MemPalace/mempalace · error · ValueError

embedding dimension must be positive

Error message

embedding dimension must be positive

What it means

Raised by MilvusCollection._ensure_remote_collection() when the dimension argument is <= 0. Before creating or validating the remote Milvus collection the backend sanity-checks the embedding dimension; zero or negative dims (which can slip through when a dim is computed from empty config or parsed as -1 'unknown') are rejected with plain ValueError. Note this is the collection-creation path; batch-level shape errors are caught earlier by _as_vector_array (error 34).

Source

Thrown at mempalace/backends/milvus.py:393

            return None
        fields = []
        if isinstance(info, dict):
            fields = info.get("fields") or (info.get("schema") or {}).get("fields") or []
        for field in fields:
            name = field.get("name") or field.get("field_name")
            if name != FIELD_VECTOR:
                continue
            params = field.get("params") or field.get("type_params") or {}
            dim = field.get("dim") or params.get("dim")
            try:
                return int(dim)
            except (TypeError, ValueError):
                return None
        return None

    def _ensure_remote_collection(self, dimension: int) -> None:
        if dimension <= 0:
            raise ValueError("embedding dimension must be positive")
        with self._lock:
            self._ensure_open()
            if self._known_dimension is not None:
                if self._known_dimension != dimension:
                    raise DimensionMismatchError(
                        f"milvus collection {self._collection_name!r} expects "
                        f"embedding dimension {self._known_dimension}, got {dimension}"
                    )
                return
            if not self._remote_exists():
                self._backend._create_remote_collection(
                    self._client,
                    self._remote_collection,
                    dimension,
                    consistency_level=self._config.consistency_level,
                )
                self._backend._load_remote_collection(self._client, self._remote_collection)
                self._known_dimension = dimension

View on GitHub (pinned to 06cb6987f0)

Solutions

1. Ensure the embedder is loaded and reports its true dimension (e.g. 384/768) before first add()
2. Validate config: dimension must be a positive int, not 0/-1/None
3. Lazy-create the collection only after the first real embedding is available

Example fix

# before
dim = embedder.dimension  # 0 because model not loaded
collection._ensure_remote_collection(dim)

# after
embedder.load()
dim = embedder.dimension  # e.g. 768
assert dim > 0, f"bad embedding dimension: {dim}"
collection._ensure_remote_collection(dim)

def valid_dimension(dim) -> bool:
    return isinstance(dim, int) and not isinstance(dim, bool) and dim > 0

try:
    collection.add(ids=ids, documents=docs, embeddings=embs)
except ValueError as e:
    if "dimension must be positive" in str(e):
        raise RuntimeError("embedder dimension not initialized — load the model before first insert") from e
    raise

When it happens

Trigger: Calling add()/query() where the embedder reports dimension 0 (uninitialized model config), or code passing dim=-1 as an 'unknown' sentinel into the collection bootstrap path.

Common situations: Embedder not yet loaded so its dim attribute is 0/None→0; misconfigured embedding model section; custom embedders whose dimension property is computed before initialization.

Related errors

• embedding must be a non-empty 1D vector
• Milvus filter field {name!r} is not a safe identifier
• $in requires a non-empty list for {field!r}
• $nin requires a non-empty list for {field!r}
• milvus batch cannot mix embedding dimensions {sorted(dims)}

AI-assisted analysis of MemPalace/mempalace@06cb6987f0 (2026-08-15). Data as JSON: /api/errors/172e57acc42d9e9b. Report an issue: GitHub.