{
    "error_type": "quant_accuracy_error",
    "description": "INT8量化后精度不达标",
    "error_patterns": [
        "accuracy threshold exceeded: expected <0.1, got 0.35",
        "cosine distance: 0.85 (threshold: 0.99)",
        "precision loss detected after quantization",
        "量化精度下降: 输出差异过大",
        "verification failed: quantized model accuracy below threshold"
    ],
    "diagnosis": {
        "cause": "INT8量化导致模型精度损失超过预期阈值",
        "check_points": [
            "量化校准数据集是否合理",
            "量化参数配置是否正确",
            "模型中是否有不适合量化的敏感层"
        ]
    },
    "suggestions": [
        "检查量化校准数据集是否合理且覆盖典型输入",
        "考虑使用混合精度量化或部分量化",
        "不使用量化重新转换模型对比精度差异"
    ],
    "example_log": "[WARNING] quant accuracy error: INT8 quantization caused significant precision loss. Cosine distance between FP32 and INT8 outputs: 0.85 (threshold: 0.99). Accuracy threshold exceeded. Consider adjusting calibration data or using mixed precision."
}