sgl-project/sglang · error · ValueError

Invalid combination of query/items types for score_request.

Error message

Invalid combination of query/items types for score_request.

What it means

score_request dispatches on the shapes of query and items: it supports str/str, str/list, list/str, and list/list (plus embed-override variants). Any other type combination (e.g. None, int, nested lists of unequal depth, numpy arrays, dicts) falls through to this catch-all ValueError.

Source

Thrown at python/sglang/srt/managers/tokenizer_manager_score_mixin.py:579

                    item_embed_overrides,
                )
            )
        elif has_embeds:
            # Text inputs with embed overrides — need to tokenize first to resolve positions
            query_ids, items_ids = self._batch_tokenize_query_and_items(query, items)
            _, input_ids, positional_embed_overrides, delimiter_indices = (
                self._build_token_id_inputs(
                    query_ids,
                    items_ids,
                    item_first,
                    use_multi_item_scoring,
                    embed_override_token_id,
                    query_embed_overrides,
                    item_embed_overrides,
                )
            )
        else:
            raise ValueError(
                "Invalid combination of query/items types for score_request."
            )

        if return_pooled_hidden_states:
            if is_generation:
                raise ValueError(
                    "return_pooled_hidden_states is not supported for CausalLM models. "
                    "It requires a model with a task-specific head "
                    "(e.g. SequenceClassification or RewardModel)."
                )
            model_config = self.model_config
            if model_config is not None:
                archs = getattr(model_config.hf_config, "architectures", []) or []
                if is_cross_encoding_pooler_model(archs):
                    raise ValueError(
                        f"return_pooled_hidden_states is not supported for "
                        f"{archs[0]}. This model uses CrossEncodingPooler which "
                        f"does not expose pre-head hidden states."

View on GitHub (pinned to 0132848349)

Solutions

1. Normalize query/items to plain Python str or list[str] (call .tolist() on numpy arrays)
2. Ensure outer list lengths match when both are lists
3. Check for None/empty inputs before calling score and skip or default them

Example fix

# before
await engine.async_score(np.array(texts), [item])
# after
await engine.async_score(list(np.array(texts)), [item])

assert isinstance(query, (str, list)) and isinstance(items, (str, list))
assert all(isinstance(x, str) for x in ([query] if isinstance(query, str) else query))

def valid_score_args(query, items) -> bool:
    ok = lambda v: isinstance(v, str) or (isinstance(v, list) and v and all(isinstance(x, str) for x in v))
    if isinstance(query, list) and isinstance(items, list) and isinstance(items[0], list):
        return ok(query) and len(query) == len(items) and all(ok(i) for i in items)
    return ok(query) and ok(items)

When it happens

Trigger: Calling score with query=None, items being a generator or numpy array, mismatched nesting (list vs list-of-lists with unequal lengths in cross-scoring mode), or passing embeddings in a shape the dispatcher does not recognize.

Common situations: Dynamically built request payloads where query or items can be None; converting arrays to numpy; passing a list of queries with a list of list-of-items of different length.

Understand the failure class

Background: Invalid argument type errors: "must be of type string", "expected X, got Y", and ERR_INVALID_ARG_TYPE explained — this error's family across 15 libraries.

Related errors

• actions must be a list[list[str]]
• Inkling reasoning_effort must not be a boolean
• cuda-graph mode cu_seqlens should be a list
• flashinfer_cudnn expects packed indptrs as a torch.Tensor
• Invalid prompts type for score_prompts.

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/69856a8380bc075c. Report an issue: GitHub.