Skip to main content
Ctrl+K
CloudAI - Home CloudAI - Home

CloudAI

CloudAI - Home CloudAI - Home

CloudAI

Table of Contents

  • Overview
  • Get started
    • Install CloudAI
    • Run your first experiment
    • Understand experiment configuration
  • Run experiments
    • Set up your execution environment
    • Run on Slurm
    • Run on Kubernetes
    • Compose a scenario
    • Prepare and clean up with hooks
    • Prepare model weights
    • Prepare workload prerequisites
    • Inspect commands before execution
    • Read results and generate reports
  • Optimize configurations
    • Run a configuration search
    • Choose an objective and search space
    • Plan compute usage
    • Compare trade-offs and reuse configurations
  • Automate experiments
    • Run experiments with Python
    • Handle results and failures
    • Publish results and integrate with CI
  • Workloads
    • Training
      • MegatronBridge
      • Nemo Run
      • DDLB
    • Inference
      • vLLM
      • SGLang
      • AI Dynamo
      • SRT-Slurm
    • Networking and communication
      • NCCL
      • UCC
      • OSU
      • DeepEP Benchmark
      • NIXL Bench
      • NIXL EP
      • NIXL KVBench
      • NIXL CTPerf
    • Storage
      • fio
    • Simulation and prediction
      • AIConfigurator
      • DynamoMocker
      • Chakra Replay
    • General-purpose execution
      • Bash Command
      • Slurm Container
      • Sleep
    • Legacy integrations
      • NeMo Launcher (deprecated)
      • JaxToolbox workloads (DEPRECATED)
  • Extend CloudAI
    • Architecture and extension points
    • Add a workload
    • Add a reporter
    • Add an optimization agent
    • Add an execution system or runner
    • Develop and contribute
  • Reference
    • CLI reference
    • Workload and system support
    • Systems
      • NCCL on RunAI
    • System configuration example
    • Shared system settings
    • Test definitions
    • Scenario configuration
    • Agent and search settings
    • Reports
      • Training Report Schema
    • Python API
  • Troubleshooting
  • Reference
  • Reports

Reports#

Use these schemas when consuming generated reports programmatically. For the first inspection of an experiment, start with Read results and generate reports.

CloudAI report schemas describe the generated output for supported workload categories.

  • Training Report Schema

previous

Agent and search settings

next

Training Report Schema
NVIDIA NVIDIA
Privacy Policy | Your Privacy Choices | Terms of Service | Accessibility | Corporate Policies | Product Security | Contact

Copyright © 2026, NVIDIA CORPORATION & AFFILIATES.