xtekky/gpt4free · error · ResponseError

GPU token limit exceeded: {chunk.decode(errors='replace')}

Error message

GPU token limit exceeded: {chunk.decode(errors='replace')}

What it means

ResponseError raised while streaming the StabilityAI SD3.5 Large HF Space Gradio SSE feed: an 'event: error' frame was received, and the following 'data: ' line is surfaced as the error text. Despite the hardcoded 'GPU token limit exceeded' prefix, the actual cause is whatever the Space reported in the error event — most commonly exhausted ZeroGPU quota, but any Space-side error event takes this path.

Source

Thrown at g4f/Provider/hf_space/StabilityAI_SD35Large.py:81

                    num_inference_steps,
                ]
            }
            async with session.post(
                f"{cls.url}{cls.api_endpoint}", json=data, proxy=proxy
            ) as response:
                response.raise_for_status()
                event_id = (await response.json()).get("event_id")
                async with session.get(
                    f"{cls.url}{cls.api_endpoint}/{event_id}"
                ) as event_response:
                    event_response.raise_for_status()
                    event = None
                    async for chunk in event_response.content:
                        if chunk.startswith(b"event: "):
                            event = chunk[7:].decode(errors="replace").strip()
                        if chunk.startswith(b"data: "):
                            if event == "error":
                                raise ResponseError(
                                    f"GPU token limit exceeded: {chunk.decode(errors='replace')}"
                                )
                            if event in ("complete", "generating"):
                                try:
                                    data = json.loads(chunk[6:])
                                    if data is None:
                                        continue
                                    url = data[0]["url"]
                                except (json.JSONDecodeError, KeyError, TypeError) as e:
                                    raise RuntimeError(
                                        f"Failed to parse image URL: {chunk.decode(errors='replace')}",
                                        e,
                                    )
                                if event == "generating":
                                    yield ImagePreview(url, prompt)
                                else:
                                    yield ImageResponse(url, prompt)
                                    break

View on GitHub (pinned to 973504e177)

Solutions

1. Read the data line after the prefix — it states the actual cause (e.g. 'You have exceeded your GPU quota').
2. Wait for the quota window to reset (HF ZeroGPU grants quota per hour/week) before retrying.
3. Provide HF cookies or a valid zerogpu token/api_key so the request is billed to an account with remaining quota.
4. If the error text is unrelated to quota, treat it as a Space-side failure: retry later or switch to another SD3.5 provider.

try:
    ...
except ResponseError as e:
    if 'GPU token limit exceeded' in str(e):
        await asyncio.sleep(3600)  # wait out the ZeroGPU quota window, or switch provider
        ...

When it happens

Trigger: Joining the gradio_api queue with a ZeroGPU token whose GPU-time quota is exhausted (the Space emits event: error with a quota message); calling the Space anonymously when it requires a ZeroGPU-eligible account; or any Space-side exception surfacing as an error event over the SSE stream.

Common situations: Many image generations in a short window on HF ZeroGPU Spaces draining the free quota; using an unauthenticated session after HF tightened anonymous ZeroGPU access; expired zerogpu token obtained earlier in the session.

Related errors

• Failed to parse message: {chunk.decode(errors='replace')}
• Failed to parse image URL: {chunk.decode(errors='replace')}
• No media files provided for image generation.
• No coins left. Log in with a different account or wait a whi
• No response

AI-assisted analysis of xtekky/gpt4free@973504e177 (2026-08-14). Data as JSON: /api/errors/28aff528866e7481. Report an issue: GitHub.