Identifying name of the assistant that can be customized according to your preferences.
Detailed description of the purpose and capabilities of your assistant. This is a place to provide information about how the assistant enhances the end user experience.
Indicates whether the assistant is enabled or disabled for use.
An embedding model is a type of machine learning model designed to convert data—like words, phrases, or sentences—into dense numerical vectors, known as embeddings. These embeddings capture the semantic meaning of the input.
Specific parameters related to embeddings and related model characteristics:
Determines the embeddings model service provider used by your RAG assistant.
This is a mandatory parameter that can take any of the following values:
- azure: To use Azure OpenAI.
- cohere: To use cohere models.
- vertex_ai: To use Google Vertex AI, check valid models here, you can set for example embedding-001, text-embedding-004, text-embedding-005 embeddings models(1) of 768 Dimensions.
- awsbedrock: To use embeddings models from AWS Bedrock.
- openai: To use OpenAI.
Specific name of the model being used; supported values depend on the Provider Name selected.
For example, if the Provider Name parameter has the "openai" value, Model Name takes the value 'text-embedding-ada-002' for assistants created up to 4/11/2024 and 'text-embedding-3-small' onwards.
Note: If you need to use a model not detailed on the list, use the
overide mechanism.
API authentication key provided to access the language model service.
For example, if the Provider parameter is "azureopenai", you must specify "apiKey" with the authentication key.
It allows configuring the information processing approach for the embeddings model. It only applies for the following providers:
- CoHere (using awsprovider): search_document (default), search_query, classification, clustering.
- Google: RETRIEVAL_DOCUMENT (default for embeddings), RETRIEVAL_QUERY (default for querying), SEMANTIC_SIMILARITY, CLASSIFICATION, CLUSTERING, TASK_TYPE_UNSPECIFIED
- nvidia: passage, query; check here for more detail.
Use Globant Enterprise AI proxy to get embeddings, all requests will be logged (enabled by default).
Cache the embeddings for the current Organization/Project (disabled by default), needs to enable the useProxy property too.
Determines how many items are grouped in each batch when calling the embeddings model during the ingestion process.
By default, batchSize is set to 100 items. However, you may need to adjust this parameter based on the specific embeddings model you are using.
Defined the mode options are:
- empty (default) will use text embeddings.
- multimodal support for image embeddings (notice the model must support it).
The URL pointing to the language model service (advanced).
It allows configuring the index dimensionality in the vectorStore. Each model has a specific dimension, and the default value is 1536, associated with the OpenAI provider.
(1): Depending on the Provider Name and Model Name selected for your embeddings, you will need to check the Dimensions value. By default, it assumes 1536, which is valid for OpenAI text-embedding-ada-002, text-embedding-3-small and text-embedding-3-large (3072) shrunk to that value. Check the troubleshooting section for more detail.
Notes:
- Please contact the help desk if you want to use a Model Name with a Dimensions value other than 1536, as it must match the Vector Store Provider configuration in the Index section.
- If you change the Provider Name, Model Name, Dimensions, it is recommended to reingest all your data.