MCPcopy Create free account
hub / github.com/Hunyuan-PromptEnhancer/PromptEnhancer / build_instruction

Function build_instruction

t2i_keypoints_eval.py:160–177  ·  view source on GitHub ↗

Build the judge instruction for a single key point.

(prompt: str, key_point: str, key_point_desc: str, language: str)

Source from the content-addressed store, hash-verified

158# Evaluation
159# --------------------------------------------------------------------------- #
160def build_instruction(prompt: str, key_point: str, key_point_desc: str, language: str) -> str:
161 """Build the judge instruction for a single key point."""
162 if language == "zh":
163 return (
164 f"该图片由 prompt:[{prompt}] 通过 AIGC 生成。请判断,基于图片画面中的可见部分,"
165 f"只需要考虑 [{key_point}] 是否严格满足 [{key_point_desc}] 要求。"
166 f"accept 代表满足要求,reject 代表不满足要求。"
167 f"仅根据图片视觉呈现的可见部分(不要联想画面,不要猜测画面),"
168 f"详细分析得到结果的依据并给出结果:"
169 )
170 return (
171 f"This image is generated by prompt:[{prompt}] through AIGC. Please judge, "
172 f"based on the visible part of the image, only consider whether [{key_point}] "
173 f"strictly meets the [{key_point_desc}] requirement. accept means meeting the "
174 f"requirement, reject means not meeting the requirement. Only rely on the "
175 f"visible part of the image (do not associate the scene & do not guess the "
176 f"scene), analyze the basis for the result in detail and give the result:"
177 )
178
179
180def eval_sample(sample: dict, judge: GeminiJudge) -> dict:

Callers 1

eval_sampleFunction · 0.85

Calls

no outgoing calls

Tested by

no test coverage detected