chroma-core/chroma · error · ValueError

Expected each embedding in the embeddings to be a numpy arra

Error message

Expected each embedding in the embeddings to be a numpy array, got {list(set([type(e).__name__ for e in embeddings]))}

What it means

Every element of the embeddings list must be a numpy ndarray. validate_embeddings checks isinstance(e, np.ndarray) for each element and reports the distinct offending type names. Plain Python lists/tuples inside embeddings are rejected by this specific check (the public client normally converts, so this fires on direct calls to validation or custom API/EF implementations that skip conversion).

Source

Thrown at chromadb/api/types.py:1383

    if n_results <= 0:
        raise TypeError(
            f"Number of requested results {n_results}, cannot be negative, or zero."
        )
    return n_results


def validate_embeddings(embeddings: Embeddings) -> Embeddings:
    """Validates embeddings to ensure it is a list of numpy arrays of ints, or floats"""
    if not isinstance(embeddings, (list, np.ndarray)):
        raise ValueError(
            f"Expected embeddings to be a list, got {type(embeddings).__name__}"
        )
    if len(embeddings) == 0:
        raise ValueError(
            f"Expected embeddings to be a list with at least one item, got {len(embeddings)} embeddings"
        )
    if not all([isinstance(e, np.ndarray) for e in embeddings]):
        raise ValueError(
            "Expected each embedding in the embeddings to be a numpy array, got "
            f"{list(set([type(e).__name__ for e in embeddings]))}"
        )
    for i, embedding in enumerate(embeddings):
        if embedding.ndim == 0:
            raise ValueError(
                f"Expected a 1-dimensional array, got a 0-dimensional array {embedding}"
            )
        if embedding.size == 0:
            raise ValueError(
                f"Expected each embedding in the embeddings to be a 1-dimensional numpy array with at least 1 int/float value. Got a 1-dimensional numpy array with no values at pos {i}"
            )

        if embedding.dtype not in [
            np.float16,
            np.float32,
            np.float64,
            np.int32,

View on GitHub (pinned to aecdd12c8a)

Solutions

Convert each item: embeddings=[np.asarray(e, dtype=np.float32) for e in embeddings]
Convert wholesale: embeddings=list(np.array(embeddings, dtype=np.float32))
Fix a custom EmbeddingFunction to return np.ndarray instances

Example fix

# before
embeddings = [[0.1, 0.2, 0.3], [0.4, 0.5, 0.6]]  # plain lists

# after
import numpy as np
embeddings = [np.asarray(e, dtype=np.float32) for e in embeddings]

Defensive patterns

Strategy: type-guard

Validate before calling

import numpy as np

def normalize_embeddings(embeddings):
    if isinstance(embeddings, np.ndarray):
        embeddings = list(embeddings)
    return [e if isinstance(e, np.ndarray) else np.asarray(e, dtype=np.float32) for e in embeddings]

embeddings = normalize_embeddings(embeddings)  # every element now np.ndarray

Type guard

import numpy as np

def all_ndarray(embeddings) -> bool:
    return all(isinstance(e, np.ndarray) for e in embeddings)

Try / catch

try:
    validate_embeddings(embeddings)
except ValueError as e:
    if "to be a numpy array" in str(e):
        embeddings = [np.asarray(e, dtype=np.float32) for e in embeddings]
        validate_embeddings(embeddings)
    else:
        raise

Prevention

Always np.asarray(..., dtype=np.float32) embeddings loaded from JSON/CSV/REST
Custom embedding functions should return np.ndarray vectors, not lists
Run validation once in a shared preprocessing helper instead of trusting every call site

When it happens

Trigger: Calling validate_embeddings (directly, or via a custom SegmentAPI/server or EmbeddingFunction path that does not np.array-convert) with embeddings=[[0.1, 0.2], [0.3, 0.4]] — raw lists of floats — or mixing one np.ndarray with plain lists.

Common situations: Custom embedding functions returning list-of-lists; loading embeddings from JSON and passing them unconverted; older Chroma versions or self-built server layers that call validation on raw input.

Related errors

AI-assisted analysis of chroma-core/chroma@aecdd12c8a (2026-08-16). Data as JSON: /api/errors/bdd8bf3602ed3f17. Report an issue: GitHub.