xtekky/gpt4free · error · RuntimeError

LMArena: {json.loads(line[3:])}

Error message

LMArena: {json.loads(line[3:])}

What it means

In LMArena's streaming loop, a line prefixed 'a3:' carries a JSON error payload for the model-A side of the battle; g4f raises it verbatim as RuntimeError. Unlike hasArenaError, this includes structured error details from the arena backend (a3: is the counterpart of the 'ad:' finish-reason line).

Source

Thrown at g4f/Provider/needs_auth/LMArena.py:709

                                ]
                                if __images:
                                    yield ImageResponse(
                                        __images, prompt, {"model": modelB}
                                    )

                            elif line.startswith("ad:"):
                                yield JsonConversation(
                                    evaluationSessionId=evaluationSessionId
                                )
                                finish = json.loads(line[3:])
                                if "finishReason" in finish:
                                    yield FinishReason(finish["finishReason"])
                                if "usage" in finish:
                                    yield Usage(**finish["usage"])
                            elif line.startswith("bd:"):
                                ...
                            elif line.startswith("a3:"):
                                raise RuntimeError(f"LMArena: {json.loads(line[3:])}")
                            elif line.startswith("b3:"):
                                ...
                            else:
                                debug.log(f"LMArena: Unknown line prefix: {line[:2]}")
                break
            except (CloudflareError, MissingAuthError) as error:
                args = None
                debug.error(error)
                debug.log(f"{cls.__name__}: Cloudflare error")
                continue
            except RateLimitError as error:
                if "prompt failed" in str(error):
                    raise
                args = None
                _need_clear_cookies = True
                debug.error(error)
                continue
            except:

View on GitHub (pinned to 973504e177)

Solutions

1. Read the JSON payload inside the message — it names the actual backend cause; act on that (shorten context, change model, etc.)
2. Start a fresh conversation (drop the saved JsonConversation/evaluationSessionId) and retry
3. Reduce message history or media size if the payload mentions limits
4. Update g4f and retry later if the payload indicates a platform-side error

Example fix

// before
# opaque failure
await client.chat.completions.create(model=..., messages=long_history)

// after
# surface and log the a3 payload, then trim history
except RuntimeError as e:
    logging.error('LMArena a3 error detail: %s', str(e))
resp = await client.chat.completions.create(model=..., messages=messages[-6:])

try:
    async for chunk in stream:
        handle(chunk)
except RuntimeError as e:
    # str(e) contains the a3 JSON payload with the backend reason
    logging.error('LMArena a3 error: %s', e)
    if 'context' in str(e).lower():
        messages = messages[-6:]  # trim and retry
        raise Retryable(messages)

When it happens

Trigger: During response.iter_lines() of the chat POST, a line starting with 'a3:' arrives whose JSON body describes the failure — for example token/context limits exceeded, malformed conversation state, or backend exceptions surfaced mid-stream.

Common situations: Very long conversations that exceed the arena model's context; a stale/invalid evaluationSessionId reused from an old conversation object; upstream backend errors during heavy load.

Related errors

• LMArena Beta encountered an error: hasArenaError
• line.get("error")
• {r.status} {response body text}
• No audio chunks
• Install or update "curl_cffi" package | pip install -U curl_

AI-assisted analysis of xtekky/gpt4free@973504e177 (2026-08-14). Data as JSON: /api/errors/7e8cd934214ccc30. Report an issue: GitHub.