chore: add handle multple user

src/agents_framework/integrations.py

@@ -12,10 +12,11 @@ import datetime
 
 from .tools import (
     load_retrieval_tool,
-    classifer_question_execution,
     search_searxng_engine,
     generate_response_base_context,
-    openai_client_answer
+    openai_client_answer,
+    data_prompts,
+    create_execution_w_custom_message
 )
 from api.schemas import OutputLocation
 from .utils import get_data
@@ -26,18 +27,6 @@ from typing import List, Dict, Tuple, Union
 OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
 
 
-paths = {
-    "question_type_classification": "data/prompts/question_type_classification.txt",
-    "questions_generation": "data/prompts/questions_generation.txt",
-    "normal_conversation": "data/prompts/normal_conversation.txt",
-    "rephrase_query": "data/prompts/rephrase_query.txt",
-    "interpreter_web": "data/prompts/interpreter_web.txt",
-    
-}
-
-# Load the data
-data = get_data(paths)
-
 
 
 class AgentIntegrationService:
@@ -63,7 +52,7 @@ class AgentIntegrationService:
                 query, 
                 chat_history, 
                 model="gpt-3.5-turbo", 
-                template=data["rephrase_query"]
+                template=data_prompts["rephrase_query"]
             )
             if query == 'not_need': query_reformat = query
             answer = agent_using(query_reformat)
@@ -114,34 +103,37 @@ class AgentIntegrationService:
             context += result['snippet'] + "\n"
         date = datetime.datetime.now().isoformat()
         print(context)
-        answer = generate_response_base_context(context.strip(), date=date, template=data['interpreter_web'])
+        answer = generate_response_base_context(context.strip(), date=date, template=data_prompts['interpreter_web'])
         # print(answer.content)
         return answer.content
 
     
     def answer_normal(self, user_input: str, chat_history, model_name="gpt-3.5-turbo") -> str:
         """Extract the token name or name from the user input"""
-        set_conversation = [("system", data["normal_conversation"])]
+
+        messages = [{"role": "system", "content": data_prompts['normal_conversation']}]
         if chat_history and len(chat_history) > 0:
-            for history in chat_history:
-                set_conversation.append(("human", history['question']))
-                set_conversation.append(("system", history['answer']))
-        set_conversation.append(("human", "{input}"))
-        final_prompt = ChatPromptTemplate.from_messages(set_conversation) | ChatOpenAI(model=model_name, temperature=0)
-        return final_prompt.invoke(user_input)
+            for mes in chat_history:
+                messages.append({"role": "user", "content": mes['question']})
+                messages.append({"role": "assistant", "content": mes['answer']})
+        messages.append({"role": "user", "content": user_input})
+
+        answer = create_execution_w_custom_message(messages).content
+
+        return answer
 
 
     def create_questions(self, user_input: str, chat_history) -> str:
         """Create questions from the user input"""
-        set_conversation = [("system", data['questions_generation'])]
+        messages = [{"role": "system", "content": data_prompts['questions_generation']}]
         if chat_history:
             for mes in chat_history:
-                set_conversation.append(("human", mes['question']))
-                set_conversation.append(("ai", mes['answer']))
-        set_conversation.append(("human", "{input}"))
+                messages.append({"role": "user", "content": mes['question']})
+                messages.append({"role": "assistant", "content": mes['answer']})
+        messages.append({"role": "user", "content": user_input})
+
+        answer = create_execution_w_custom_message(messages).content
 
-        create_questions_execution = ChatPromptTemplate.from_messages(set_conversation) | ChatOpenAI(model="gpt-4o", temperature=0)
-        answer = create_questions_execution.invoke({"input": user_input}).content
         print(answer)
         data_loaded = yaml.safe_load(answer.replace("```", "").replace("yaml", ""))
         return data_loaded

src/agents_framework/tools.py

@@ -13,6 +13,12 @@ from langchain.retrievers import ParentDocumentRetriever
 from langchain_core.prompts import ChatPromptTemplate
 from langchain_community.utilities import SearxSearchWrapper
 
+from tenacity import (
+    retry,
+    stop_after_attempt,
+    wait_random_exponential,
+)  # for exponential backoff
+
 from .utils import get_data
 
 OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
@@ -33,18 +39,31 @@ PROJECT_ROOT = "./"  # insert your project root directory name here
 
 
 paths = {
+    "normal_conversation": "data/prompts/normal_conversation.txt",
     "question_type_classification": "data/prompts/question_type_classification.txt",
     "questions_generation": "data/prompts/questions_generation.txt",
-    "interpreter_web": "data/prompts/interpreter_web.txt"
+    "interpreter_web": "data/prompts/interpreter_web.txt",
+    "rephrase_query": "data/prompts/rephrase_query.txt",
 }
 
 # Load the data
-data = get_data(paths)
+data_prompts = get_data(paths)
 current_date = datetime.datetime.now().isoformat()
 
 # Initial search api
 searxng_api = SearxSearchWrapper(searx_host=SEARXNG_PORT, k = 20)
 
+
+
+
+
+
+
+
+
+
+
+
 def load_retrieval_tool(embedding_model):
     parent_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=0)
     child_splitter = RecursiveCharacterTextSplitter(chunk_size=200, chunk_overlap=0)
@@ -68,6 +87,7 @@ def load_retrieval_tool(embedding_model):
     return retriever_tool
 
 
+@retry(wait=wait_random_exponential(min=1, max=60), stop=stop_after_attempt(6))
 def search_searxng_engine(query, engines: list=["google"], enabled_engines: list=["google", "apple_maps"]):
     results = searxng_api.results(
         query,
@@ -80,6 +100,7 @@ def search_searxng_engine(query, engines: list=["google"], enabled_engines: list
     return results
 
 
+@retry(wait=wait_random_exponential(min=1, max=60), stop=stop_after_attempt(6))
 def openai_client_answer(query, history=None, model="gpt-3.5-turbo", template=""):
     chat_completion = openai_client.chat.completions.create(
         messages=[
@@ -93,6 +114,7 @@ def openai_client_answer(query, history=None, model="gpt-3.5-turbo", template=""
     return chat_completion.choices[0].message.content 
 
 
+@retry(wait=wait_random_exponential(min=1, max=60), stop=stop_after_attempt(6))
 def generate_response_base_context(context, date, template, model="gpt-4o"):
     chat_completion = openai_client.chat.completions.create(
         messages=[
@@ -106,20 +128,42 @@ def generate_response_base_context(context, date, template, model="gpt-4o"):
     return chat_completion.choices[0].message
 
 
-classifer_question_execution = ChatPromptTemplate.from_messages(
-    [
-        ("system", data['question_type_classification']),
-        ("human", "{input}"),
-    ]
-) | ChatOpenAI(model="gpt-3.5-turbo", temperature=0)
+@retry(wait=wait_random_exponential(min=1, max=60), stop=stop_after_attempt(6))
+def create_execution_wout_edit_template(input_text, template, model="gpt-3.5-turbo"):
+    chat_completion = openai_client.chat.completions.create(
+        messages=[
+            {"role": "system", "content": template},
+            {"role": "user", "content": input_text}
+        ],
+        model=model,
+        temperature=0
+    )
+    return chat_completion.choices[0].message
+
+
+@retry(wait=wait_random_exponential(min=1, max=60), stop=stop_after_attempt(6))
+def create_execution_w_custom_message(messages: list, model="gpt-3.5-turbo"):
+    chat_completion = openai_client.chat.completions.create(
+        messages=messages,
+        model=model,
+        temperature=0
+    )
+    return chat_completion.choices[0].message
+
+# classifer_question_execution = ChatPromptTemplate.from_messages(
+#     [
+#         ("system", data['question_type_classification']),
+#         ("human", "{input}"),
+#     ]
+# ) | ChatOpenAI(model="gpt-3.5-turbo", temperature=0)
 
 
-create_questions_execution = ChatPromptTemplate.from_messages(
-    [
-        ("system", data['questions_generation']),
-        ("human", "{input}"),
-    ]
-) | ChatOpenAI(model="gpt-4o", temperature=0)
+# create_questions_execution = ChatPromptTemplate.from_messages(
+#     [
+#         ("system", data_prompts['questions_generation']),
+#         ("human", "{input}"),
+#     ]
+# ) | ChatOpenAI(model="gpt-4o", temperature=0)
 
 
 if __name__ == "__main__":
@@ -134,7 +178,7 @@ if __name__ == "__main__":
 
 
     def generate_response(context: str) -> str:
-        prompt = data["interpreter_web"].format(context=context, date=current_date)
+        prompt = data_prompts["interpreter_web"].format(context=context, date=current_date)
         print(prompt)
         location_answer_execution = ChatPromptTemplate.from_messages(
             [

src/api/routes.py

@@ -3,7 +3,7 @@ from fastapi import APIRouter, HTTPException, Depends
 from fastapi.security import APIKeyHeader
 
 from agents.models import service
-from agents_framework.tools import classifer_question_execution
+from agents_framework.tools import create_execution_wout_edit_template, data_prompts
 from .schemas import MessageInput ,ChatAgentResponse, OutputLocation
 
 
@@ -70,9 +70,8 @@ async def chat_completion(
     # return {"text": text, "ids_location": ids_location}
 
 def classify_question_type(query: str): 
-
-    question_type = classifer_question_execution.invoke(query).content.replace('`', "").replace("Tags", "").strip()
-    # log.info(f"question_type: {question_type}")
+    question_type = create_execution_wout_edit_template(query, data_prompts["question_type_classification"]).content.replace('`', "").replace("Tags", "").strip()
+    # question_type = classifer_question_execution.invoke(query).content.replace('`', "").replace("Tags", "").strip()
     print(question_type)
     return question_type