microsoft/semantic-kernel · error · ServiceInvalidResponseError

The response from Cohere model does not contain embeddings.

Error message

The response from Cohere model does not contain embeddings.

What it means

Raised by the Cohere (Bedrock) text-embedding response parser when the response dict has no 'embeddings' key, that key is not a list, or the list is empty. The parser then dereferences response['embeddings'][0], so it defends all three conditions up front.

Source

Thrown at python/semantic_kernel/connectors/ai/bedrock/services/model_provider/bedrock_cohere.py:96

# endregion

# region Text Embedding


def get_text_embedding_request_body(text: str, settings: BedrockEmbeddingPromptExecutionSettings) -> Any:
    """Get the request body for text embedding for Cohere Command models."""
    return remove_none_recursively({
        "texts": [text],
        "input_type": settings.extension_data.get("input_type", "search_document"),
        "truncate": settings.extension_data.get("truncate", None),
        "embedding_types": settings.extension_data.get("embedding_types", None),
    })


def parse_text_embedding_response(response: dict[str, Any]) -> list[float]:
    """Parse the response from text embedding for Cohere Command models."""
    if "embeddings" not in response or not isinstance(response["embeddings"], list) or len(response["embeddings"]) == 0:
        raise ServiceInvalidResponseError("The response from Cohere model does not contain embeddings.")

    return response.get("embeddings")[0]  # type: ignore


# endregion

View on GitHub (pinned to c028a0c7dc)

Solutions

1. Confirm the model_id is a Cohere embedding model such as cohere.embed-english-v3 or cohere.embed-multilingual-v3, not a Command chat model.
2. Check the raw response body; if Cohere returns a validation error (e.g. invalid input_type/truncate), fix the extension_data settings accordingly.
3. Ensure the text passed to embed is non-empty so remove_none_recursively does not strip the 'texts' field.
4. Upgrade semantic-kernel and boto3 so the Cohere response schema the parser reads matches what Bedrock returns.

Example fix

// before
settings.extension_data["input_type"] = ""
// after
settings.extension_data["input_type"] = "search_document"

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

from typing import Any

def is_cohere_embedding_response(resp: Any) -> bool:
    return isinstance(resp, dict) and isinstance(resp.get("embeddings"), list) and bool(resp["embeddings"])

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("Cohere embedding call failed; check input_type/truncate extension data")
    raise

When it happens

Trigger: Called from BedrockTextEmbeddingService -> parse_text_embedding_response for a Cohere embedding model (e.g. cohere.embed-english-v3). Fires when the Cohere payload omits 'embeddings', returns it as a non-list, or returns an empty list (Cohere rejected the input text, or returned an error/status object instead).

Common situations: Using a Cohere chat model ID instead of a Cohere embedding model ID; Cohere rejecting input because input_type extension data is invalid; Bedrock throttling returning an error body; the 'texts' array in the request was empty after remove_none_recursively; version skew between Cohere embedding response schema and the parser.

Related errors

• The response from Amazon Titan model does not contain embedd
• Unsupported Cohere model: {modelId}
• Unsupported service type
• Response is null
• An error occurred while initializing the {nameof(IEmbeddingG

AI-assisted analysis of microsoft/semantic-kernel@c028a0c7dc (2026-08-13). Data as JSON: /api/errors/c8325d2558762e8d. Report an issue: GitHub.