run trees directly, the app stores the call hierarchy for that run. You can then see this hierarchy visually in the app. This allows you to dive into details like the inputs and outputs of each component, the parameters used, how long it took, feedback received, token usage, and other crucial details for inspecting your run. Additionally, you can rate the run, which helps gather data for training, testing, and other analysis.
  • Sharing the work: You can easily share any of the runs you’ve recorded in LangSmith. This simplifies the process of publishing and replicating your work. For example, if you come across a bug or unexpected result in a specific setup, sharing it with your team or in a LangChain Issue makes it more convenient to address and resolve.

2. Datasets: Datasets are examples used to assess or enhance a chain, agent, or model. Each example in a dataset represents an interaction and includes inputs and optional expected outputs. Currently, there are three types of datasets, each indicating common input and output structures. There are three types of datasets:

  • Key-value datasets: Default datasets use the “kv” data type, where inputs and outputs are like pairs of information with keys and values. This setup is handy when dealing with chains and agents that need or produce multiple inputs or outputs.
  • LLM datasets: Datasets labeled as “llm” type match with the string inputs and outputs typical of “completion” style LLMs (string in, string out). In these datasets, there’s an “inputs” dictionary with a key “input” connected to a single prompt string. Likewise, the “outputs” dictionary holds a single “output” key tied to a single response string.
  • Chat datasets: Datasets labeled as “chat” type align with messages and outputs from LLMs that operate with structured “chat” messages. In each example, the “inputs” dictionary holds a key “input” linked to a list of chat messages. The “outputs” dictionary has a key “output” tied to a single list of chat messages.

3. Evaluation:

  • LangChain Evaluators: LangChain evaluators are components or modules within the LangChain framework designed to assess and evaluate the performance of language models. These evaluators are specifically tailored to different types of evaluation tasks and criteria. Some examples of LangChain evaluators include question answering evaluators, string evaluators, criteria evaluators, and trajectory evaluators.
  • Custom Evaluators: Custom evaluators refer to evaluators that users or developers create to meet their specific evaluation needs. LangSmith provides guidance and resources for creating custom evaluators outside the LangChain framework. These custom evaluators can be designed to evaluate language models based on unique criteria, tasks, or domains. They offer flexibility and customization options for evaluating models beyond the predefined evaluators provided by LangChain.

4. Human Annotation: Human annotation in LangSmith refers to adding human-generated labels or annotations to text data. It involves tasks like categorizing, tagging, or rating examples to create labeled datasets for training or evaluation purposes. By incorporating human annotation, you can improve the accuracy and quality of language models. The labeled datasets created through human annotation serve as valuable training material for the models, enabling them to learn from human expertise and make better predictions or classifications.

5. Hub: In LangSmith, the ‘Hub’ is like a central place or repository where you can discover, share, and access different resources related to language models. It’s a platform where users can find pre-defined setups, configurations, or examples that can be used to build and enhance their language models. The Hub also allows users to collaborate and exchange ideas. You can join discussions, provide feedback, and contribute to the community. It’s a place where developers and researchers come together to learn from each other and improve their language models.


LangSmith is a platform that offers tools, resources, and guidance for working with language models and analyzing language-related data. It provides features such as fine-tuning language models, evaluating their performance, and incorporating human annotation to improve accuracy.

LangSmith also offers logging and tracing functionalities for debugging and monitoring language model applications. The Hub in LangSmith is a central repository for sharing and accessing language model configurations and examples.

Overall, LangSmith aims to support developers, researchers, and organizations in harnessing the power of language models for various applications.

Drop a query if you have any questions regarding LangSmith and we will get back to you quickly.

Making IT Networks Enterprise-ready – Cloud Management Services

  • Accelerated cloud migration
  • End-to-end view of the cloud environment
Get Started

About CloudThat

CloudThat is a leading provider of Cloud Training and Consulting services with a global presence in India, the USA, Asia, Europe, and Africa. Specializing in AWS, Microsoft Azure, GCP, VMware, Databricks, and more, the company serves mid-market and enterprise clients, offering comprehensive expertise in Cloud Migration, Data Platforms, DevOps, IoT, AI/ML, and more.

CloudThat is recognized as a top-tier partner with AWS and Microsoft, including the prestigious ‘Think Big’ partner award from AWS and the Microsoft Superstars FY 2023 award in Asia & India. Having trained 650k+ professionals in 500+ cloud certifications and completed 300+ consulting projects globally, CloudThat is an official AWS Advanced Consulting Partner, AWS Training Partner, AWS Migration Partner, AWS Data and Analytics Partner, AWS DevOps Competency Partner, Amazon QuickSight Service Delivery Partner, Amazon EKS Service Delivery Partner, Microsoft Gold Partner, AWS Microsoft Workload Partners, Amazon EC2 Service Delivery Partner, and many more.

To get started, go through our Consultancy page and Managed Services PackageCloudThat’s offerings.


1. What is the difference between ‘Run’ and ‘Trace’?

ANS: – A ‘Run’ represents a single unit of work or operation within your LLM application, whereas a ‘Trace’ is a collection of runs organized in a tree or graph.

2. How do I share a prompt with my teammates in LangSmith?

ANS: – You can share prompts within a LangSmith organization by uploading them within a shared organization.

WRITTEN BY Yaswanth Tippa

Yaswanth Tippa is working as a Research Associate - Data and AIoT at CloudThat. He is a highly passionate and self-motivated individual with experience in data engineering and cloud computing with substantial expertise in building solutions for complex business problems involving large-scale data warehousing and reporting.



    Click to Comment

Get The Most Out Of Us

Our support doesn't end here. We have monthly newsletters, study guides, practice questions, and more to assist you in upgrading your cloud career. Subscribe to get them all!