Boost the performance of your Keras models by integrating with New Relic, providing comprehensive monitoring and optimization for the entire inference process.

What's included?

Keras observability quickstart contains 3 alerts. These alerts detect changes in key performance metrics. Integrate these alerts with your favorite tools (like Slack, PagerDuty, etc.) and New Relic will let you know when something needs your attention.
This alert is triggered if prediction exceeds 1000 for 5 minutes.
Model Drift
This alert is triggered if model drift exceeds 1 for 5 minutes.
Data Drift
This alert is triggered if data drift exceeds 1 for 5 minutes.
This quickstart doesn't include any dashboards. Do you think it should?
You can edit this quickstart to add helpful components. View the repository and open a pull request.
View repo View repo

Why monitor your Keras models during inference?

Once Keras models are trained and deployed, monitoring during the inference phase becomes paramount for various reasons:

Improve performance:

Inference, especially with large models, can be computationally intensive. Monitoring metrics like inference latency, memory usage, and GPU utilization can identify bottlenecks, allowing for necessary optimizations.

Model health:

Over time, the distribution of input data might change, a phenomenon known as data drift. Monitoring allows for the early detection of such scenarios, ensuring the model remains robust and accurate.

Data quality:

As models make predictions on new, unseen data, ensuring the quality and consistency of input data is vital. Monitoring can help detect anomalies or inconsistencies in real-time data that could adversely affect model output.

Comprehensive monitoring for Keras models during inference

Properly monitoring your Keras models during the inference phase ensures optimal performance and sustained accuracy. Tracking key metrics will help maintain model health and swiftly detect potential issues.

What’s included in the Keras monitoring quickstart?

The New Relic Keras monitoring quickstart offers specialized out-of-the-box reporting for inference:

  • Real-time visibility: Gain insights into the performance of your Keras models, tracking metrics like inference latency and throughput in real-time.
  • Data Integrity Checks: Ensure that the input data for inference aligns with expected formats and does not have any anomalies that could affect model outputs.
  • Proactive alerts: Receive immediate notifications about critical issues affecting your Keras model's performance or reliability during inference.
  • Resource Utilization: Monitor the computational resources (CPU, GPU, memory) being used during the inference process to optimize deployments and control costs.

How to use this quickstart

  • Sign Up for a free New Relic account or Log In to your existing account.
  • Click the install button.
  • Install the quickstart to get started or improve how you monitor your environment. They're filled with pre-built resources like dashboards, instrumentation, and alerts.
New Relic
Jyothi Surampudi
Need help? Visit our Support Center or check out our community forum, the Explorers Hub.
Related resources