What causes "No embeddings to aggregate" in run-llama/llama_index?

Skip aggregation when the embedding list is empty: `if embeddings: ... else: continue`.

run-llama/llama_index · error · ValueError

No embeddings to aggregate

Error message

No embeddings to aggregate

What it means

Raised by mean_agg when it is called with an empty list of embeddings. Mean aggregation over zero vectors is undefined, so the function fails fast instead of returning NaN or a zero vector. This helper backs similarity averaging in BaseEmbedding.

Source

Thrown at llama-index-core/llama_index/core/base/embeddings/base.py:50

    EmbeddingStartEvent,
)
import llama_index.core.instrumentation as instrument

dispatcher = instrument.get_dispatcher(__name__)


class SimilarityMode(str, Enum):
    """Modes for similarity/distance."""

    DEFAULT = "cosine"
    DOT_PRODUCT = "dot_product"
    EUCLIDEAN = "euclidean"


def mean_agg(embeddings: List[Embedding]) -> Embedding:
    """Mean aggregation for embeddings."""
    if not embeddings:
        raise ValueError("No embeddings to aggregate")

    return np.array(embeddings).mean(axis=0).tolist()


def similarity(
    embedding1: Embedding,
    embedding2: Embedding,
    mode: SimilarityMode = SimilarityMode.DEFAULT,
) -> float:
    """Get embedding similarity."""
    if mode == SimilarityMode.EUCLIDEAN:
        # Using -euclidean distance as similarity to achieve same ranking order
        return -float(np.linalg.norm(np.array(embedding1) - np.array(embedding2)))
    elif mode == SimilarityMode.DOT_PRODUCT:
        return np.dot(embedding1, embedding2)
    else:
        product = np.dot(embedding1, embedding2)
        norm = np.linalg.norm(embedding1) * np.linalg.norm(embedding2)

View on GitHub (pinned to afd0fef371)

Solutions

Skip aggregation when the embedding list is empty: `if embeddings: ... else: continue`.
Fix upstream node construction so empty-text nodes are dropped (e.g. filter nodes with not node.get_content().strip()).
If an empty aggregate must exist for your schema, decide an explicit policy (skip node, or raise a clearer domain error) rather than relying on this generic one.

Example fix

# before
emb = mean_agg(get_text_embeddings_for_node_batch(node_batch))

# after
embs = get_text_embeddings_for_node_batch(node_batch)
if embs:
    emb = mean_agg(embs)
else:
    node_batch = [n for n in node_batch if n.get_content().strip()]

Defensive patterns

Strategy: validation

Validate before calling

texts = [t for t in texts if t and t.strip()]
if not texts:
    return  # nothing to aggregate

Type guard

def has_embeddings(embs: list) -> bool:
    return len(embs) > 0

Try / catch

try:
    agg = mean_agg(embs)
except ValueError:
    logger.warning("empty embedding batch, skipping node")

Prevention

Filter empty/whitespace text before embedding.
Drop empty nodes at node-parser time rather than at aggregation time.

When it happens

Trigger: Aggregating embeddings for a node whose text is empty, or calling get_text_embedding over a chunk list that filtered down to zero items before aggregation; passing [] to mean_agg directly.

Common situations: Indexing documents that produce empty text nodes (blank pages, whitespace-only extraction); pipelines that batch texts and pass an exhausted/filtered batch; splitting logic yielding zero chunks for a document.

Related errors

AI-assisted analysis of run-llama/llama_index@afd0fef371 (2026-08-15). Data as JSON: /api/errors/8377c8622cd70ce7. Report an issue: GitHub.