Validating Generative AI — Read our original technical brief created in collaboration with Parker & Lawrence Growth Advisory
Get started
Published
November 20, 2024
ValidMind is designed to streamline the management of risk for AI models, including those used in machine learning (ML), natural language processing (NLP), and large language models (LLMs). ValidMind offers tools that cater to both model developers and validators, simplifying key aspects of model risk management.
What do I use ValidMind for?
Model developers and validators play important roles in managing model risk, including risk that stems from generative AI and machine learning models. From complying with regulations to ensuring that institutional standards are followed, your team members are tasked with the careful documentation, testing, and independent validation of models.
The purpose of these efforts is to ensure that good risk management principles are followed throughout the model lifecycle. To assist you with these processes of documenting and validating models, ValidMind provides a number of tools that you can employ regardless of the technology used to build your models.
The two main components of ValidMind. The ValidMind Library that integrates with your existing developer environment, and the ValidMind Platform.
The ValidMind AI risk platform provides two main product components:
The ValidMind Library is a Python library of tools and methods designed to automate generating model documentation and running validation tests. The library is designed to be platform agnostic and integrates with your existing development environment.
For Python developers, a single installation command provides access to all the functions:
%pip install validmind
The ValidMind Platform is an easy-to-use web-based interface that enables you to track the model lifecycle:
Customize workflows to adhere to and oversee your model risk management process.
Review and edit the documentation and test metrics generated by the library.
Collaborate with and capture feedback from model developers and model validators.
Generate validation reports and approvals.
For more information about the benefits that ValidMind can offer, check out the ValidMind overview.
Key ValidMind concepts
model documentation
A structured and detailed record pertaining to a model, encompassing key components such as its underlying assumptions, methodologies, data sources, inputs, performance metrics, evaluations, limitations, and intended uses.
Within the realm of model risk management, this documentation serves to ensure transparency, adherence to regulatory requirements, and a clear understanding of potential risks associated with the model’s application.
validation report
A formal document produced after a model validation process, outlining the findings, assessments, and recommendations related to a specific model’s performance, appropriateness, and limitations. Provides a comprehensive review of the model’s conceptual framework, data sources and integrity, calibration methods, and performance outcomes.
Within model risk management, the validation report is crucial for ensuring transparency, demonstrating regulatory compliance, and offering actionable insights for model refinement or adjustments.
template, documentation template
Functions as a test suite and lays out the structure of model documentation, segmented into various sections and sub-sections. Documentation templates define the structure of your model documentation, specifying the tests that should be run, and how the results should be displayed.
ValidMind templates come with pre-defined sections, similar to test placeholders, including boilerplates and spaces designated for documentation and test results. When rendered, produces a document that model developers can use for model validation.
test
A function contained in the library, designed to run a specific quantitative test on the dataset or model. Test results are sent to the ValidMind Platform to generate the model documentation according to the template that is associated with the documentation.
Tests are the building blocks of ValidMind, used to evaluate and document models and datasets, and can be run individually or as part of a suite defined by your model documentation template.
metrics, custom metrics
Metrics are a subset of tests that do not have thresholds. Custom metrics are functions that you define to evaluate your model or dataset. These functions can be registered via the ValidMind Library to be used with the ValidMind Platform.
In the context of ValidMind’s Jupyter Notebooks, metrics and tests can be thought of as interchangeable concepts.
inputs
Objects to be evaluated and documented in the ValidMind Library. They can be any of the following:
model: A single model that has been initialized in ValidMind with vm.init_model(). See the Model Documentation or the for more information.
dataset: Single dataset that has been initialized in ValidMind with vm.init_dataset(). See the Dataset Documentation for more information.
models: A list of ValidMind models - usually this is used when you want to compare multiple models in your custom tests.
datasets: A list of ValidMind datasets - usually this is used when you want to compare multiple datasets in your custom tests. See this example for more information.
parameters
Additional arguments that can be passed when running a ValidMind test, used to pass additional information to a test, customize its behavior, or provide additional context.
outputs
Custom tests can return elements like tables or plots. Tables may be a list of dictionaries (each representing a row) or a pandas DataFrame. Plots may be matplotlib or plotly figures.
test suite
A collection of tests which are run together to generate model documentation end-to-end for specific use cases.
For example, the classifier_full_suite test suite runs tests from the tabular_dataset and classifier test suites to fully document the data and model sections for binary classification model use cases.
How do I get started?
On the ValidMind Platform, everything starts with the model inventory — you first register a new model and then manage the model lifecycle through the different activities that are part of your existing model risk management processes.
Approval workflow
A typical high-level model approval workflow looks like this:
graph LR
A[Model<br>registration] --> B[Initial<br>validation]
B --> C[Validation<br>approval]
C --> D[In production]
D --> E[Periodic review<br>and revalidation]
E --> B
Select a documentation template when registering a new inventory model to start your model documentation. You then use the model inventory to manage the metadata associated with the model, including all compliance and regulatory attributes.
Initial validation
Triggers a new workflow2 to yield a model that will be ready for production deployment after its documentation and validation reports have been approved.
Perform validation of the model to ensure that it meets the needs for which it was designed. You can also connect to third-party systems to send events when a model has been approved for production.
In production
Use the model in production while ensuring its ongoing reliability, accuracy, and compliance with regulations by monitoring the model’s performance.
Periodic review and revalidation
As part of regular performance monitoring or change management, you follow a process similar to that seen in the Initial validation step.
