ONNXEmbeddings Syntax Elements | Vantage BYOM - ONNXEmbeddings Syntax Elements - Teradata Vantage

Teradata Vantage™ - Bring Your Own Model User Guide
Deployment
VantageCloud
VantageCore
Edition
VMware
Enterprise
IntelliFlex
Lake
Product
Teradata Vantage
Release Number
7.0
Published
October 2025
ft:locale
en-US
ft:lastEdition
2025-11-07
dita:mapPath
fee1607120608274.ditamap
dita:ditavalPath
ayr1485454803741.ditaval
dita:id
fee1607120608274
Accumulate
Specify one or more column names to add to the output table in a comma-separated list. Use an asterisk (*) to specify all columns.
An accumulate_column cannot specify column Ids as integers, ranges, or allow the data type BLOB or CLOB.
ModelOutputTensor
Specify which tensor model to use for output.
EncodedMaxLength
Enter a value to set the encoded tokenizer maximum length for the output tokens.
Only available for models that do not have fixed dimensions.
Default: 512
ShowModelProperties
[Optional] Display the input and output tensor properties of the model as a varchar column.
When set to true, scoring is not run and only the current model properties are shown.
Default: False
OutputColumnPrefix
[Optional] Display the column prefix for each of the output columns when using FLOAT32 OutputFormat.
Default: emb_
OutputFormat
[Optional] Use the TYPE(INTEGER) or vector format for the model embeddings output.
Valid values are: VARBYTE, BLOB, FLOAT32, VARCHAR, and VECTOR.
The integer value for VARBYTE, BLOB, VARCHAR, and VECTOR are the number of bytes (or characters) for each type. The VECTOR number must be a multiple of 8. The value for FLOAT32 is the number of FLOAT32 output columns.
Default: VARBYTE(3072)
Default when used as OutputFormat ('VECTOR'): 32000
When VECTOR is used with a byte size parameter such as OutputFormat ('VECTOR(3072)'), it must be divisible by 8 since the vector data type is made of 8 byte values.
To display the VECTOR value in BTEQ, you must cast it to VARCHAR to display its binary value.
OverwriteCachedModel
[Optional] Valid values are: 'current_cached_model', '*', 'true', 't', 'yes', '1', 'false', 'f', 'no', 'n', or '0'. All of these values are equivalent as this argument applies only to the model specified in the model ON clause, which is assumed to be in the cache.
Important: Do not use the OverwriteCachedModel argument except when you are trying to replace a previously cached model. This applies to all models supported by BYOM. Using the argument in other cases, including in concurrent queries or multiple times within a short period of time, may lead to an OOM error from garbage collection not being fast enough.
If a model loaded into the memory of the node fits in the cache, it stays in the cache until being evicted to make space for another model that needs to be loaded or after 7 days if not used. Therefore, a model can remain in the cache after the query that loaded it completes. Other queries that use the same model can use it, saving the cost of reloading it into memory.
You want to overwrite a cached model only when it has been updated, to make sure that the Predict function uses the updated model instead of the cached model.
Default behavior: The function does not overwrite cached models.
UseCache
[Optional] When true, this statement loads or uses the model from the cache.
When false, the model is loaded into a temporary cache that is cleared once the query completes.
String enclosed in single quotes 'false' or 'true'.
Default: False
EnableMemoryCheck
[Optional] When true, enables the memory check for large models to verify there is enough native memory. This parameter only applies to ModelType DAI.
Only valid with ModelType ONNX.
Default: True