run-llama/llama_index · error · ValueError

Unknown query mode: {query_mode}

Error message

Unknown query mode: {query_mode}

What it means

get_top_similar_embeddings_by_query only supports three classifier modes: SVM, LINEAR_REGRESSION, and LOGISTIC_REGRESSION. Passing any other VectorStoreQueryMode (DEFAULT, MMR, HYBRID, etc.) falls through to the else and raises, because the function is a specialized embedding-ranking helper, not a general query dispatcher.

Source

Thrown at llama-index-core/llama_index/core/indices/query/embedding_utils.py:86

    embeddings_np = np.array(embeddings)
    # create dataset
    dataset_len = len(embeddings) + 1
    dataset = np.concatenate([query_embedding_np[None, ...], embeddings_np])
    y = np.zeros(dataset_len)
    y[0] = 1

    if query_mode == VectorStoreQueryMode.SVM:
        # train our SVM
        # TODO: make params configurable
        clf = svm.LinearSVC(
            class_weight="balanced", verbose=False, max_iter=10000, tol=1e-6, C=0.1
        )
    elif query_mode == VectorStoreQueryMode.LINEAR_REGRESSION:
        clf = linear_model.LinearRegression()
    elif query_mode == VectorStoreQueryMode.LOGISTIC_REGRESSION:
        clf = linear_model.LogisticRegression(class_weight="balanced")
    else:
        raise ValueError(f"Unknown query mode: {query_mode}")

    clf.fit(dataset, y)  # train

    # infer on whatever data you wish, e.g. the original data
    similarities = clf.decision_function(dataset[1:])
    sorted_ix = np.argsort(-similarities)
    top_sorted_ix = sorted_ix[:similarity_top_k]

    result_similarities = similarities[top_sorted_ix]
    result_ids = [embedding_ids[ix] for ix in top_sorted_ix]

    return result_similarities, result_ids


def get_top_k_mmr_embeddings(
    query_embedding: List[float],
    embeddings: List[List[float]],
    similarity_fn: Optional[Callable[..., float]] = None,

View on GitHub (pinned to afd0fef371)

Solutions

1. Use one of VectorStoreQueryMode.SVM, .LINEAR_REGRESSION, .LOGISTIC_REGRESSION when this code path is hit
2. For nearest-neighbour semantics use VectorStoreQueryMode.DEFAULT (handled by the vector store, not this helper)
3. Validate query_mode against the supported set before constructing the retriever

Example fix

# before
retriever = index.as_retriever(vector_store_query_mode=VectorStoreQueryMode.MMR)
# if this path is entered: ValueError

# after
retriever = index.as_retriever(vector_store_query_mode=VectorStoreQueryMode.SVM)  # or DEFAULT

SUPPORTED = {VectorStoreQueryMode.SVM, VectorStoreQueryMode.LINEAR_REGRESSION, VectorStoreQueryMode.LOGISTIC_REGRESSION}
if query_mode in SUPPORTED and not has_sklearn():
    query_mode = VectorStoreQueryMode.DEFAULT
retriever = index.as_retriever(vector_store_query_mode=query_mode)

def is_classifier_mode(mode: VectorStoreQueryMode) -> bool:
    return mode in {
        VectorStoreQueryMode.SVM,
        VectorStoreQueryMode.LINEAR_REGRESSION,
        VectorStoreQueryMode.LOGISTIC_REGRESSION,
    }

try:
    sims, ids = get_top_similar_embeddings_by_query(qe, embeddings, query_mode=mode)
except ValueError as e:
    if 'Unknown query mode' in str(e):
        raise ValueError(f'{mode} not supported here; use SVM/LINEAR_REGRESSION/LOGISTIC_REGRESSION or DEFAULT') from e
    raise

When it happens

Trigger: Setting vector_store_query_mode to something other than the three supported values on a retriever whose code path routes into this helper — e.g. SVM-mode embeddings but the mode string was overwritten later, or passing an unvalidated string/int as query_mode.

Common situations: Copy-pasting retriever configs between engines where the same enum value routes to different internals; building a mode selector UI that forwards arbitrary enum values; refactoring from QueryMode (old enum) to VectorStoreQueryMode and passing a stale value.

Related errors

• Max iterations of {max_iterations} reached! Either something
• CitableBlock content must contain exactly one block when pro
• Eval mode {eval_mode} not supported.
• Cannot specify both system_prompt and prefix_messages
• Cannot specify both system_prompt and prefix_messages

AI-assisted analysis of run-llama/llama_index@afd0fef371 (2026-08-15). Data as JSON: /api/errors/dffc5eee36b2d862. Report an issue: GitHub.