better:优化表达方式提取的token消耗

This commit is contained in:
SengokuCola
2025-12-03 00:21:02 +08:00
parent 34129bafad
commit 7f66d5588d
6 changed files with 155 additions and 332 deletions

View File

@@ -429,15 +429,36 @@ class ChatHistorySummarizer:
# 2. 构造编号后的消息字符串和参与者信息
numbered_lines, index_to_msg_str, index_to_msg_text, index_to_participants = self._build_numbered_messages_for_llm(messages)
# 3. 调用 LLM 识别话题,并得到 topic -> indices
# 3. 调用 LLM 识别话题,并得到 topic -> indices(失败时最多重试 3 次)
existing_topics = list(self.topic_cache.keys())
success, topic_to_indices = await self._analyze_topics_with_llm(
numbered_lines=numbered_lines,
existing_topics=existing_topics,
)
max_retries = 3
attempt = 0
success = False
topic_to_indices: Dict[str, List[int]] = {}
while attempt < max_retries:
attempt += 1
success, topic_to_indices = await self._analyze_topics_with_llm(
numbered_lines=numbered_lines,
existing_topics=existing_topics,
)
if success and topic_to_indices:
if attempt > 1:
logger.info(
f"{self.log_prefix} 话题识别在第 {attempt} 次重试后成功 | 话题数: {len(topic_to_indices)}"
)
break
logger.warning(
f"{self.log_prefix} 话题识别失败或无有效话题,第 {attempt} 次尝试失败"
+ ("" if attempt >= max_retries else ",准备重试")
)
if not success or not topic_to_indices:
logger.warning(f"{self.log_prefix} 话题识别失败或无有效话题,本次检查忽略")
logger.error(
f"{self.log_prefix} 话题识别连续 {max_retries} 次失败或始终无有效话题,本次检查放弃"
)
# 即使识别失败,也认为是一次“检查”,但不更新 no_update_checks保持原状
return