BerriAI/litellm · error · PerplexityEmbeddingError

raw_response.text

Error message

raw_response.text

What it means

LiteLLM's Perplexity embedding handler calls raw_response.json(); when the body is not JSON it raises PerplexityEmbeddingError with the raw body text as the message and the real HTTP status code. Note Perplexity's API is primarily search/chat, so embedding models are limited and invalid model names commonly produce non-JSON errors.

Source

Thrown at litellm/llms/perplexity/embedding/transformation.py:151

            int8_values: Final = struct.unpack(f"{count}b", raw_bytes)
            return [float(v) / 127.0 for v in int8_values]
        return embedding_value

    def transform_embedding_response(
        self,
        model: str,
        raw_response: httpx.Response,
        model_response: EmbeddingResponse,
        logging_obj: LiteLLMLoggingObj,
        api_key: str | None = None,
        request_data: dict = {},
        optional_params: dict = {},
        litellm_params: dict = {},
    ) -> EmbeddingResponse:
        try:
            raw_response_json: Final = raw_response.json()
        except Exception:
            raise PerplexityEmbeddingError(message=raw_response.text, status_code=raw_response.status_code)

        model_response.model = raw_response_json.get("model", model)
        model_response.object = raw_response_json.get("object", "list")

        raw_data: Final = raw_response_json.get("data", [])
        decoded_data: Final[list[dict[str, Any]]] = []
        for item in raw_data:
            decoded_item = dict(item)
            decoded_item["embedding"] = self._decode_base64_embedding(item.get("embedding"))
            decoded_data.append(decoded_item)
        model_response.data = decoded_data

        usage_data: Final = raw_response_json.get("usage", {})
        usage: Final = Usage(
            prompt_tokens=usage_data.get("prompt_tokens", 0) or usage_data.get("total_tokens", 0),
            total_tokens=usage_data.get("total_tokens", 0),
        )
        model_response.usage = usage

View on GitHub (pinned to 77b7c6c40c)

Solutions

1. Read the exception message - it is the raw body naming the real problem
2. Confirm the embedding model actually exists on Perplexity's API
3. Verify/rotate PERPLEXITYAI_API_KEY
4. Route embeddings to a provider that supports them (e.g., OpenAI) if Perplexity lacks the model

Catch PerplexityEmbeddingError around litellm.embedding(); log the message (raw body) to identify auth vs. model-availability issues, and fall back to an OpenAI-compatible embedding provider when Perplexity rejects the model.

When it happens

Trigger: Calling litellm.embedding(model='perplexity/...') with an unavailable embedding model, an invalid/expired API key, or an api_base pointing at something that returns HTML or plain text.

Common situations: Assuming Perplexity serves OpenAI-compatible embeddings it does not offer; expired PERPLEXITYAI_API_KEY; gateway/CDN error pages during incidents.

Related errors

• raw_response.text
• Error parsing OpenRouter response: {e}
• Error parsing OpenRouter response: {e}
• raw_response.text
• judge response is not a JSON object

AI-assisted analysis of BerriAI/litellm@77b7c6c40c (2026-08-18). Data as JSON: /api/errors/57782a87c93dc4ca. Report an issue: GitHub.