We are excited to unveil the latest update to our platform. This release introduces a variety of new features, significant enhancements, and backend updates. Below are the specifics of the updates included in this release.
New Features
Data Access Selector
The AI Suite now includes a newly implemented "Data Fabric Selector" which facilitates the creation of customized data fabrics, managing connected Layar systems, and adding or removing Layar systems to Data Fabrics. This feature also enables the restriction of data access and sharing of data fabrics among users while assigning varying levels of read/write permissions.
š Users configured with access to any of the Applications with the Data Selector must have a seat for Cortex.
Applications with the Data Access Selector
The following applications have the Data Access Restriction setting.
Single Sign-On (SSO)
Single Sign-On functionality via Reverse Proxy has been added as an option for the AI Suite.
Account
The administrators are now empowered with a new capability to produce a password reset link for their users. This addition to the system is intended to provide an efficient solution for users who may have forgotten their passwords or need to reset them for security reasons. With this feature, administrators can easily generate a secure link that facilitates password reset without requiring manual intervention from IT support personnel.
š”The "reset password" option is hidden if SSO is enabled.
Layar
New parameters to control Rag
Add an API option to override the RAG query terms that are used to return information to feed to the LLM
Hybrid search ratio parameters to allow heavier weighting on keyword or vector search depending on the use case
Ability to turn reranker on or off
Enable/Disable text chunk expansion
Composer
Includes support for Mixtral.
Requires 2x the compute for inference and has 6x the parameters.
Signal
The Question bar has been added to the side panel.
A sorting option has been integrated to facilitate sorting by date and relevance.
Codex
This release includes the option to add custom links for user manuals within a dataset.
Improvements
Layar
The API documentation for the Generative AI endpoint has been updated to include new parameters and options, enhancing the handling of LLM query phrases.
Enhancements to the reranking logic have been made to optimize the delivery of pertinent context chunks to the LLM.
An enhanced embedding model for document-centric Retrieval-Augmented Generation (RAG) systems has been developed. This model aims to enhance the retrieval process by accurately identifying and incorporating relevant information from documents into the generation workflow, thereby improving the overall performance of RAG systems.
Enhancements in the performance of workers processing requests to reduce memory consumption.
Enhancements to the Data Fabric include the ability to restrict data access, share data fabrics among users, and assign varying levels of read/write permissions.
Enhance email templates to provide improved support for deep links.
We have implemented an option for Single Sign-On through Reverse Proxy.
The link to the Certara AI Help Center has been updated within the application.
Enhanced support for the upload of Rich Text Format (RTF) files.
The user interface sidebar of Layar has been simplified to provide a more streamlined and user-friendly experience.
Codex
Table and plot downloads contain the indication abbreviation in the filename
Endpoint Correlation Plot
To form a spline, increase the number of unique values required.
Covariate vs Endpoint Plot
Displays error plot when data is missing.
Funnel plot
There was an update to the converge algorithm settings.
Backend Updates
This release includes an enhanced text similarity search performance for Retrieval Augmented Generation (RAG) and related applications for improving the accuracy and efficiency of information retrieval systems. Techniques such as chunking, re-ranking, and query transformations significantly improve RAG systems by providing more relevant results and tailored responses.
The capability to override chemistry models via configuration settings has been implemented.
The LLM model was enabled to operate in clustered mode across multiple GPUs.
Includes the capability to activate a configuration option that permits the execution of transformers on GPUs to enhance performance.
Includes the ability to enable a configuration option that supports Active/Standby deployments of the software stack.
Incorporates a feature to enable user authentication through SAML Single Sign-On.
Enhancements to the model loading logic have been made, including improvements for offline mode functionality.
ā
ā
ā





