LangChain v0.2 Release: Versioned Docs, New Evaluator Prompts, and More
Joerg Hiller Jun 02, 2024 01:11
LangChain v0.2 introduces versioned docs, improved dataset handling, and new evaluator prompts.
LangChain, a prominent player in the AI and machine learning space, has rolled out its latest update, v0.2. This release includes enhanced documentation, improved dataset handling, and new evaluator prompts, according to the LangChain Blog.
Product Updates
💡
Highlighting the latest product updates and news for LangChain, LangSmith, and LangGraph
LangChain
📝 The LangChain v0.2 release has versioned docs, with clearer structure and content.
The updated documentation is now organized into tutorials, how-to guides, conceptual guides, and API docs. This new structure aims to make it easier for users to find the information they need.
LangChain v0.2 docs walkthrough
Users can access the new docs in Python and JavaScript. Migration scripts are available to help users transition their existing codebases.
📞 Multi-modal function calling in JavaScript lets you pass images, video, and audio and get back structured output.
The new feature, available in the @langchain/google-genai or @langchain/google-vertexai packages, enhances the ability to build more dependable applications with multimodal support. More details can be found in the documentation.
LangSmith
✂️ Dataset splits for evaluation and filtering/editing dataset examples.
LangSmith now supports dataset splits, making it easier to run evaluations on subsets of data. This feature is particularly useful for testing new use cases and maintaining evaluation workflows. More information is available here.
🔁 Repetitions to build confidence in your experiment results.
LangSmith now allows users to run multiple repetitions of experiments to reduce noise and build more confidence in their results. A detailed guide on how to evaluate datasets with repetitions is available here.
🔧 Off-the-shelf online evaluator prompts to catch bad retrieval and hallucinations for RAG.
LangSmith’s new off-the-shelf evaluators can flag document relevance, check for hallucinations, and assess answer helpfulness. These evaluators can be customized to meet specific criteria, such as toxicity. More details can be found here.
💬 Manage private prompts without a handle.
The SDK and LangSmith Prompts UI have been improved to simplify prompt navigation. Now, only public prompts require a handle on creation, making it easier for new accounts to get started.
Upcoming Events
🎫
Meet up with LangChain enthusiasts, employees, and eager AI app builders at IRL events this coming month
🐻 June 18 (San Francisco): Berkeley LLM meetup. This event is open to Berkeley PhD students, faculty, and alumni. Attendees can learn about LangGraph and other OSS projects built at UC Berkeley. Sign up here.
🏙️ June 26 (NYC): LangChain and Elastic NYC meetup. The event will feature lightning talks, networking opportunities, and refreshments. Sign up here.
Speak the Lang
🦜
Real-life use cases and examples of how folks have used LangChain, LangSmith, or LangGraph to build LLM apps with high-quality and accuracy — even in production
Auto-correction for agents
Reflective loops prompt LLMs to reflect on and critique past actions, improving the quality and success rate of agents. A notable example includes self-correcting code assistants using LangGraph and Codestral.
Financial analysis with agents
LangChain agents are being used to reimagine traditional financial analysis. For instance, a GPT-4-powered AI agent has shown high accuracy rates in performance attribution analysis, as detailed in this paper by Bruno Veras De Melo.
Another example is the LangGraph expense tracker, a project that allows users to manage expenses by sending pictures of invoices for automated extraction and categorization.
Collaboration & Integrations
🤝
We 💚 helping users leverage partner features in the ecosystem by using our integrations
Generative AI Agents Developer Contest by NVIDIA and LangChain
Participants can create a text-based or multimodal agent using NVIDIA’s LLM stack and the LangChain framework. The contest runs until June 17th, with GPU prizes at stake. Enter the contest.
Code interpreter with Azure Container App Dynamic Sessions
The new integration with Dynamic Sessions allows agents to execute Python code safely. More information is available in this blog post.
Forced tool calls with Claude 3
Anthropic’s forced tool calling feature is now available for LangChain users, enabling reliable tool calls for various features. More details can be found here.
Nomic Embed local support for Python
LangChain users can now access a local version of Nomic Embed, allowing dynamic switching between local and remote inference based on input size and complexity. More information is available here.
For more updates, visit the LangChain blog and YouTube channel. Join the discussion on Discord or email support@langchain.dev for assistance.
Image source: Shutterstock.jpg)