agentscope-ai/agentscope · error · RuntimeError

Failed to call embedding model {self.model} after {self.max_

Error message

Failed to call embedding model {self.model} after {self.max_retries + 1} retries.

What it means

Raised by EmbeddingModelBase._call_with_retry when the embedded provider call raised no exception on the final pass but also produced no usable outcome path (defensive terminal error after max_retries+1 attempts), reporting the model name and retry count.

Source

Thrown at src/agentscope/embedding/_embedding_base.py:354

                        "Batch attempt %d failed for embedding model "
                        "%s: %s. Retrying in %.1fs...",
                        attempt + 1,
                        self.model,
                        str(e),
                        self.retry_delay,
                    )
                    await asyncio.sleep(self.retry_delay)
                else:
                    logger.warning(
                        "All %d attempt(s) failed for a batch of "
                        "embedding model %s.",
                        self.max_retries + 1,
                        self.model,
                    )

        if last_error is not None:
            raise last_error
        raise RuntimeError(
            f"Failed to call embedding model {self.model} after "
            f"{self.max_retries + 1} retries.",
        )

    # ------------------------------------------------------------------
    # Abstract — subclasses implement this for a single batch
    # ------------------------------------------------------------------

    @abstractmethod
    async def _call_api(
        self,
        inputs: list[Any],
        **kwargs: Any,
    ) -> EmbeddingResponse:
        """Call the underlying embedding API for a **single batch**.

        Subclasses must implement this method.  The batch splitting,
        concurrency, and retry logic are handled by :meth:`__call__`

View on GitHub (pinned to e90f1c7592)

Solutions

1. Inspect the underlying provider error (last_error / logs) and fix credentials, quota, or batch size
2. Increase max_retries and use exponential backoff for flaky networks
3. Cache embeddings and enqueue failures for later replay instead of retrying inline forever

Example fix

# before
vecs = await model(texts)
# after
try:
    vecs = await model(texts)
except RuntimeError:
    vecs = await replay_later(texts)  # queue and degrade gracefully

for attempt in range(N):\n    try:\n        vecs = await model(texts); break\n    except RuntimeError as e:\n        if \"after\" in str(e) and \"retries\" in str(e) and attempt < N - 1:\n            await asyncio.sleep(2 ** attempt); continue\n        raise

When it happens

Trigger: Exhausting all retry attempts for __call__ on an embedding model — each attempt failing (network/API errors) — after which the wrapper gives up; also reachable if retries complete without success and without a captured last_error.

Common situations: Sustained provider outages or rate limiting; invalid credentials causing every retry to fail; very large batches that consistently time out.

Related errors

• DashScope text embedding API error: {response}
• DashScope multimodal embedding API error: {res}
• Failed to fetch bytes from URL `{url}` after {max_retries} r
• Model call failed after retries, but no exception was raised
• Text embedding model {self.model!r} only accepts str inputs,

AI-assisted analysis of agentscope-ai/agentscope@e90f1c7592 (2026-08-28). Data as JSON: /api/errors/b12893ccd3c1c333. Report an issue: GitHub.