import numpy as np


def validate_vectors(values, dimension=None, normalize=False):
    try:
        array = np.asarray(values, dtype=np.float32)
    except (TypeError, ValueError) as exc:
        raise ValueError('Embedding response is not a rectangular numeric matrix.') from exc
    if array.ndim != 2 or array.shape[0] == 0 or array.shape[1] == 0:
        raise ValueError('Embeddings must be a nonempty matrix.')
    if dimension and array.shape[1] != dimension:
        raise ValueError(f'Embedding dimension {array.shape[1]} does not match {dimension}.')
    norms = np.linalg.norm(array, axis=1)
    if not np.isfinite(array).all() or not np.isfinite(norms).all() or np.any(norms == 0):
        raise ValueError('Embeddings contain non-finite values or zero vectors.')
    if normalize:
        array = array / norms[:, None]
    return array.tolist()