microsoft/semantic-kernel · error · ServiceInvalidResponseError

The response from Amazon Titan model does not contain embedd

Error message

The response from Amazon Titan model does not contain embeddings.

What it means

Raised by the Amazon Titan (Bedrock) text-embedding response parser when the parsed response dict has no 'embedding' key, or that key is not a list. Semantic Kernel expects the Titan embedding endpoint to return a JSON body containing an 'embedding' array of floats; any other shape is treated as an upstream contract violation.

Source

Thrown at python/semantic_kernel/connectors/ai/bedrock/services/model_provider/bedrock_amazon_titan.py:108


def get_text_embedding_request_body(text: str, settings: BedrockEmbeddingPromptExecutionSettings) -> dict[str, Any]:
    """Get the request body for text embedding for Amazon Titan models."""
    return remove_none_recursively({
        "inputText": text,
        # Extension data: https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-titan-embed-text.html
        "dimensions": settings.extension_data.get("dimensions", None),
        "normalize": settings.extension_data.get("normalize", None),
        "embeddingTypes": settings.extension_data.get("embeddingTypes", None),
        # Extension data: https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-titan-embed-mm.html
        "embeddingConfig": settings.extension_data.get("embeddingConfig", None),
    })


def parse_text_embedding_response(response: dict[str, Any]) -> list[float]:
    """Parse the response from text embedding for Amazon Titan models."""
    if "embedding" not in response or not isinstance(response["embedding"], list):
        raise ServiceInvalidResponseError("The response from Amazon Titan model does not contain embeddings.")

    return response.get("embedding")  # type: ignore


# endregion

View on GitHub (pinned to c028a0c7dc)

Solutions

Verify the model_id is an Amazon Titan embedding model such as amazon.titan-embed-text-v2:0, not a text-generation Titan model.
Inspect the raw Bedrock response (enable SDK debug logging) to confirm it actually contains an 'embedding' list; if the body is an error, fix the underlying IAM/quota/access issue.
Upgrade semantic-kernel and boto3 to compatible versions so the Bedrock response schema matches what the parser expects.
If you configured 'embeddingTypes' extension data, ensure the value is a type Titan returns in the flat 'embedding' field, not a nested batch structure.

Example fix

// before
service = BedrockTextEmbeddingService(model_id="amazon.titan-text-premier-v1:0")
// after
service = BedrockTextEmbeddingService(model_id="amazon.titan-embed-text-v2:0")

Defensive patterns

Strategy: validation

Validate before calling

def is_valid_titan_embedding_response(response: dict) -> bool:
    return isinstance(response, dict) and isinstance(response.get("embedding"), list) and len(response["embedding"]) > 0

Type guard

from typing import Any

def is_titan_embedding_response(resp: Any) -> bool:
    return isinstance(resp, dict) and isinstance(resp.get("embedding"), list)

Try / catch

from semantic_kernel.exceptions.service_exceptions import ServiceInvalidResponseError

try:
    embeddings = await service.generate_embeddings([text])
except ServiceInvalidResponseError as e:
    if "does not contain embeddings" in str(e):
        logger.error("Titan returned no embeddings; raw response needs inspection: %s", e)
    raise

Prevention

Always use a Titan embedding model ID (amazon.titan-embed-text-v2:0).
Enable boto3 debug logging to inspect the raw Bedrock response when first integrating.
Wrap the embedding call in a ServiceInvalidResponseError catch and surface the raw response for diagnostics.

When it happens

Trigger: Called from BedrockTextEmbeddingService -> parse_text_embedding_response after Amazon Titan embedding invoke. Fires when response.get('embedding') is missing, is None, or is not a list (e.g. the model returned an error object, a different field name like 'vector', or an empty body).

Common situations: Using a Titan model ID that is not an embedding model (e.g. amazon.titan-text-premier) for embeddings; Bedrock returning an access/throttling error payload that shadows the embedding field; mismatch between Bedrock SDK response version and SK parser; model returns 'embeddingTypes' batch format instead of flat 'embedding' list.

Related errors

AI-assisted analysis of microsoft/semantic-kernel@c028a0c7dc (2026-08-13). Data as JSON: /api/errors/a7d6bfe9c7f4dcf3. Report an issue: GitHub.