Learn more — ValidMind Library
The ValidMind Library is a Python library and documentation engine designed to streamline the process of documenting various types of models.
EU AI Act Compliance — Read our original regulation brief on how the EU AI Act aims to balance innovation with safety and accountability, setting standards for responsible AI use
December 6, 2024
We’ve chosen some new names for our product offerings, expanded our training courses, introduced new large language model (LLM) features — and much, much, more!
To reduce ambiguity and highlight the capabilities of our developer tools, as of v2.5.25
what was previously referred to as the ValidMind Developer Framework is now known as the ValidMind Library.
In addition to the enhancements in this release, references to the developer framework have been replaced across the ValidMind Library package and open-source repository, as well as the ValidMind Platform and the user documentation.
Check out our RAG Model Documentation Demo Jupyter Notebook!
Learn how to build a basic RAG (Retrieval-Augmented Generation) model to help automatically answer RFP (Request for Proposal) questions using GenAI.
From in-depth tests assessing generation quality to checks detecting bias and toxicity, use LangChain2 and ValidMind to build, evaluate, and document a simple RAG model as it is being developed.
To reduce ambiguity and highlight the capabilities of our cloud-hosted interface, as of v1.27.7
what was previously referred to as the ValidMind Platform UI is now simply known as the ValidMind Platform.
In addition to the enhancements in this release, references to the platform UI have been replaced across the ValidMind Library package and open-source repository, as well as the ValidMind Platform and the user documentation.
You can now submit feature requests, bug reports, or any other feedback directly within the ValidMind Platform.
Click on the Talk to us. tab on the right-hand side of any screen and follow three easy steps:
You can now add, edit, or remove custom data to your analytics within the ValidMind Platform in two steps:
In concert with our product renaming across the ValidMind Library and the ValidMind Platform, we’ve edited all user guides and external facing documentation to reflect these updates.
Of note is the Developers drop-down menu, previously called Developer Framework.
Along with keeping our content up to date,4 we’ve refreshed the look and feel of our documentation to reflect ValidMind’s new brand colors used on our main website5 and within the ValidMind Platform.
We’ve overhauled the documentation homepage. Check it out:
The ValidMind logo has been updated across the documentation, and a new style applied to our side navigation and table of contents to unify the experience across our products:
To complement our rebranding,6 our ValidMind Academy training now features its own theme and landing page to distinguish the material from the rest of the documentation along with content updates.7
ValidMind offers several specialized features that use large language models (LLMs), such as our Document Checker. These features streamline model risk management and ensure regulatory compliance.
We’ve created a guide on how we approach these features and what you need to know:
In addition to comparison tests using the input_grid
parameter in run_test()
functionality, we’ve added similar support for param
s:
run_test()
function allows you to pass a param_grid
.Example
For this param
grid:
param_grid = {
"param1": [1],
"param2": [0.1, 0.2],
}
A test runs once for each of the following param
groups:
{"param1": 1, "param2": 0.1}
{"param1": 1, "param2": 0.2}
With a new name8 comes a fresh skin — the ValidMind Platform’s theme now matches our main website and documentation, featuring our new logo and accent color:
We now support including attachment fields as a request field on a User Action step within workflows, allowing you to directly request supporting documentation for transitioning a model into its next lifecycle status.
We continue to add new functionality to model workflows within the ValidMind Platform, including the ability to now select a User Role or a User Type inventory field in an Approval step, and the ability to configure diverging paths with Condition Branch steps.
Custom model inventory fields are now more flexible than ever — introducing the ability to define a formula(params)
function that automatically calculates and returns a value based on the params dictionary, which includes selected custom field keys retrieved from your other inventory model fields.
Example formula:
def formula(params):
# High Risk: If materiality is high risk, return high risk regardless of complexity
if params.materiality == "High Risk":
return "High Risk"
# Medium Risk: If materiality is low risk but complexity is high risk, return medium risk
if params.materiality == "Low Risk" and params.complexity == "High Risk":
return "Medium Risk"
# Low Risk: Both materiality and complexity are low risk
return "Low Risk"
You can now annotate attachments uploaded to both models or model findings, allowing you to describe the contents of the file or provide commentary.
If more than one set of test results has been logged with the ValidMind Library to the ValidMind Platform, you can now filter your historical test results for ease of access.
When working with validation reports, linking developer evidence is now associated with a documentation section, making validation reports clearer and the process easier for you to assess compliance.
We’ve been working hard to keep our documentation up to date with our lightning-fast feature releases. Below is a round-up of some notable user guide improvements not covered by independent feature release announcements:
Signing up for ValidMind is free — and now you can follow our step-by-step registration guide to get you started.
After you sign up for ValidMind, refer to our guides on setting up your organization, including options for establishing a private connection between ValidMind and your company network.
We’ve clarified prerequisites for tasks and the permissions default roles have and improved our information on user management to help you understand how access to ValidMind works.
To complement the ability to add your own model inventory fields, model registration within ValidMind Platform is now a two-step process, broken down into Basic Model Information and Additional Model Information with a matching user guide.
To better align documentation with what users see within the ValidMind Platform, our previous article titled “View documentation activity” has been renamed to View model activity and now lives under the model inventory sub-section under Guides.
This new guide features expanded information on the auditing feature, including information on how to filter tracked activity.
We’ve expanded our reference guides on working with documentation templates, including updated information on customizing templates, and swapping between versions of different types of templates.
We’ve expanded our reference guide on working with content blocks, including updated information on different content block types, an overview of our content editing toolbar, and a section on how to insert math formulas right into your documentation.
Our reference material on model findings has been expanded to include key concepts, and have been broken down into two sub-guides detailing viewing and filtering model findings, and adding and managing model findings.
On top of support for custom analytics,10 we’ve updated our feature documentation for reporting to include overviews of default report pages provided out-of-the-box by ValidMind.
Information on available analytics and setting up reporting has also moved into its own section under Guides in reporting.
In our Jupyter Notebook samples, we’ve now included information within the notebooks themselves on how to upgrade to the latest version of the ValidMind Library.
Model lifecycle statuses have been renamed in the ValidMind Platform and documentation updated to reflect this shift.
Custom fields have been renamed in the ValidMind Platform and documentation updated to reflect this change.
Did you know that along with the ValidMind Library, the code for our documentation site is also open-source?
Below are some improvements we’ve made to our authoring tools to streamline the process for contributors:
To support the release of ValidMind Academy, we’ve included a section in our style guide to cover guidelines unique to our training materials.
.button
and .video
CSS classesPreviously, visiting a non-existing page within the documentation would present you with a poor user experience in the form of a default XML error.
We’ve now introduced a custom 404 page where you can perform a search instead.
To access the latest version of the ValidMind Platform,12 hard refresh your browser tab:
Ctrl
+ Shift
+ R
OR Ctrl
+ F5
⌘ Cmd
+ Shift
+ R
OR hold down ⌘ Cmd
and click the Reload
buttonTo upgrade the ValidMind Library:13
In your Jupyter Notebook:
Then within a code cell or your terminal, run:
You may need to restart your kernel after running the upgrade package for changes to be applied.