mirror of
https://github.com/linshenkx/prompt-optimizer.git
synced 2026-07-01 23:44:42 +08:00
2cdd095c2b
- 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
46 lines
1.5 KiB
JSON
46 lines
1.5 KiB
JSON
{
|
|
"type": "compare",
|
|
"score": {
|
|
"overall": 78,
|
|
"dimensions": [
|
|
{
|
|
"key": "goalAchievementRobustness",
|
|
"label": "目标达成稳定性",
|
|
"score": 70
|
|
},
|
|
{
|
|
"key": "outputQualityCeiling",
|
|
"label": "输出质量上限",
|
|
"score": 90
|
|
},
|
|
{
|
|
"key": "promptPatternQuality",
|
|
"label": "提示词模式质量",
|
|
"score": 65
|
|
},
|
|
{
|
|
"key": "crossSnapshotRobustness",
|
|
"label": "跨快照鲁棒性",
|
|
"score": 60
|
|
},
|
|
{
|
|
"key": "workspaceTransferability",
|
|
"label": "对工作区的可迁移性",
|
|
"score": 85
|
|
}
|
|
]
|
|
},
|
|
"improvements": [
|
|
"在原始提示词中补充明确的任务类型（如七言律诗）、表达方式要求（含蓄、不解释）和意象范围，可以显著提升模型输出的质量与一致性。",
|
|
"为提高跨快照鲁棒性，建议在提示词中加入格式规范（如八句四联、押平声韵），以减少模型因理解偏差导致的输出不稳定。",
|
|
"在需要控制输出风格时，应避免模糊指令（如“写一首诗”），而应使用更具体的任务描述，引导模型聚焦于特定文学形式或情感表达。"
|
|
],
|
|
"summary": "快照 B 的输出质量更高，主要得益于其明确了任务类型、格式约束及禁止项，这些差异直接影响了生成结果的文学性和一致性。",
|
|
"patchPlan": [],
|
|
"metadata": {
|
|
"model": "dashscope",
|
|
"timestamp": 1773732635109,
|
|
"duration": 10153
|
|
}
|
|
}
|