BerriAI/litellm · error · ValueError

Shape must be of length 2.

Error message

Shape must be of length 2.

What it means

Triton embedding splitter guard: split_embedding_by_dims received an embedding shape whose length is not 2 (expected [batch, dim]), so the response tensor cannot be split per dimension batch as the codepath assumes.

Source

Thrown at litellm/llms/triton/embedding/transformation.py:139

                continue
            try:
                prompt_tokens += token_counter(model=model, text=text)
            except Exception:
                prompt_tokens += len(text.split())

        return Usage(
            prompt_tokens=prompt_tokens,
            completion_tokens=0,
            total_tokens=prompt_tokens,
        )

    def get_error_class(self, error_message: str, status_code: int, headers: dict | httpx.Headers) -> BaseLLMException:
        return TritonError(message=error_message, status_code=status_code, headers=headers)

    @staticmethod
    def split_embedding_by_shape(data: list[float], shape: list[int]) -> list[list[float]]:
        if len(shape) != 2:
            raise ValueError("Shape must be of length 2.")
        embedding_size: Final = shape[1]
        return [data[i * embedding_size : (i + 1) * embedding_size] for i in range(shape[0])]

View on GitHub (pinned to 77b7c6c40c)

Solutions

1. Pass a shape list/tuple of exactly length 2 for the embedding request.

Example fix

shape=[batch_size, seq_len]

When it happens

Trigger: Triggered when the Triton embedding request shape does not have exactly length 2.

Common situations: See trigger scenarios.

AI-assisted analysis of BerriAI/litellm@77b7c6c40c (2026-08-18). Data as JSON: /api/errors/b070804f3766a335. Report an issue: GitHub.