AWS Bedrock Embeddings

​

Node Details

• Name: AWSBedrockEmbeddings
• Type: AWSBedrockEmbeddings
• Version: 5.0
• Category: Embeddings

​

Base Classes

• AWSBedrockEmbeddings
• BedrockEmbeddings (from LangChain)

​

Parameters

​

Credential (Optional)

• Type: AWS API Credential

​

Inputs

1. Region
   • Type: Async Options (listRegions)
   • Default: “us-east-1”
   • Description: AWS region for the Bedrock service
2. Model Name
   • Type: Async Options (listModels)
   • Default: “amazon.titan-embed-text-v1”
   • Description: The embedding model to use
3. Custom Model Name
   • Type: String
   • Optional: Yes
   • Description: If provided, overrides the selected Model Name
4. Cohere Input Type
   • Type: Options
   • Optional: Yes
   • Description: Specifies the type of input for Cohere models (v3+)
   • Options:
     • search_document
     • search_query
     • classification
     • clustering
5. Batch Size
   • Type: Number
   • Optional: Yes
   • Default: 50
   • Description: Document batch size for Titan model API calls
6. Max AWS API retries
   • Type: Number
   • Optional: Yes
   • Default: 5
   • Description: Maximum number of API call retries for Titan model

​

Functionality

1. Model Initialization:
   • Sets up the BedrockEmbeddings model with provided parameters
   • Configures AWS credentials if provided
2. Embedding Generation:
   • For single queries: Uses embedQuery method
   • For multiple documents: Uses embedDocuments method
   • Handles different processing for Titan and Cohere models
3. Batch Processing:
   • Implements a batch processing system for Titan models to manage API rate limits
   • Includes retry logic with exponential backoff for handling throttling exceptions
4. Error Handling:
   • Provides specific error messages for invalid responses or exceeded retry limits

​

Use Cases

1. Semantic search in vector databases
2. Text classification tasks
3. Document clustering
4. Enhancing natural language processing pipelines

​

Notes

• The node dynamically loads available models and regions
• It supports both Amazon Titan and Cohere embedding models
• Special handling is implemented for Cohere models, requiring input type specification
• Batch processing and retry logic are implemented to handle API rate limits efficiently