These enhancements were added to VantageCloud Lake for August 2025.
Bring Your Own Large Language Model and Deep Learning Model Workloads (Google Cloud)
BYO-LLM extends the power of the Open Analytics Framework to allow you to work with open-source LLMs from providers like Hugging Face. Using teradataml, you bring the models into VantageCloud Lake to run on the new GPU accelerated analytic GPU compute clusters configured with NVIDIA GPU accelerated containers designed to run LLM and DL workloads. See Bring Your Own LLM and DL Workloads.
Bring Your Own Model
BYOM 6.02 supports the following functionality updates:
- ONNXEmbeddings and ONNXSeq2Seq functions, including additional items for language models
- H2OPredict support for NLP models
- DataRobot scoring library version 2.2.4
- JVM memory size and JVMMStatus function changes
See Bring Your Own Model.
Teradata Package for Generative AI Updates
Teradata Package for Generative AI has been updated to support Vector Store and expanded Text Analytics functions.
- Teradata Enterprise Vector Store is engineered to efficiently store, index, and search high-dimensional vector embeddings. The teradatagenai library offers Python APIs that simplify access to and management of the vector store so you can build your own natural language applications using Vantage as the foundational compute/storage engine. Operations include providing methods to manage the vector store and facilitate the creation and management of patterns to create metadata-based vector stores.
- Text Analytics functions enables inference using large language model (LLM) through:
- AWS Bedrock
- Azure OpenAI
- Google Gemini
- VantageCloud Lake BYO-LLM
- Teradata AI Factory integrated with NVIDIA AI Enterprise
- In-database Text Analytics Functions
Function BYO-LLM (Hugging Face) AWS Bedrock Azure OpenAI Google Gemini NVIDIA NIM BYOM (ONNX) analyze_sentiment() classify() detect_language() embeddings() extract_key_phrases() mask_pii() recognize_entities() recognize_pii_entities() summarize() translate() sentence_similarity() ask()
Restrict Public Internet Access to the Console using PrivateLink (AWS)
You can create a private URL using PrivateLink and disable public internet access to the console. See Organization Admin: Restrict Public Internet Access to the Console.
New Authorization Type for Identity Provider
When selecting an identity provider, you now have the option of selecting a Private key as the authorization type for OpenID Connect.
Open Analytics Framework (OAF) Auth Sign-In
A customer organization admin can disable or enable OAuth2.0 Device Authorization Grant flow based authentication to access Open Analytics REST APIs from their VantageCloud Lake Console. This toggle functions at an organization level impacts Open Analytics REST API access to all VantageCloud Lake environments provisioned for the customer organization.
- Generate JSON Web Tokens (JWTs) using OAuth 2.0 Device Authorization Grant flow and pass along JWTs as Bearer tokens.
- Use Teradata Package for Python to access the Open Analytics APIs.
When disabled, customers can use Personal Access Token based authentication.
See Open Analytics Framework Authorization Sign In.
Autoschema Generic data type for NOS
As part of Autoschema Generic data type for NOS feature, three new Named Value Pairs (NVP) are added - GENERIC_DATATYPE, STRING_SIZE and BINARY_SIZE to use when creating foreign table.
See Example: GENERIC_DATATYPE, STRING_SIZE, and BINARY_Size for more information.
New Workload Management Ruleset API: TDWMSetRulesetAttr
A new workload managements ruleset API is added which allows updating attributes of a specific ruleset.
See TDWMSetRulesetAttr for more.
Identify Target Tables in DBQL Logging
The DBQL OBJECTS logging option now supports identifying base tables that are the target of write operations from SQL statements or load utilities. This feature helps you analyze data lineage flows more effectively within your system.
To facilitate this enhancement, a new system view column, TargetIndicator, has been added.
For more information, see dbqlobjV .
Extending Open Analytics to access objects from external data sources
The method to provide input data in on-prem object storage or OTF tables to run Python or R analytics using the Apply table operator has been extended to support files and objects in external cloud object storage. The use of the proxy_client Python package enables seamless access to and analysis of unstructured and structured data within Teradata at scale.See Extending Open Analytics to Access Objects from External Data Sources.