history init

application/app.py

@@ -23,6 +23,7 @@ from langchain.prompts.chat import (
     ChatPromptTemplate,
     SystemMessagePromptTemplate,
     HumanMessagePromptTemplate,
+    AIMessagePromptTemplate,
 )
 from pymongo import MongoClient
 from werkzeug.utils import secure_filename
@@ -107,6 +108,8 @@ def run_async_chain(chain, question, chat_history):
     result["answer"] = answer
     return result
 
+    
+
 
 @celery.task(bind=True)
 def ingest(self, directory, formats, name_job, filename, user):
@@ -164,16 +167,6 @@ def api_answer():
             docsearch = FAISS.load_local(vectorstore, CohereEmbeddings(cohere_api_key=embeddings_key))
 
         # create a prompt template
-        if history:
-            history = json.loads(history)
-            template_temp = template_hist.replace("{historyquestion}", history[0]).replace("{historyanswer}",
-                                                                                           history[1])
-            c_prompt = PromptTemplate(input_variables=["summaries", "question"], template=template_temp,
-                                      template_format="jinja2")
-        else:
-            c_prompt = PromptTemplate(input_variables=["summaries", "question"], template=template,
-                                      template_format="jinja2")
-
         q_prompt = PromptTemplate(input_variables=["context", "question"], template=template_quest,
                                   template_format="jinja2")
         if settings.LLM_NAME == "openai_chat":
@@ -182,6 +175,18 @@ def api_answer():
                 SystemMessagePromptTemplate.from_template(chat_combine_template),
                 HumanMessagePromptTemplate.from_template("{question}")
             ]
+            if history:
+                tokens_current_history = 0
+                tokens_max_history = 1000
+                #count tokens in history
+                for i in history:
+                    if "prompt" in i and "response" in i:
+                        tokens_batch = llm.get_num_tokens(i["prompt"]) + llm.get_num_tokens(i["response"])
+                        if tokens_current_history + tokens_batch < tokens_max_history:
+                            tokens_current_history += tokens_batch
+                            messages_combine.append(HumanMessagePromptTemplate.from_template(i["prompt"]))
+                            messages_combine.append(SystemMessagePromptTemplate.from_template(i["response"]))
+
             p_chat_combine = ChatPromptTemplate.from_messages(messages_combine)
         elif settings.LLM_NAME == "openai":
             llm = OpenAI(openai_api_key=api_key, temperature=0)

frontend/src/conversation/conversationApi.ts

@@ -7,6 +7,7 @@ export function fetchAnswerApi(
   question: string,
   apiKey: string,
   selectedDocs: Doc,
+  history: Array<any> = [],
 ): Promise<Answer> {
   let namePath = selectedDocs.name;
   if (selectedDocs.language === namePath) {
@@ -37,7 +38,7 @@ export function fetchAnswerApi(
       question: question,
       api_key: apiKey,
       embeddings_key: apiKey,
-      history: localStorage.getItem('chatHistory'),
+      history: history,
       active_docs: docPath,
     }),
   })

frontend/src/conversation/conversationSlice.ts

@@ -19,6 +19,7 @@ export const fetchAnswer = createAsyncThunk<
     question,
     state.preference.apiKey,
     state.preference.selectedDocs!,
+    state.conversation.queries,
   );
   return answer;
 });