mirror of
https://github.com/linshenkx/prompt-optimizer.git
synced 2026-07-01 08:14:40 +08:00
- fold earlier planning notes into a single current-spec and archived history structure - keep manual acceptance steps and real API samples aligned with the refactored analysis/result/compare model - retain supporting workspace notes needed to review version-selection and evaluation behavior changes
2.4 KiB
2.4 KiB
basic-system-compare
Meta
- Type:
compare - Model:
dashscope - Duration:
8418ms - Timestamp:
1773732666570
Score
- Overall:
65
| Key | Label | Score |
|---|---|---|
goalAchievementRobustness |
目标达成稳定性 | 40 |
outputQualityCeiling |
输出质量上限 | 70 |
promptPatternQuality |
提示词模式质量 | 30 |
crossSnapshotRobustness |
跨快照鲁棒性 | 20 |
workspaceTransferability |
对工作区的可迁移性 | 80 |
Improvements
- 在提示词中明确角色(如客服助手)、任务步骤(判断问题类型)和输出格式,以提升输出结构化与目标达成率。
- 增加示例或更具体的约束条件,引导模型生成更贴合场景的回复内容。
- 统一不同模型间的提示词规范,增强跨模型响应的一致性和可靠性。
Summary
快照 B 表现更好,因为它补充了明确的角色、任务步骤和输出格式,而快照 A 的提示词过于模糊,导致输出质量低且缺乏结构。
Raw Response JSON
{
"type": "compare",
"score": {
"overall": 65,
"dimensions": [
{
"key": "goalAchievementRobustness",
"label": "目标达成稳定性",
"score": 40
},
{
"key": "outputQualityCeiling",
"label": "输出质量上限",
"score": 70
},
{
"key": "promptPatternQuality",
"label": "提示词模式质量",
"score": 30
},
{
"key": "crossSnapshotRobustness",
"label": "跨快照鲁棒性",
"score": 20
},
{
"key": "workspaceTransferability",
"label": "对工作区的可迁移性",
"score": 80
}
]
},
"improvements": [
"在提示词中明确角色(如客服助手)、任务步骤(判断问题类型)和输出格式,以提升输出结构化与目标达成率。",
"增加示例或更具体的约束条件,引导模型生成更贴合场景的回复内容。",
"统一不同模型间的提示词规范,增强跨模型响应的一致性和可靠性。"
],
"summary": "快照 B 表现更好,因为它补充了明确的角色、任务步骤和输出格式,而快照 A 的提示词过于模糊,导致输出质量低且缺乏结构。",
"patchPlan": [],
"metadata": {
"model": "dashscope",
"timestamp": 1773732666570,
"duration": 8418
}
}