Clarification Needed: Embeddings Model Selection

[elina-ashi]

Hi,

In our chatbot bedrock knowledge base, we are using Titan Text Embeddings v2 as the embeddings model.

However, in the CloudFormation parameter, we selected amazon.nova-2-multimodal-embeddings-v1 as the EmbeddingsBedrockModelId.

So which model is actually being used as the embeddings model in this bot? When answering questions, which embedding model is being used? How does it work in this case?"
Could anyone please reply on this?

[advaitamazon]

Thank you for reporting this! Our team will take a look and get back shortly.

[aijunpeng]

These are two separate embeddings models that serve different purposes — they don't conflict with each other.

• EmbeddingsBedrockModelId (amazon.nova-2-multimodal-embeddings-v1) — Used by QnABot for semantic matching against your Q&A items stored in OpenSearch. This is the model used when answering questions that match items you've added in the Content Designer.
• Bedrock Knowledge Base embeddings model (Titan Text Embeddings v2) — Used internally by your Bedrock Knowledge Base to index and search its own data sources. QnABot calls the Knowledge Base as a fallback when no match is found in the QnABot question bank.

Both models are active but operate independently. QnABot first tries to match using amazon.nova-2-multimodal-embeddings-v1 against your Q&A items. If no match is found, it falls back to the Bedrock Knowledge Base (which uses Titan Text Embeddings v2 internally).

Hope this clarifies the issue. Please let us know if there are any more questions.