MLflow

MLflow's ChatModel API makes it easy to build production-ready chat applications. Today we're sharing a beginner-friendly guide to help you get started. ChatModel provides a standardized way to create and deploy conversational AI models that are: 🔌 OpenAI-compatible out of the box 📊 Fully integrated with MLflow tracking and evaluation 🔄 Easy to share via Model Registry 🚀 Ready to deploy as REST APIs The guide walks you through the basics with practical examples, showing how to: - Map your application logic to ChatModel's standardized interface - Handle common inference parameters (temperature, max_tokens, etc.) - Pass custom parameters This is especially valuable when working with custom chat applications or locally hosted models, where you want MLflow's ecosystem features without complex integration work. Learn more: 📚 Guide: https://lnkd.in/gRpxA_2M #MLOps #MachineLearning #AI #MLflow #LLMOps

MLflow 2.18's enhanced Trace UI brings major usability improvements to GenAI observability, helping you better visualize and debug your LLM application flows. The new Trace UI brings significant improvements to help you investigate LLM application behavior: 📝 Enhanced content rendering with Markdown support—view formatted text, code blocks, and structured data exactly as they appear in your application 🔍 Standardized span component structure—navigate complex traces more intuitively ⚡️ Improved span content visualization—quickly identify and analyze key information in your execution traces These UI improvements streamline the process of auditing your GenAI applications, helping you pinpoint the source of unexpected behaviors and debug complex multi-agent interactions more efficiently. MLflow Tracing now provides an even better experience for understanding your application's behavior at every step. Try it out in MLflow 2.18: https://lnkd.in/gNbzX8_R #MLflow #LLMOps #AI #MLOps

Curious about the latest advancements in GenAI application lifecycle management? Our latest session dives into the newest integrations for GenAI, covering everything from application tracking and evaluation to deployment monitoring. 💡Discover where GenAI support in MLflow is headed and get a glimpse of our vision for integrating advanced AI-driven solutions. Don’t miss out on this essential roadmap for GenAI innovation! https://lnkd.in/gHkQD2Xa

MLflow 2.18 introduces native support for DSPy, making it easier to develop, track, and deploy LLM-powered applications that learn from examples. DSPy helps you build modular, optimizable LLM programs. Now with MLflow integration, you get: 🔄 Version control and reproducibility for your DSPy programs 📊 Automatic logging of prompts and completions during optimization 📦 Easy packaging of compiled DSPy programs for deployment 🔍 Built-in evaluation using MLflow's GenAI metrics This means you can focus on building your DSPy application while MLflow handles the ML lifecycle management—from experiment tracking to production deployment. Learn more: 📚 Documentation: https://lnkd.in/g__va9ss 💻 Quickstart Guide: https://lnkd.in/gPd7yiaZ 📝 Release Notes: https://lnkd.in/gNbzX8_R #MLflow #LLMOps #MachineLearning #AI #DSPy

Concerned about increased date rates? We got you! Harness data growth effectively, enabling agile data management and operational efficiency with the latest updates to MLflow. For the last year, our team has been focusing on building out a robust lifecycle management tooling into MLflow to support the Gen AI workstream. This allows users to have contextually aware responses to what they’re asking about, leveraging the data that input in an indexable system that can be supplied to a general language model. Users can now: ✅ Integrate ML flow tracing, allowing you to see inside the black box.  ✅ Have access to evaluate your retrieval relevance (MLflow evaluate) Explore the latest in MLflow and discover what the future holds for GenAI lifecycle management. https://lnkd.in/gHkQD2Xa

MLflow now supports Anthropic, Bedrock, Mistral, and TogetherAI as LLM judge providers for evaluating GenAI model outputs in MLflow 2.18. When evaluating LLM outputs, you can now use your preferred foundation model as the judge across MLflow's built-in metrics like answer_correctness and faithfulness. Here's what's new: 🤖 New provider support: Anthropic (Claude models), AWS Bedrock, Mistral, TogetherAI, alongside existing OpenAI support 🔄 Simple provider switching 🔐 Enterprise-ready with new proxy_url and extra_headers options for accessing providers through your security infrastructure Get started with the new providers: 📚 Documentation: https://lnkd.in/gvCGTep7 🔖 Release notes: https://lnkd.in/gNbzX8_R #MLflow #LLMOps #MachineLearning #AI

MLflow 2.18 brings major enhancements to GenAI development, tracing capabilities, and core APIs. This release adds thread and process-safety to MLflow's fluent APIs for tracking and registry operations, so you no longer need Client APIs for multi-threaded applications. Key updates include: 🔗 DSPy integration with built-in logging, loading, and tracing support 🖥️ Overhauled trace UI with markdown rendering and improved span components 📊 New one-line tracing for DSPy, LiteLLM, and Google Gemini ⚖️ Expanded LLM-as-a-Judge support including Anthropic, Bedrock, Mistral, and TogetherAI ⏰ Environment variable detection for smoother model deployment Note: The release notes include important information about upcoming changes to the ChatModel interface to standardize GenAI application development. Full release notes: https://lnkd.in/gNbzX8_R #MLflow #MachineLearning #AI #LLMOps #MLOps #DSPy

MLflow Tracing preserves error data and partial execution state when exceptions occur, making it easier to diagnose problems in GenAI applications. Tracing: 📊 Records all data captured up to the failure point 🔍 Stores exception details within trace events ❌ Shows error status clearly in the UI to identify failed operations This approach to error handling complements MLflow's broader tracing capabilities, ensuring full observability even when something goes wrong. Learn more: https://lnkd.in/ewefKiMJ #MLOps #MachineLearning #AI #LLMOps

🔥 Major upgrade in Unity Catalog v0.2! 🔥 You can now use Unity Catalog as the MLflow model registry backing resource to store, access, and govern registered models and model versions. Using MLflow and Unity Catalog together allows you to experiment with training runs and models in #MLflow and then easily register your final model(s) in Unity Catalog. ✔️ The diagram below shows how Unity Catalog and MLflow work together. 👇 Want to know more about v0.2's exciting upgrades? Read our full blog post! 🔗 https://hubs.la/Q02XlXyR0 #unitycatalog #opensource #oss #linuxfoundation

Our latest MLflow blog, by Jas Bali, shows how to integrate AWS Bedrock Agents with the MLflow ChatModel interface and tracing capabilities. This technical guide demonstrates implementing Bedrock Agent as a ChatModel, configuring Action Groups with Lambda functions, utilizing Knowledge Bases, and adding comprehensive tracing within MLflow. Illustrated through a space-geek-approved spacecraft launch window calculator, developers will learn to build AI applications with step-by-step traces of model reasoning and tool usage. The guide includes complete code for a custom ChatModel implementation, Lambda functions, OpenAPI schemas, and tracing configurations. https://lnkd.in/gBEg7Sqz #mlops #llmops #machinelearning #ai #aws