sgl-project/sglang · error · ValueError

Required: indexer, forward_batch, x, q_lora, positions

Error message

Required: indexer, forward_batch, x, q_lora, positions

What it means

DeepSeek DSA sparse retrieval requires five kwargs: indexer, forward_batch, x, q_lora, and positions. Any None among them raises this ValueError, because the DeepSeek indexer call needs the hidden states (x), low-rank query projection (q_lora), and position ids to compute retrieval scores.

Source

Thrown at python/sglang/srt/mem_cache/sparsity/algorithms/deepseek_dsa.py:39

    def retrieve_topk(
        self,
        queries: torch.Tensor,
        layer_id: int,
        req_pool_indices: torch.Tensor,
        sparse_mask: torch.Tensor,
        attn_metadata: Optional[Any],
        **kwargs,
    ) -> tuple:
        indexer, forward_batch, x, q_lora, positions = (
            kwargs.get("indexer"),
            kwargs.get("forward_batch"),
            kwargs.get("x"),
            kwargs.get("q_lora"),
            kwargs.get("positions"),
        )

        if any(v is None for v in [indexer, x, q_lora, positions, forward_batch]):
            raise ValueError("Required: indexer, forward_batch, x, q_lora, positions")

        return (
            indexer(
                x=x,
                q_lora=q_lora,
                positions=positions,
                forward_batch=forward_batch,
                layer_id=layer_id,
            ),
            None,
        )

    def initialize_representation_pool(
        self,
        start_layer: int,
        end_layer: int,
        token_to_kv_pool,
        req_to_token_pool,

View on GitHub (pinned to 0132848349)

Solutions

1. Supply all five kwargs from the model's forward: retrieve_topk(queries, indexer=self.indexer, x=hidden_states, q_lora=q_lora, positions=positions, forward_batch=forward_batch)
2. Ensure the model actually has the DSA indexer components (DeepSeek V3.2 DSA architecture); otherwise pick a different sparse algorithm
3. Check integration examples for the DeepSeek DSA path to copy the exact argument wiring

Example fix

# before
out = dsa.retrieve_topk(queries, forward_batch=forward_batch)
# after
out = dsa.retrieve_topk(queries, forward_batch=forward_batch,
                        indexer=self.indexer, x=hidden_states,
                        q_lora=q_lora, positions=positions)

required = {k: kwargs.get(k) for k in ("indexer", "forward_batch", "x", "q_lora", "positions")}
missing = [k for k, v in required.items() if v is None]
assert not missing, f"missing DSA kwargs: {missing}"

When it happens

Trigger: Calling retrieve_topk on the DeepSeek DSA algorithm without supplying indexer (the DeepSeek indexer module), x (hidden states), q_lora (LoRA-projected queries), or positions.

Common situations: Wiring the DSA algorithm into a custom model forward where the indexer module or q_lora projection is not plumbed through; using the generic sparse-algorithm API without model-specific arguments for DeepSeek V3.2-style DSA.

Understand the failure class

Background: "missing required argument" and "the following required arguments were not provided": what required-argument errors mean and how to fix them — this error's family across 20 libraries.

Related errors

• forward_batch with seq_lens is required for TopK retrieval
• bad compress_ratio {compress_ratio}
• DSA indexer weights_proj LoRA is incompatible with piecewise
• DSA indexer only supports CUDA, HIP, and NPU
• Quest query hidden size {hidden} not divisible by head_dim {

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/ba214eb8b466e2f1. Report an issue: GitHub.