sgl-project/sglang · error · ValueError

Either input_ids or inputs_embeds must be provided.

Error message

Either input_ids or inputs_embeds must be provided.

What it means

CLIP text/position embedding forward requires exactly one of input_ids or inputs_embeds to compute the sequence length. Passing None for both is invalid.

Source

Thrown at python/sglang/srt/models/clip.py:123

        # position_ids (1, len position emb) is contiguous in memory and exported when serialized
        self.register_buffer(
            "position_ids",
            torch.arange(config.max_position_embeddings).expand((1, -1)),
            persistent=False,
        )

    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
    ) -> torch.Tensor:
        if input_ids is not None:
            seq_length = input_ids.shape[-1]
        elif inputs_embeds is not None:
            seq_length = inputs_embeds.shape[-2]
        else:
            raise ValueError("Either input_ids or inputs_embeds must be provided.")

        max_positions = self.position_embedding.weight.shape[0]
        if seq_length > max_positions:
            raise ValueError(
                f"Sequence length {seq_length} exceeds the maximum {max_positions}."
            )

        if position_ids is None:
            position_ids = self.position_ids[:, :seq_length]

        if inputs_embeds is None:
            inputs_embeds = self.token_embedding(input_ids)

        position_embeddings = self.position_embedding(position_ids)
        embeddings = inputs_embeds + position_embeddings

        return embeddings

View on GitHub (pinned to 0132848349)

Solutions

1. Pass input_ids (token ids) or precomputed inputs_embeds, never neither
2. Fix the upstream caller that failed to supply one of the two tensors

Example fix

# before
emb.forward(None, None)
# after
emb.forward(input_ids=input_ids)

assert input_ids is not None or inputs_embeds is not None

def has_embed_input(ids, embeds) -> bool:
    return ids is not None or embeds is not None

try: emb.forward(input_ids, inputs_embeds)
except ValueError as e: raise UserInputError(str(e))

When it happens

Trigger: Calling CLIPEmbeddings.forward(input_ids=None, inputs_embeds=None).

Common situations: Wrapping CLIP in a custom pipeline where the embedder is skipped and ids are dropped; a bug upstream leaves both None.

Related errors

• You have to specify input_ids
• Sequence length {seq_length} exceeds the maximum {max_positi
• unsupported input for causal Conv3D cat/pad CUDA
• unsupported input for usp_merge_heads CUDA
• unsupported input for modulate_scale_shift CUDA

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/c29e28e310e09167. Report an issue: GitHub.