AI 服務暫時不可用，以下為來源正文，待恢復後補全翻譯。
AI Runtime is Generally Available | Databricks Blog Skip to main content AI Runtime is now generally available, bringing high-performance GPU infrastructure directly into Databricks. Scale reliably and experiment faster with seconds-level launches, flexible environments, and integrated observability — without the overhead of managing GPU infrastructure. Keep every training workload secure and governed with built-in Unity Catalog controls and workload isolation — in a new era of agent-led training. Today, we’re excited to announce the General Availability of AI Runtime (AIR), bringing fully managed GPU infrastructure directly into Databricks so customers can train their most demanding models where their enterprise data already lives. Too often, AI teams spend more time securing GPU capacity, managing failures, and waiting for experiments to launch than improving their models. Running GPU infrastructure separately from their Lakehouse adds governance gaps and puts sensitive training data at risk. As coding agents drive more training experiments, these bottlenecks and risks grow — making reliable, fast, and governed infrastructure essential. AIR brings GPU infrastructure, developer tools, and data governance together so teams can train at scale, iterate faster, and secure every workload without managing a separate GPU stack. Hundreds of companies, including Robinhood, PicPay, Merck, Flo Health, TinyFish, and FactSet, already use AIR to train and ship deep learning models to production, with use cases spanning healthcare, manufacturing, and consumer-facing agents. Our own Databricks AI Research team uses AIR to train the models behind KARL and AI Parse. With AIR, customers can: Train at scale with hassle-free GPU infrastructure: Serverless, on-demand access to advanced GPUs backed by reliable multinode training, optimized performance, and battle-tested by 100,000+ GPU-hour training runs. Experiment faster with near-instant launch latency and easy debugging: Bring your own code, containers, and Ray workloads – and obtain seconds-level launch latency on RDMA-enabled GPUs. Environment caching and built-in debugging tools to save time on every training loop. Secure and govern every GPU workload: AIR operates within the governance perimeter of Databricks and Unity Catalog, and is backed by years of investment in Serverless security. AIR helps ensure GPU workloads protect valuable training data, comply with access policies, and are tracked/auditable. Hassle-free GPU infrastructure Managing GPU infrastructure requires significant operational overhead, from capacity planning and node maintenance to networking and monitoring. With AIR, Databricks takes on the operational work of provisioning GPUs, maintaining healthy infrastructure, and optimizing the workload, so your team can focus on training: Get the GPU capacity you need, on demand. Databricks handles GPU procurement, cluster configuration, and networking so your team can start training without building and operating a full GPU platform. Scale with confidence on healthy infrastructure. Databricks runs dozens of checks for GPU and network health before and during training, so your workloads can scale to hundreds of GPUs reliably. Efficient training by default: Built-in data loaders stream and cache training data to reduce time spent waiting on I/O. Use tools such as UCVolumeDataset to improve data throughput without building a custom loading pipeline. AI Runtime validates GPUs before scheduling, continuously monitors active workloads, and periodically tests idle GPUs to detect and remove unhealthy nodes. Using Databricks AI Runtime, we moved our large-scale training from a single-node setup with 8 A100 GPUs to a multi-node setup with 16 H100 GPUs. This reduced training time from roughly seven days to 48 hours, a 3.5x speedup, making it much easier for our team to iterate on larger model experiments and train more efficiently.—Kostia Omelianchuk, Applied Research Scientist and Area Tech Lead, Superhuman Use GPUs wherever you work Access GPUs via CLI or interactively within Notebooks or IDEs. Your first workload just requires a single CLI command. With Databricks AIR Skills, migrating from a K8S or SLURM cluster just takes a single natural language prompt. Orchestrate GPU workloads with confidence with Lakeflow AIR integrates natively with Lakeflow so you can keep GPU workloads in sync with upstream and downstream pipelines. Moving to production requires just a few extra lines of YAML, with built in priority/pre-emption that minimize idle GPU time and monitoring tools like Genie ZeroOps that lower operational toil. Experiment, debug, and iterate faster Distributed training workloads are painful to set up and launch. AI teams waste time configuring training environments, waiting for GPU clusters to spin up, and debugging issues across the stack. AIR helps you get from code to training in seconds. You can use pre-built environments or bring your own container, and benefit from highly optimized environment caching so you just need to build environments once. AIR is compatible with all major frameworks, including Ray, so you can bring code over with minimal configuration. Databricks AI Runtime lets us train where our data already lives, with the reliability, security, and performance we need for large-scale model training. We can spend less time managing GPU infrastructure and more time improving the models themselves.—Vladislav Nedosekin, Director of Engineering, Data & AI Platform, Flo Health Launch your next experiment in seconds When you iterate and develop a training script, every second of delay is felt repeatedly. And so, we’ve optimized every layer of the provisioning stack to make workloads, even large scale RDMA-enabled clusters with dozens of GPUs, spin up in seconds. Unlike legacy GPU stacks that require full container image downloads before running code, AI Runtime uses warm serverless GPU pools and lazy image loading to start workloads in seconds, while continuing to load the remaining image data in the background. Debug distributed training in one place We’ve built an entirely new observability experience in MLflow, the industry’s leading experiment tracking tool. With this new experience, debugging your logs, which stream seconds after submission, is as simple as adding –watch to your AIR command. For deeper investigation, you can use our new logs viewer, tailor-built for multinode training. Genie Code accelerates debugging by analyzing AI Runtime training logs, identifying issues, and recommending next steps directly within Databricks. At Sonar, we help organizations build code confidence in their agentic software factories through independent code verification and governance. A team of our research engineers recently used Databricks AI Runtime to build LLM training pipelines and rapidly test different hypotheses. AI Runtime gave us the ability to move quickly from ideas to experiments, with observability of each stage in the pipeline.—Joe Tyler, Lead AI Researcher, Sonar Databricks' AI Runtime greatly streamlined the process of training a custom Text To Formula (TTF) model. With no infrastructure setup or delays, it was easy to choose the right compute based on prompt size and output token generation. This allowed us to move quickly, maintain our Lakehouse workflows, and deliver a high-quality model with full governance, reducing time to setup, train and deploy our model from days to hours.—Nikhil Sunderraj, Principal Machine Learning Engineer, FactSet Agentic debugging with built-in Genie Code Genie Code is embedded into the entire user experience, and comes pre-packaged with agent skills to help you debug issues and optimize GPU training performance. These skills include insights that our research and performance teams discover, like how we reduced Qwen training time by 6x on AIR. At Robinhood, AI Runtime helps our teams move from experimentation to training faster, without the infrastructure overhead. Fine-tuning is seamless, letting us train where our data lives while keeping governance and data controls built in. That means our engineers spend more time building better models and less time managing complex GPU setups.—Robert Macy III, Machine Learning Engineer, Robinhood Databricks AI Runtime lets us scale distributed training on serverless GPUs across millions of records, right inside our existing data workflows. As more of the top 20 biopharma companies rely on our models to inform development decisions, that speed and simplicity lets our team keep improving the models instead of managing the infrastructure.—Eran Barash, VP of Artificial Intelligence and Machine Learning, QuantHealth Secure and govern every GPU workload as part of the Lakehouse Managing GPU infrastructure separately from enterprise data creates governance gaps, fragmented access controls, and unnecessary data copies. As agents run hundreds of training experiments in parallel, teams need to ensure sensitive data stays governed and workloads remain securely isolated, without relying on manual oversight. The OpenAI Huggingface incident underscores this risk. AIR extends your trusted data controls and governance to every GPU workload. Unity Catalog enforces consistent read/write permissions on all aspects of a GPU workload (data access, checkpoints, logs, etc.), so your GPU platform and data governance layer now speak the exact same language. Isolate workloads and control network access AIR is backed by years of investment in serverless security, isolation, and private networking. Unlike GPU isolation mechanisms that expose the host kernel, AIR has VM-level isolation, supports PrivateLink and serverless egress controls. Thus AIR offers much stronger protections for training workloads compared to conventional systems. Understand where your GPU budget goes Track AIR usage and costs through Databricks billing system tables, alongside your other Databricks workloads. Use workload-level usage records to identify which training workloads consume your GPU budget and guide decisions about future runs. With Databricks AI Runtime and Serverless, we’ve been able to scale LLM training and inference more quickly, working with larger models, multiple H100 GPUs, and large volumes of data. The integration with our existing pipelines, together with Databricks’ orchestration and monitoring capabilities, has made the end-to-end workflow simpler and more efficient, reducing infrastructure complexity and development cycle time.—Vitor Rolim, Staff Machine Learning Engineer, PicPay For us, training the model is only part of the problem. We need to know whether a change actually makes the agent better at completing real tasks, and then be able to get that improvement into production quickly. With AI Runtime, we can train on GPUs, evaluate with MLflow, and serve our web-native model without building a separate stack for every stage.—Zach Richardson, Head of AI, TinyFish Get started with your first GPU workload To help you get started, we’ve put together several template notebooks and starter guides: Please see our quickstart to get started, and connect to AI Runtime in seconds. Please visit our web page to learn more about AI Runtime. Interested in trying AIR out on an example? Clone our cookbook repo to get started in just one CLI command! Please reach out to your account team to learn more or if you have any questions! Get the latest posts in your inbox Subscribe to our blog and get the latest posts delivered to your inbox. Sign up View all blogs