BerriAI/litellm · error · HuggingFaceError

sentence transformers requires 2+ sentences

Error message

sentence transformers requires 2+ sentences

What it means

Raised when the model name contains 'sentence-transformers' and the input list is empty. Note the message says '2+ sentences' but the code only checks len(input) == 0, so in practice it fires only for an empty input list; a single sentence is accepted (and silently sends an empty 'sentences' array to HF).

Source

Thrown at litellm/llms/huggingface/embedding/handler.py:148

            else:
                data[k] = v

        return data

    def _transform_input(
        self,
        input: list,
        model: str,
        call_type: Literal["sync", "async"],
        optional_params: dict,
        embed_url: str,
    ) -> dict:
        data: dict = {}

        ## TRANSFORMATION ##
        if "sentence-transformers" in model:
            if len(input) == 0:
                raise HuggingFaceError(
                    status_code=400,
                    message="sentence transformers requires 2+ sentences",
                )
            data = {"inputs": {"source_sentence": input[0], "sentences": input[1:]}}
        else:
            data = {"inputs": input}

            task_type: Final = optional_params.pop("input_type", None)

            if call_type == "sync":
                hf_task: Final = get_hf_task_embedding_for_model(model=model, task_type=task_type, api_base=HF_HUB_URL)
            elif call_type == "async":
                return self._async_transform_input(model=model, task_type=task_type, embed_url=embed_url, input=input)

            data = self._transform_input_on_pipeline_tag(input=input, pipeline_tag=hf_task)

        if len(optional_params.keys()) > 0:
            data = self._process_optional_params(data=data, optional_params=optional_params)

View on GitHub (pinned to 6c2dcb801b)

Solutions

1. Guard your pipeline: skip or log-and-continue when the list of texts is empty before calling embedding().
2. If you genuinely need pairwise similarity, pass 2+ sentences (input[0] is source_sentence, the rest are compared).
3. For plain embeddings of N texts, be aware this path sends {source_sentence, sentences} — for a single text use a model without 'sentence-transformers' in the name or verify the response shape.

Example fix

# before
texts = chunk(doc)  # may be []
resp = litellm.embedding(model='huggingface/sentence-transformers/all-MiniLM-L6-v2', input=texts)

# after
texts = chunk(doc)
if not texts:
    return []
resp = litellm.embedding(model='huggingface/sentence-transformers/all-MiniLM-L6-v2', input=texts)

def safe_embed_texts(model: str, texts: list[str]):
    if not texts:
        return None  # nothing to embed; skip the API call entirely
    return texts

When it happens

Trigger: litellm.embedding(model='huggingface/sentence-transformers/all-MiniLM-L6-v2', input=[]) — an empty list triggers the raise. A single-element input does NOT raise, despite the message.

Common situations: Upstream batching/chunking code produces an empty list (empty document, filtered-out batch) that is passed straight through to the embedding call.

Related errors

• Invalid task_type={task_type}. Expected one of={hf_tasks_emb
• sentence-similarity requires 2+ sentences
• Empty image_url string is not valid.
• reranker requires 2+ sentences
• {embeddings[error]}

AI-assisted analysis of BerriAI/litellm@6c2dcb801b (2026-08-15). Data as JSON: /api/errors/c4113a0b2e25c95a. Report an issue: GitHub.