MLflow

The Data Engineering and Machine Learning (DEML) Summit 2024 is happening this Thursday and Friday (Oct 3rd - Oct 4th)! ⭐ Speaker Highlight - Benjamin Wilson ⭐ Session Title: "Simplifying GenAI Agents with MLflow" Session Time: 12:00PM on Fri, Oct 4th (PT) Session Summary: In this talk, Ben Wilson will showcase a full end-to-end workflow for creating a custom GenAI Agent that can be easily deployed to production using the features within MLflow. He'll cover the ins and outs of building a chat interface, including custom code dependencies, and explore the flexibility in the model abstraction capabilities within MLflow for GenAI applications. You'll walk away from this talk feeling a bit more confident in how to build robust solutions that leverage the latest in agentic applications and see how monitoring capabilities like MLflow Tracing can help to enhance both your development and production monitoring needs. 👉 Register for the conference and session here: https://lnkd.in/eFq5s-XF (Special thank you to Databricks for sponsoring the conference.) #dataengineering #softwareengineering #machinelearning #datascience #ai

MLflow offers advanced support for GenAI development, with many features that simplify working with Large Language Models (LLMs) and other generative AI technologies. MLflow provides: 🔧 Tools for custom GenAI model creation 📊 LLM evaluation capabilities 🐍 Custom PyFunc integration with LLMs 🔍 Evaluation for Retrieval Augmented Generation (RAG) 🔗 Integration with popular GenAI libraries like LangChain and LlamaIndex These tools help AI Engineers enhance their GenAI workflows, from model development to deployment and evaluation. Want to use MLflow for your GenAI projects? Check out our tutorials and guides covering various use cases to simplify your work with generative AI: https://lnkd.in/ghUmyezm #MLflow #GenAI #MachineLearning #AI #LLMOps

This new blog post by Awadelrahman Ahmed offers a comprehensive guide to MLflow's models from code approach to model logging. It explains what models from code logging is, how it differs from serialization-based logging, when it is appropriate to use, and how to start using it in your own model-logging workflows. Key takeaways: 🔹 Models from code logging allows you to log the code that represents your model, rather than a serialized object 🔹 It's particularly useful for models that rely on external services, combine multiple models, or don't work with traditional serialization 🔹 Implementation involves creating two files: one for the model code and another for logging 🔹 The post provides a simple example of logging a circle area calculation as a model This approach opens up new possibilities for managing compound AI systems and non-traditional models within the MLflow ecosystem. Whether you're working with GenAI, rule-based algorithms, or multi-model systems, models from code logging offers a flexible solution for tracking and versioning your work. Read the full post here: https://lnkd.in/gBcqT6eV #llms #ai #mlflow #machinelearning #mlops #llmops

Using MLflow's custom PyFunc with AutoGen enables AI agent system versioning and reproducibility, allowing for easier tracking and deployment of multi-agent frameworks. In this technical guide, Michael Berk demonstrates how to: 🤖 Develop a multi-agent framework with AutoGen 📦 Package the agent as an MLflow custom PyFunc model 📊 Use MLflow for model logging, loading, and artifact management 🔍 Implement MLflow tracing for detailed execution analysis The post walks through creating an image generation agent that adds cats to scenes, showcasing practical MLOps techniques for complex AI systems. Read the full post here: https://lnkd.in/dqF_FY6n #MLOps #AI #LLM #MachineLearning

"Models from Code Logging" is a new feature in MLflow that I believe will help with many use cases! I invested time last month to dig into this feature, learning its ins and outs, and what it can be used for! I found a couple of good use cases and I wanted to share what I have learnt about the WHATs, WHYs, and HOWs of the new Models from Code Logging feature! So in this post you will learn What its, Why and When to use it and How to implement it! https://lnkd.in/dW9vAzFj Hope this to be a good read for my #ML, #AI, and #Data colleagues 🙂 #MLOps #MLflow

MLflow tracking adapts to your project's scale, from solo experiments to team-wide deployments. Here are three key configurations: 💻 Localhost (default): Designed for individual work and quick setup. All data saves locally, providing a straightforward start. 🗃️ Local tracking with database: Enhances solo projects with improved data management. Connecting a database allows for more organized experiment tracking. 🌐 Remote tracking with MLflow Tracking Server: Built for team collaboration. Centrally store and share artifacts and metadata securely. Each configuration has distinct advantages. Our guides provide detailed setup instructions: - QuickStart: https://lnkd.in/gUWMyTZB - Local database tracking: https://lnkd.in/gtQD_7ev - Remote tracking server: https://lnkd.in/gyr_PBHK #MLOps #MachineLearning #MLflow

Listen to Michael Berk and Jerry Liu talk about mitigating LLM hallucinations in our recent webinar with LlamaIndex! Jerry talks about the importance of human-in-the-loop and UX in managing hallucinations. Hallucinations of some form will probably be an issue for some time to come—accepting this fact and building mitigations into your compound ai systems is very important! Check out the full webinar here: https://lnkd.in/gCbzpYrU #mlflow #llamaindex #llm #ai #machinelearning

Another absolute gem by Médéric HURIER. A cookie cutter template to easily start your MLOps projects. It's suitable for Kubernetes, Databricks, Azure ML, Vertex AI, and AWS SageMaker! 𝗧𝗵𝗲 𝘁𝗲𝗺𝗽𝗹𝗮𝘁𝗲 𝗰𝗼𝗻𝘁𝗮𝗶𝗻𝘀 - Streamlined project structure - Automated testing - Pre-commit / linting - Poetry integration - MLflow readiness out of the box - Invoke task automation - Dockerized deployment The great thing about this template is that it sticks to a few very good and reusable basics, which can be deployed to different platforms. Also, the documentation is excellent! With lots of links to other relevant documentation. Give this repo a ⭐ to easily jumpstart your MLOps projects! 🔗 found ⬇️ #machinelearning #datascience #softwareengineering

The recent MLflow 2.16 release included a new, in-depth guide to models-from-code logging. This feature allows you to pass a script path instead of an object reference when logging custom models, bringing several advantages: 🚀 Improved portability and compatibility across Python versions 🔧 Better handling of complex objects and system resources 📖 Human-readable model definitions ⚡ Enhanced performance, especially for complex implementations By executing the script in the target environment, MLflow sidesteps limitations of serialization libraries like pickle or cloudpickle. This approach offers more flexibility and reliability in model deployment and sharing. Check out the full guide to learn how this feature can streamline your ML workflows and make your models more portable and maintainable: https://lnkd.in/gtK7phFn #MLflow #MachineLearning #AI

Unlock the power of your data and LLMs with LlamaIndex and MLflow! This recent webinar covers, among other things, how to use MLflow Evaluate with LlamaIndex. With MLflow evaluate, you can: ⏱️ Measure response times and optimize latency for better user experience 📊 Assess output quality using built-in metrics like relevance and coherence 🔄 Compare different models or prompts to find the best fit for your use case The webinar also shows how to track LlamaIndex components in MLflow, package LlamaIndex engines for deployment, and leverage MLflow Tracing for better visibility into your application's behavior. Full Webinar: https://lnkd.in/gCbzpYrU MLflow LlamaIndex Flavor Docs: https://lnkd.in/ddqxZiB3 #ai #llm #mlflow #llamaindex