BerriAI/litellm · error · Exception

Invalid task_type= . Expected one of=

Error message

Invalid task_type={task_type}. Expected one of={hf_tasks_embeddings}

What it means

The sync HF embedding helper validates an explicitly passed task_type against the allowed set ('sentence-similarity', 'feature-extraction', 'rerank', 'embed', 'similarity', matching HF pipeline tags and TEI endpoints). Any other value raises a plain Exception before any request is made. If task_type is omitted, it is auto-derived from the model's pipeline_tag via the Hub API.

Solutions

Use one of the exact allowed values: sentence-similarity, feature-extraction, rerank, embed, similarity
Omit task_type entirely to auto-detect from the model's pipeline_tag on the Hub
For plain embeddings, 'embed' (TEI endpoint) or 'feature-extraction' are the right choices
Check the literal hf_tasks_embeddings in litellm/llms/huggingface/embedding/handler.py if unsure

Example fix

# before
litellm.embedding(model="hf/BAAI/bge-m3", input=["hello"], task_type="text-embedding")

# after
litellm.embedding(model="hf/BAAI/bge-m3", input=["hello"], task_type="embed")
# or omit task_type to auto-detect the pipeline tag

Defensive patterns

Strategy: validation

Validate before calling

ALLOWED = {"sentence-similarity", "feature-extraction", "rerank", "embed", "similarity"}

if task_type is not None and task_type not in ALLOWED:
    raise ValueError(f"task_type must be one of {sorted(ALLOWED)}")

Type guard

HF_TASK_TYPES = {"sentence-similarity", "feature-extraction", "rerank", "embed", "similarity"}

def is_valid_hf_task_type(t: str | None) -> bool:
    return t is None or t in HF_TASK_TYPES

Prevention

Centralize the allowed set as a constant shared across all embedding call sites
Prefer omitting task_type and letting pipeline_tag auto-detection choose
Add a unit test asserting your config's task_type values are within the literal set

When it happens

Trigger: litellm.embedding(model='hf/<model>', input=[...], task_type='<bad>') with a value outside the set — e.g. 'text-embedding', 'embedding', or an OpenAI-style task name. Raised from the synchronous path (line 37).

Common situations: Copy-pasting task_type names from other providers (OpenAI/Azure/Vertex); assuming a free-form string; mismatched casing ('Embed'); passing the task name of a different HF model family.

Understand the failure class

Background: "Must be a positive integer", "Invalid value", "Unsupported": the invalid-argument-value error family, when a library rejects the value you pass — this error's family across 35 libraries.

Related errors

AI-assisted analysis of BerriAI/litellm@77b7c6c40c (2026-08-18). Data as JSON: /api/errors/c7ec8fd8c8e5c71a. Report an issue: GitHub.

Appendix: source

Thrown at litellm/llms/huggingface/embedding/handler.py:37

from .transformation import HuggingFaceEmbeddingConfig

config: Final = HuggingFaceEmbeddingConfig()

HF_HUB_URL: Final = "https://huggingface.co"

hf_tasks_embeddings: Final = (
    Literal[  # pipeline tags + hf tei endpoints - https://huggingface.github.io/text-embeddings-inference/#/
        "sentence-similarity", "feature-extraction", "rerank", "embed", "similarity"
    ]
)


def get_hf_task_embedding_for_model(model: str, task_type: str | None, api_base: str) -> str | None:
    if task_type is not None:
        if task_type in get_args(hf_tasks_embeddings):
            return task_type
        else:
            raise Exception(f"Invalid task_type={task_type}. Expected one of={hf_tasks_embeddings}")
    http_client: Final = HTTPHandler(concurrent_limit=1)

    model_info: Final = http_client.get(url=f"{api_base}/api/models/{model}")

    model_info_dict: Final = model_info.json()

    pipeline_tag: Final[str | None] = model_info_dict.get("pipeline_tag", None)

    return pipeline_tag


async def async_get_hf_task_embedding_for_model(model: str, task_type: str | None, api_base: str) -> str | None:
    if task_type is not None:
        if task_type in get_args(hf_tasks_embeddings):
            return task_type
        else:
            raise Exception(f"Invalid task_type={task_type}. Expected one of={hf_tasks_embeddings}")
    http_client: Final = get_async_httpx_client(

View on GitHub (pinned to 77b7c6c40c)