RyanCodrai/turbovec · error · ValueError

Document {doc.id!r} has no embedding. TurboQuantDocumentStor

Error message

Document {doc.id!r} has no embedding. TurboQuantDocumentStore only stores documents with precomputed embeddings — run an embedder component before writing.

What it means

Raised in _write_documents_locked when a Document has embedding=None. TurboQuantDocumentStore stores only precomputed embeddings — there is no embedder inside the store — so writing an un-embedded document would create an unsearchable entry. It fires in both the FAIL path and the SKIP/OVERWRITE validation pass.

Source

Thrown at turbovec-python/python/turbovec/haystack.py:266

            # stays persisted — a partial write. Mirror that observable
            # state exactly: validate and commit per document, raising on
            # the first collision. A repeated id within a single call
            # collides with its already-committed first instance, the same
            # way a cross-call repeat would. Each individual commit is
            # still all-or-nothing — validation precedes any mutation, so
            # a failing document mid-batch never leaves the index and the
            # id maps inconsistent (#89/#139 apply per document).
            written = 0
            for doc in documents:
                if doc.id in self._str_to_u64:
                    # Checked before embedding validation: the reference
                    # raises DuplicateDocumentError for a colliding id
                    # regardless of the document's other fields.
                    raise DuplicateDocumentError(
                        f"ID '{doc.id}' already exists in the document store."
                    )
                if doc.embedding is None:
                    raise ValueError(
                        f"Document {doc.id!r} has no embedding. "
                        "TurboQuantDocumentStore only stores documents with precomputed "
                        "embeddings — run an embedder component before writing."
                    )
                self._commit_batch([doc])
                written += 1
            return written

        # SKIP / OVERWRITE: first pass validates and resolves duplicates
        # against the batch-so-far as well as the existing store:
        # InMemoryDocumentStore writes into its dict as it iterates, so a
        # repeated id *within a single call* is resolved the same way a
        # cross-call repeat would be. Without tracking the batch, every
        # duplicate row still gets its own vector while _str_to_u64 keeps
        # only the last handle, orphaning the earlier vectors.
        to_write: List[Document] = []
        batch_pos: Dict[str, int] = {}  # doc.id -> index into to_write
        to_remove: List[str] = []  # existing ids to drop, deferred past add

View on GitHub (pinned to ccab9f325e)

Solutions

1. Run a haystack Embedder component (e.g. SentenceTransformersDocumentEmbedder) on the documents before write_documents.
2. Check the pipeline wiring: the embedder step may have been skipped or produced None for some documents.
3. Catch the ValueError in ingestion code to identify which document ids lack embeddings.

When it happens

Trigger: Thrown at turbovec-python/python/turbovec/haystack.py:266 when the library encounters an invalid state.

Common situations: See trigger scenarios.

AI-assisted analysis of RyanCodrai/turbovec@ccab9f325e (2026-09-06). Data as JSON: /api/errors/68f82fd6c96a4e2d. Report an issue: GitHub.