better:优化prompt中时间的内容

This commit is contained in:
SengokuCola
2025-06-02 18:19:16 +08:00
parent a45c626b13
commit d957aef202
6 changed files with 21 additions and 118 deletions

View File

@@ -616,6 +616,8 @@ def translate_timestamp_to_human_readable(timestamp: float, mode: str = "normal"
"""
if mode == "normal":
return time.strftime("%Y-%m-%d %H:%M:%S", time.localtime(timestamp))
if mode == "normal_no_YMD":
return time.strftime("%H:%M:%S", time.localtime(timestamp))
elif mode == "relative":
now = time.time()
diff = now - timestamp
@@ -634,111 +636,4 @@ def translate_timestamp_to_human_readable(timestamp: float, mode: str = "normal"
return time.strftime("%Y-%m-%d %H:%M:%S", time.localtime(timestamp)) + ":\n"
else: # mode = "lite" or unknown
# 只返回时分秒格式,喵~
return time.strftime("%H:%M:%S", time.localtime(timestamp))
def parse_text_timestamps(text: str, mode: str = "normal") -> str:
"""解析文本中的时间戳并转换为可读时间格式
Args:
text: 包含时间戳的文本,时间戳应以[]包裹
mode: 转换模式传递给translate_timestamp_to_human_readable"normal""relative"
Returns:
str: 替换后的文本
转换规则:
- normal模式: 将文本中所有时间戳转换为可读格式
- lite模式:
- 第一个和最后一个时间戳必须转换
- 以5秒为间隔划分时间段每段最多转换一个时间戳
- 不转换的时间戳替换为空字符串
"""
# 匹配[数字]或[数字.数字]格式的时间戳
pattern = r"\[(\d+(?:\.\d+)?)\]"
# 找出所有匹配的时间戳
matches = list(re.finditer(pattern, text))
if not matches:
return text
# normal模式: 直接转换所有时间戳
if mode == "normal":
result_text = text
for match in matches:
timestamp = float(match.group(1))
readable_time = translate_timestamp_to_human_readable(timestamp, "normal")
# 由于替换会改变文本长度,需要使用正则替换而非直接替换
pattern_instance = re.escape(match.group(0))
result_text = re.sub(pattern_instance, readable_time, result_text, count=1)
return result_text
else:
# lite模式: 按5秒间隔划分并选择性转换
result_text = text
# 提取所有时间戳及其位置
timestamps = [(float(m.group(1)), m) for m in matches]
timestamps.sort(key=lambda x: x[0]) # 按时间戳升序排序
if not timestamps:
return text
# 获取第一个和最后一个时间戳
first_timestamp, first_match = timestamps[0]
last_timestamp, last_match = timestamps[-1]
# 将时间范围划分成5秒间隔的时间段
time_segments = {}
# 对所有时间戳按15秒间隔分组
for ts, match in timestamps:
segment_key = int(ts // 15) # 将时间戳除以15取整作为时间段的键
if segment_key not in time_segments:
time_segments[segment_key] = []
time_segments[segment_key].append((ts, match))
# 记录需要转换的时间戳
to_convert = []
# 从每个时间段中选择一个时间戳进行转换
for _, segment_timestamps in time_segments.items():
# 选择这个时间段中的第一个时间戳
to_convert.append(segment_timestamps[0])
# 确保第一个和最后一个时间戳在转换列表中
first_in_list = False
last_in_list = False
for ts, _ in to_convert:
if ts == first_timestamp:
first_in_list = True
if ts == last_timestamp:
last_in_list = True
if not first_in_list:
to_convert.append((first_timestamp, first_match))
if not last_in_list:
to_convert.append((last_timestamp, last_match))
# 创建需要转换的时间戳集合,用于快速查找
to_convert_set = {match.group(0) for _, match in to_convert}
# 首先替换所有不需要转换的时间戳为空字符串
for _, match in timestamps:
if match.group(0) not in to_convert_set:
pattern_instance = re.escape(match.group(0))
result_text = re.sub(pattern_instance, "", result_text, count=1)
# 按照时间戳原始顺序排序,避免替换时位置错误
to_convert.sort(key=lambda x: x[1].start())
# 执行替换
# 由于替换会改变文本长度,从后向前替换
to_convert.reverse()
for ts, match in to_convert:
readable_time = translate_timestamp_to_human_readable(ts, "relative")
pattern_instance = re.escape(match.group(0))
result_text = re.sub(pattern_instance, readable_time, result_text, count=1)
return result_text
return time.strftime("%H:%M:%S", time.localtime(timestamp))