Perpetual ML
Perpetual ML is an AI tool that leverages a unique technology, known as Perpetual Learning, to drastically accelerate model training. This acceleration is chief...
Last verified:
What is Perpetual ML?
Perpetual ML Suite is a unified, batteries-included ML studio designed for solo developers and data science teams. It leverages PerpetualBooster, the #1 algorithm on AutoML benchmarks, to automatically train models without requiring hyperparameter optimization. The platform enables continual learning that reduces total training time from O(n^2) to O(n), allowing models to continuously learn from real-time data without starting over. Users can track and compare all model experiments, deploy models for batch or real-time inference, and monitor metrics, data drift, and model drift proactively.
Key features include Auto Train with PerpetualBooster, Continual Learning, Optimal Business Decisioning that directly optimizes user-defined business objectives (maximizing profit, minimizing risk), Experiment Tracking, Model Registry with version control, Monitoring without retraining or ground truth, Deployment for batch and real-time inference, Marimo Notebooks for reactive collaborative development, and native Data Platform integration with Snowflake (with upcoming Databricks support). The suite integrates directly with existing data warehouses so data never leaves your environment.
Perpetual ML is ideal for solo developers, data science teams, and organizations using modern data warehouses like Snowflake. It works with various machine learning tasks including classification, regression, time-series, ranking, and text classification. The Rust-based backend provides compatibility with Python, C++, Java, and more. The platform operates efficiently without needing specialized hardware like GPUs or TPUs, making it accessible for teams without expensive infrastructure.
Perpetual ML pricing
Pricing model: Free
Perpetual ML offers pay-as-you-go pricing with two options. Perpetual ML Cloud uses usage-based pricing: Shared CPU at $0.00014 per vCPU second for training, batch inference, real-time ML endpoints, and Marimo notebook execution; Persistent Storage at $0.000069 per GB hour for model artifacts, logs, datasets; and Egress at $0.10 per GB. The Marketplace option has tiered pricing based on monthly task volume: 1-9 Tasks is Free, 10-99 Tasks is $9.9 per task per month, and 100+ Tasks is $0.9 per task per month. Included tasks are auto train, continual learning, metric monitoring, data drift monitoring, concept drift monitoring, and optimal business decisioning. Enterprise support and all other features are free. AWS and Snowflake infrastructure costs are billed separately by those providers. Custom pricing options are available by contacting the company.
Perpetual ML pros
- No hyperparameter optimization needed - self-generalizing algorithm
- PerpetualBooster is #1 on AutoML benchmark
- 100x faster model training than traditional methods
- Continual learning reduces training time from O(n^2) to O(n)
- Update models with new data without restarting training
- Directly optimizes user-defined business objectives like profit or risk
- Monitor data drift and model drift without retraining
- Monitoring works without ground truth data
- Native Snowflake integration - data never leaves warehouse
- Upcoming Databricks support expanding platform compatibility
- No GPU or TPU required - works on standard hardware
- Rust-based backend compatible with Python, C++, Java
- Secure version-controlled model registry for collaboration
- Marimo notebooks for reactive collaborative development
- Unified platform for batch and real-time inference deployment
- Supports classification, regression, time-series, ranking, text classification
- Experiment tracking to reproduce all Auto Train experiments
- Low-code/no-code interface accessible to non-experts
- Enterprise support included free of charge
- Scalable and explainable ML platform
Perpetual ML cons
- Currently only supports Snowflake - Databricks still upcoming
- AWS and Snowflake infrastructure costs billed separately
- Pay-per-task pricing can become expensive at higher volumes
- $9.9 per task for 10-99 tasks is costly for mid-volume users
- Cloud egress fees at $0.10 per GB add to total cost
- Shared CPU pricing at $0.00014 per vCPU second for training
- Persistent storage at $0.000069 per GB hour adds ongoing costs
- No annual billing discounts mentioned - purely pay-as-you-go
- Custom pricing requires contacting sales - not self-service
- Marimo notebooks feature listed as coming soon in some areas
Frequently asked questions about Perpetual ML
What is PerpetualBooster and why is it special?
PerpetualBooster is a gradient boosting machine (GBM) algorithm that doesn't need hyperparameter optimization, unlike other GBM algorithms. It is self-generalizing and ranks as the #1 algorithm on AutoML benchmarks, delivering state-of-the-art predictive performance while being easy to use.
How does continual learning work in Perpetual ML?
Continual learning significantly cuts total training time from O(n^2) to O(n), where n represents the number of batches. Instead of time-consuming periodic retraining, the platform continuously learns from real-time data, allowing you to update models with new data without starting over. This is especially critical for large datasets and applications like fraud detection.
What data platforms does Perpetual ML support?
Perpetual ML is currently natively integrated with Snowflake, meaning your data never leaves your Snowflake data warehouse. You get the same security and governance policies with powerful ML tools. Upcoming support includes Databricks and other data platforms.
Do I need special hardware like GPUs to use Perpetual ML?
No, Perpetual ML is designed to be efficient without needing special hardware. It operates without the requirement for GPUs or TPUs, making it accessible for teams without expensive infrastructure.
What machine learning tasks does Perpetual ML support?
Perpetual ML works with a variety of machine learning tasks including classification, regression, time-series, ranking, and text classification. It handles both batch inference and real-time inference scenarios.
How does monitoring work without ground truth?
Perpetual ML's monitoring feature effortlessly tracks metrics, data drift, and model drift without the need for retraining or ground truth data. This allows you to proactively detect and respond to changes in your data environment without waiting for labeled outcomes.
What programming languages can I use with Perpetual ML?
The Rust-based backend makes Perpetual ML compatible with many programming languages including Python, C++, Java, and more, providing flexibility for different development environments.
Is there a free tier available?
Yes, the Marketplace pricing includes a free tier for 1-9 tasks per month. Additionally, enterprise support and all other features beyond the included tasks are free of charge. You can also try everything with no time limit on the free plan.
How does Optimal Business Decisioning work?
Optimal Business Decisioning directly optimizes your user-defined business objective such as maximizing profit or minimizing risk. This ensures your ML models drive truly optimal business actions rather than just optimizing for traditional metrics like accuracy.
Where can I get Perpetual ML Suite?
Perpetual ML Suite is available as a Native App on the Snowflake Marketplace. You can check the app at app.snowflake.com/marketplace/listing/GZSYZX0EMJ/perpetual-ml-perpetual-ml-suite. The platform integrates directly with your existing Snowflake account.