alibaba/spring-ai-alibaba · error · RuntimeException

解析模型调用结果出错，请重试

Error message

解析模型调用结果出错，请重试

What it means

Thrown by EvaluatorServiceImpl.evaluatorTest after the LLM responds: the model's output string cannot be parsed into EvaluatorDebugResult via fastjson JSONObject.parseObject. The error indicates the model returned something other than the expected JSON structure.

Source

Thrown at spring-ai-alibaba-admin/spring-ai-alibaba-admin-server-start/src/main/java/com/alibaba/cloud/ai/studio/admin/service/impl/EvaluatorServiceImpl.java:190

        String userPrompt = modelConfigParser.replaceVariables(request.getPrompt(), request.getVariables());

        String prompt = userPrompt.concat(SYSTEM_PROMPT);

        log.info("evaluatorTest:prompt,{}", prompt);

        String response = Objects.requireNonNull(client.prompt(prompt).call().content()).trim();

        log.info("模型返回值:{}", response);

        String formatedResponse = extractRawText(response);
        log.info("模型返回值:{},格式化后模型返回值:{}.", response, formatedResponse);

        try {
            return JSONObject.parseObject(formatedResponse, EvaluatorDebugResult.class);
        } catch (Exception e) {
            log.info("解析失败: {}", formatedResponse, e);
            throw new RuntimeException("解析模型调用结果出错，请重试");
        }

    }

} 

View on GitHub (pinned to f82da0b50f)

Solutions

1. Retry the debug call — the message itself says 'please retry' and LLM output is nondeterministic
2. Strip markdown code fences and extract the first {...} block before parseObject
3. Enable JSON mode / structured output on the model call and tighten the prompt to mandate exact JSON schema
4. Inspect the log line '解析失败: {}' to see the raw response and fix the prompt or parser accordingly
5. Wrap parseObject with a fallback parser or default EvaluatorDebugResult on failure

Example fix

// before
return JSONObject.parseObject(formatedResponse, EvaluatorDebugResult.class);
// after
String json = extractJsonBlock(formatedResponse); // strip ``` fences / surrounding text
try {
    return JSONObject.parseObject(json, EvaluatorDebugResult.class);
} catch (Exception e) {
    throw new EvaluatorResponseParseException("Model returned non-JSON output: " + json, e);
}

String json = extractJsonBlock(response);
if (json == null || !json.trim().startsWith("{")) throw new RetryableFormatException(response);

boolean looksLikeJson(String s) { return s != null && s.replace("```","").trim().startsWith("{"); }

try { return JSONObject.parseObject(formatedResponse, EvaluatorDebugResult.class); } catch (Exception e) { log.warn("non-JSON model output: {}", formatedResponse); return retryWithStrictJsonPrompt(); }

When it happens

Trigger: The evaluator's model prompt yields free-form text, markdown fences (```json ... ```), trailing prose, or truncated JSON; formatedResponse still isn't valid JSON matching EvaluatorDebugResult fields.

Common situations: Weak/older models ignoring strict JSON output instructions; temperature too high; prompt template edited so the JSON schema example was lost; model returning Chinese text where the parser expects typed fields; max_tokens cutting JSON mid-object.

Understand the failure class

Background: JSON parse error: "Unexpected token" / "not valid JSON" / "failed to parse" — what JSON parsers are really complaining about — this error's family across 45 libraries.

Related errors

• ChatClient successfully returned, but the returned json is i
• Nacos config content is not valid JSON, but dot notation was
• 解析默认参数失败: {}
• 解析itemIds字符串失败: {}
• CreateMCPServerError

AI-assisted analysis of alibaba/spring-ai-alibaba@f82da0b50f (2026-09-09). Data as JSON: /api/errors/14d00dc271039ad3. Report an issue: GitHub.