diff --git a/rag2_0/dify/DifyQueryRetrieval.py b/rag2_0/dify/DifyQueryRetrieval.py
index 7b14e01..6e0fb64 100644
--- a/rag2_0/dify/DifyQueryRetrieval.py
+++ b/rag2_0/dify/DifyQueryRetrieval.py
@@ -189,8 +189,7 @@ class DifyQueryRetrieval:
if doc_title not in hit_doc_titles_set: # 确保不添加重复的标题
hit_doc_titles_set.add(doc_title)
query_hit_stats[query_type].append(doc_title)
-
- logging.info(f"查询命中统计: {json.dumps(query_hit_stats, ensure_ascii=False)}")
+
return {
"documents": processed_documents,
diff --git a/rag2_0/dify/DifyQueryRetrieval_api.py b/rag2_0/dify/DifyQueryRetrieval_api.py
index c421a3a..52c05c9 100644
--- a/rag2_0/dify/DifyQueryRetrieval_api.py
+++ b/rag2_0/dify/DifyQueryRetrieval_api.py
@@ -99,7 +99,7 @@ async def retrieve(request: RetrieveRequest):
query_list,
data_set_list,
query_expand_dict=query_expand_dict,
- top_k=5
+ top_k=4
)
end_time = time.time()
diff --git a/rag2_0/intent_recognition/IntentRecognition.py b/rag2_0/intent_recognition/IntentRecognition.py
index 6514585..323db3e 100755
--- a/rag2_0/intent_recognition/IntentRecognition.py
+++ b/rag2_0/intent_recognition/IntentRecognition.py
@@ -22,14 +22,14 @@ import threading
from .PromptTemplates import (classification_prompt, query_rewrite_prompt_pro,
extract_nouns_prompt, classification_info,
slot_filling_prompt, step_back_prompt,
- follow_up_questions_prompt, hyde_prompt, multi_questions_prompt)
+ hyde_prompt)
from .DataModels import (
Classification, QueryRewrite, Term, TermList,
SoftwareFunctionSlots, SoftwareTroubleShootingSlots, ProfessionalConsultingSlots,
DataProblemSlots, FileExtensionConsultingSlots, SoftwareLockSlots,
InstallationDownloadSlots, ProblemDiagnosisSlots, OtherSlots, IntentAndSlotResult,
- StepBackPrompt, FollowUpQuestions, HypotheticalDocument, MultiQuestions
+ StepBackPrompt, HypotheticalDocument
)
from .ProfessionalNounVector import ProfessionalNounRetriever, AsyncProfessionalNounRetriever
from rag2_0.tool.ModelTool import XinferenceReRankerModel, OpenAiLLM, SiliconFlowReRankerModel
@@ -147,7 +147,6 @@ class AsyncIntentRecognizer:
Returns:
分类结果
"""
- classification_start_time = time.time()
classification_parser = PydanticOutputParser(pydantic_object=Classification)
formatted_prompt = classification_prompt.format(user_input=query,
classification_info=classification_info,
@@ -159,10 +158,6 @@ class AsyncIntentRecognizer:
# 异步调用LLM
response = await self._llm.invoke_async(formatted_prompt, False)
- classification_end_time = time.time()
- classification_time = classification_end_time - classification_start_time
- logging.info(f"异步意图分类耗时统计 - 总耗时: {classification_time:.2f}秒")
-
# 尝试直接解析JSON响应
response.content = response.content.strip()
clean_output = re.sub(r'.*?', '', response.content, flags=re.DOTALL)
@@ -356,7 +351,6 @@ class AsyncIntentRecognizer:
改写结果
"""
- rewrite_start_time = time.time()
# 准备问题改写提示
terms_dict = [term.model_dump(exclude={"description"}) for term in keywords.terms]
keywords_str = json.dumps(terms_dict, ensure_ascii=False)
@@ -373,9 +367,6 @@ class AsyncIntentRecognizer:
response.content = response.content.strip()
clean_output = re.sub(r'.*?', '', response.content, flags=re.DOTALL)
parsed_output = query_rewrite_parser.parse(clean_output)
- rewrite_end_time = time.time()
- rewrite_time = rewrite_end_time - rewrite_start_time
- logging.info(f"异步问题改写耗时统计 - 总耗时: {rewrite_time:.2f}秒")
return parsed_output
except Exception as e:
raise RuntimeError(f"解析问题改写结果时出错: {e}") from e
@@ -393,11 +384,11 @@ class AsyncIntentRecognizer:
async def process_query_async(self, query: str, conversation_context: Dict = None,
- chat_history: List[Dict[str, str]] = None,
- previous_slots: Dict[str, Any] = None,
- use_jieba: bool = False,
- enable_query_expansion: bool = False,
- cur_soft_name: str = "") -> Dict[str, Any]:
+ chat_history: List[Dict[str, str]] = None,
+ previous_slots: Dict[str, Any] = None,
+ use_jieba: bool = False,
+ enable_query_expansion: bool = False,
+ cur_soft_name: str = "") -> Dict[str, Any]:
"""
异步处理用户问题的完整流程
@@ -423,26 +414,20 @@ class AsyncIntentRecognizer:
if enable_query_expansion:
# 创建异步任务并立即开始执行
query_expand_tasks = [
- # 5.1: 后退提示
+ # 后退提示
asyncio.create_task(self._generate_step_back_prompt_async(query, chat_history, conversation_context)),
- # 5.2: Follow Up Questions
- asyncio.create_task(self._generate_follow_up_questions_async(query, chat_history, conversation_context)),
-
- # 5.3: 文档查询
+ # 文档查询
asyncio.create_task(self._find_matching_software_docs_async(query, cur_soft_name, chat_history)),
-
- # 5.4: 多问题查询
- asyncio.create_task(self._generate_multi_questions_async(query, chat_history, conversation_context))
]
- # 步骤1-3: 并行执行关键词匹配、问题改写和意图分类
+ # 执行关键词匹配
keywords_task = self._match_keywords_async(query, use_jieba)
# 等待关键词匹配完成
keywords_terms, query_keys = await keywords_task
- # 步骤2: 问题改写
+ # 步骤2-3: 并行执行问题改写和意图分类
rewrite_task = self._rewrite_query_async(
query=query,
keywords=keywords_terms,
@@ -450,14 +435,14 @@ class AsyncIntentRecognizer:
chat_history=chat_history,
context=conversation_context
)
+ classification_task = self._classify_intent_async(query, conversation_context, chat_history, previous_slots)
- # 等待问题改写完成
- rewrite = await rewrite_task
+ # 并行等待问题改写和意图分类完成
+ start_time = time.time()
+ rewrite, classification = await asyncio.gather(rewrite_task, classification_task)
+ end_time = time.time()
+ logging.info(f"意图分类耗时统计 - 总耗时: {end_time - start_time:.2f}秒")
- # 步骤3: 进行意图分类
- classification_task = self._classify_intent_async(rewrite.rewrite, conversation_context, chat_history, previous_slots)
- classification = await classification_task
-
# 特殊处理 锁相关咨询
if classification.vertical_classification == "安装下载注册" and classification.sub_classification == "软件锁类":
process_lock_start_time = time.time()
@@ -465,12 +450,9 @@ class AsyncIntentRecognizer:
process_lock_end_time = time.time()
process_lock_time = process_lock_end_time - process_lock_start_time
logging.info(f"锁相关咨询正则匹配 - 总耗时: {process_lock_time:.2f}秒")
- # 步骤4: 进行槽位填充
- # 如果是有效分类，进行槽位填充
+
slot_filling_result = {}
- # if classification.vertical_classification not in ["其他", "闲聊"] and classification.sub_classification not in ["其他", "闲聊"]:
- # slot_filling_result = await self._fill_slots_async(rewrite.rewrite, classification, conversation_context, chat_history, previous_slots)
-
+
if not enable_query_expansion:
return {
"classification": classification.model_dump(),
@@ -484,28 +466,22 @@ class AsyncIntentRecognizer:
start_time = time.time()
query_expand_results = await asyncio.gather(*query_expand_tasks)
end_time = time.time()
- logging.info(f"异步问题扩展环节耗时统计 - 总耗时: {end_time - start_time:.2f}秒")
+ logging.info(f"问题扩展环节耗时统计 - 总耗时: {end_time - start_time:.2f}秒")
# 收集结果
step_back_result = query_expand_results[0] if query_expand_results[0] else StepBackPrompt(original_query=query, can_use_back_prompt=False, step_back_query=[query])
- follow_up_result = query_expand_results[1] if query_expand_results[1] else FollowUpQuestions(original_query=query, follow_up_query=query)
- wiki_result = query_expand_results[2] if query_expand_results[2] else []
- multi_questions_result = query_expand_results[3] if query_expand_results[3] else MultiQuestions(original_query=query, sub_questions=[query])
+ wiki_result = query_expand_results[1] if query_expand_results[1] else []
all_questions=[]
all_questions.append(query)
all_questions.append(rewrite.rewrite)
all_questions.extend(wiki_result)
all_questions.extend(step_back_result.step_back_query)
- all_questions.append(follow_up_result.follow_up_query)
- all_questions.extend(multi_questions_result.sub_questions)
all_questions = list(set(all_questions))
query_expand = {
"all": all_questions,
"step_back": step_back_result.step_back_query,
- "follow_up": [follow_up_result.follow_up_query],
- "multi_questions": multi_questions_result.sub_questions,
"wiki_title": wiki_result,
"original_query":query,
"rewrite_query":rewrite.rewrite
@@ -670,52 +646,13 @@ class AsyncIntentRecognizer:
parsed_output = step_back_parser.parse(clean_output)
step_back_end_time = time.time()
step_back_time = step_back_end_time - step_back_start_time
- logging.debug(f"异步后退提示生成耗时统计 - 总耗时: {step_back_time:.2f}秒")
+ logging.info(f"后退提示生成耗时统计 - 总耗时: {step_back_time:.2f}秒")
return parsed_output
except Exception as e:
# 如果解析失败，返回原始查询作为后退提示
- logging.error(f"异步后退提示生成失败: {e}", exc_info=True)
+ logging.error(f"后退提示生成失败: {e}", exc_info=True)
return StepBackPrompt(original_query=query, can_use_back_prompt=False, step_back_query=[query])
- async def _generate_follow_up_questions_async(self, query: str, chat_history: List[Dict[str, str]] = None, conversation_context: str = "") -> FollowUpQuestions:
- """
- 异步生成后续问题
-
- Args:
- query: 用户原始问题
- chat_history: 历史对话记录
- conversation_context: 会话背景信息
-
- Returns:
- 后续问题结果
- """
- follow_up_start_time = time.time()
- # 准备提示词
- follow_up_parser = PydanticOutputParser(pydantic_object=FollowUpQuestions)
- formatted_prompt = follow_up_questions_prompt.format(
- query=query,
- chat_history=json.dumps(chat_history, ensure_ascii=False) if chat_history else "[]",
- # conversation_context=conversation_context,
- output_format=follow_up_parser.get_format_instructions()
- )
-
- try:
- # 异步调用LLM
- response = await self._llm.invoke_async(formatted_prompt, False)
-
- # 解析输出
- response.content = response.content.strip()
- clean_output = re.sub(r'.*?', '', response.content, flags=re.DOTALL)
- parsed_output = follow_up_parser.parse(clean_output)
- follow_up_end_time = time.time()
- follow_up_time = follow_up_end_time - follow_up_start_time
- logging.debug(f"异步后续问题生成耗时统计 - 总耗时: {follow_up_time:.2f}秒")
- return parsed_output
- except Exception as e:
- # 如果解析失败，返回原始查询作为后续问题
- logging.error(f"异步后续问题生成失败: {e}", exc_info=True)
- return FollowUpQuestions(original_query=query, follow_up_query=query)
-
async def _find_matching_software_docs_async(self, query: str, soft_name: str,
chat_history: List[Dict[str, str]] = None,
top_k: int = 3) -> List[str]:
@@ -771,7 +708,7 @@ class AsyncIntentRecognizer:
json_response = json_parser.parse(response.content)
for match in json_response:
wiki_names.append(match["content"])
- logging.debug(f"软件文档匹配耗时: {end_time - start_time:.2f}秒")
+ logging.info(f"软件文档匹配耗时: {end_time - start_time:.2f}秒")
return wiki_names
except json.JSONDecodeError as e:
@@ -781,43 +718,4 @@ class AsyncIntentRecognizer:
except Exception as e:
logging.error(f"查找匹配软件文档时出错: {e}", exc_info=True)
# 出错时返回空列表
- return []
-
- async def _generate_multi_questions_async(self, query: str, chat_history: List[Dict[str, str]] = None, conversation_context: str = "") -> MultiQuestions:
- """
- 异步生成多角度问题
-
- Args:
- query: 用户原始问题
- chat_history: 历史对话记录
- conversation_context: 会话背景信息
-
- Returns:
- 多角度问题结果
- """
- multi_questions_start_time = time.time()
- # 准备提示词
- multi_questions_parser = PydanticOutputParser(pydantic_object=MultiQuestions)
- formatted_prompt = multi_questions_prompt.format(
- query=query,
- chat_history=json.dumps(chat_history, ensure_ascii=False) if chat_history else "[]",
- # conversation_context=conversation_context,
- output_format=multi_questions_parser.get_format_instructions()
- )
-
- try:
- # 异步调用LLM
- response = await self._llm.invoke_async(formatted_prompt, False)
-
- # 解析输出
- response.content = response.content.strip()
- clean_output = re.sub(r'.*?', '', response.content, flags=re.DOTALL)
- parsed_output = multi_questions_parser.parse(clean_output)
- multi_questions_end_time = time.time()
- multi_questions_time = multi_questions_end_time - multi_questions_start_time
- logging.debug(f"异步多角度问题生成耗时统计 - 总耗时: {multi_questions_time:.2f}秒")
- return parsed_output
- except Exception as e:
- # 如果解析失败，返回原始查询作为唯一子问题
- logging.error(f"异步多角度问题生成失败: {e}", exc_info=True)
- return MultiQuestions(original_query=query, sub_questions=[query])
\ No newline at end of file
+ return []
\ No newline at end of file
diff --git a/start_AnswerType.sh b/start_AnswerType.sh
old mode 100644
new mode 100755
index 0f57fb2..27f6b79
--- a/start_AnswerType.sh
+++ b/start_AnswerType.sh
@@ -4,14 +4,14 @@
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
# 检查是否已经存在名为AnswerType的screen会话
-if screen -ls | grep "AnswerType"; then
+if screen -ls | grep -q "\.AnswerType\s"; then
echo "Screen session 'AnswerType' already exists."
else
# 启动一个名为AnswerType的screen会话，并在其中执行后续命令
- screen -dmS AnswerType bash -c '
- cd $SCRIPT_DIR
+ screen -dmS AnswerType bash -c "
+ cd \"$SCRIPT_DIR\"
uv run uvicorn rag2_0.dify.AnswerType:app --host 0.0.0.0 --port 8003 --workers 1
- '
+ "
# 输出提示信息
echo "Started screen session 'AnswerType' and executed the command."
diff --git a/start_DifyQueryRetrieval_api.sh b/start_DifyQueryRetrieval_api.sh
old mode 100644
new mode 100755
index bc5cec1..072177b
--- a/start_DifyQueryRetrieval_api.sh
+++ b/start_DifyQueryRetrieval_api.sh
@@ -4,12 +4,12 @@
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
# 检查是否已经存在名为DifyQueryRetrieval_api的screen会话
-if screen -ls | grep "DifyQueryRetrieval_api"; then
+if screen -ls | grep -q "DifyQueryRetrieval_api"; then
echo "Screen session 'DifyQueryRetrieval_api' already exists."
else
# 启动一个名为DifyQueryRetrieval_api的screen会话，并在其中执行后续命令
screen -dmS DifyQueryRetrieval_api bash -c '
- cd $SCRIPT_DIR
+ cd \"$SCRIPT_DIR\"
uv run uvicorn rag2_0.dify.DifyQueryRetrieval_api:app --host 0.0.0.0 --port 8002 --workers 25
'
diff --git a/start_intent_recognition_api.sh b/start_intent_recognition_api.sh
index 0f48a44..3f49dd8 100755
--- a/start_intent_recognition_api.sh
+++ b/start_intent_recognition_api.sh
@@ -7,11 +7,11 @@ SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
if screen -ls | grep "intent_recognition_api"; then
echo "Screen session 'intent_recognition_api' already exists."
else
- # 启动一个名为xinference的screen会话，并在其中执行后续命令
- screen -dmS intent_recognition_api bash -c '
- cd $SCRIPT_DIR
+ # 启动一个名为intent_recognition_api的screen会话，并在其中执行后续命令
+ screen -dmS intent_recognition_api bash -c "
+ cd \"$SCRIPT_DIR\"
uv run uvicorn rag2_0.dify.intent_recognition_api:app --host 0.0.0.0 --port 8001 --workers 25
- '
+ "
# 输出提示信息
echo "Started screen session 'intent_recognition_api' and executed the command."