Update app.py

app.py

@@ -107,11 +107,10 @@ def query_hybrid_search(query: str, client: QdrantClient, collection_name: str,
         score_threshold=0.9
     )
 
-@outlines.prompt
-def build_prompt_conv(user_input: str):
-    """Generate a short, single-sentence summary of the user's intent or topic based on their question, capturing the main focus of what they want to discuss.
+def build_prompt_conv():
+    return f"""Generate a short, single-sentence summary of the user's intent or topic based on their question, capturing the main focus of what they want to discuss.
     
-    Question : {{ user_input }}
+    Question : {st.session_state.user_input}
     """
     
 @outlines.prompt
@@ -158,28 +157,28 @@ def open_query_prompt(past_messages: str, query: str):
 
 @outlines.prompt
 def route_llm(context: str, query: str):
-    """Understand the context and answer the question by 'Yes' or 'No'
+    """Based on the given context and user query, decide if the context is relevant. Return 'Yes' for relevant and 'No' for irrelevant.
 
-    Context: {{ context }}
+    Context : {{ context }}
 
-    Question: {{ query }}
+    Query: {{ query }}
     """
 
 @outlines.prompt
 def answer_with_context(context: str, query: str):
-    """You are an assistant for question-answering tasks. Use the following pieces of retrieved context to answer the question. Use three sentences maximum and keep the answer concise.
-    
-    Context: {{ context }} 
-    
-    Question: {{ query }} 
-            
+    """Context information is below.
+    ---------------------
+    {context}
+    ---------------------
+    Given the context information and not prior knowledge, answer the query.
+    Query: {query}
     Answer:
     """
 
 @outlines.prompt
 def idk(query: str):
-    """Just say that you don't know have the knowledge to answer the following question
-
+    """When you encounter a question that falls outside your knowledge or expertise, respond in a way that politely conveys you don't have the information needed to answer.
+    
     Question: {{ query }}
     """
 
@@ -502,7 +501,7 @@ def load_models_and_documents():
     time.sleep(5)
     container.empty()
     
-    return client, collection_name, model, dense_model, sparse_model, nlp, conn, cursor
+    return client, collection_name, llm, model, dense_model, sparse_model, nlp, conn, cursor
 
 def chunk_documents(texts: List[str], metadatas: List[dict], dense_model: OptimumEncoder, sparse_model: SparseTextEmbedding):
     text_splitter = SemanticChunker(
@@ -549,7 +548,7 @@ def on_change_documents_only():
 if __name__ == '__main__':
     st.set_page_config(page_title="Multipurpose AI Agent",layout="wide", initial_sidebar_state='auto')
 
-    client, collection_name, llm, dense_model, sparse_model, nlp, conn, cursor = load_models_and_documents()
+    client, collection_name, llm, model, dense_model, sparse_model, nlp, conn, cursor = load_models_and_documents()
 
     styles = {
         "nav": {
@@ -728,17 +727,13 @@ if __name__ == '__main__':
             )
 
         def generate_conv_title(llm):
-            if st.session_state.chat_id == 'New Conversation':               
-                gen_conv = outlines.generate.text(llm)
-                prompt = build_prompt_conv(st.session_state.user_input)
-                print(f'USER INPUT : {prompt}')
-                output = gen_conv(
-                    prompt,
-                    max_tokens=10,
-                    sampling_params=SamplingParams(temperature=0.3)
+            if st.session_state.chat_id == 'New Conversation':
+                output = llm.generate(
+                    build_prompt_conv(),
+                    SamplingParams(temperature=0.3, max_tokens=10)
                 )
-                print(f'OUTPUT : {output}')
-                st.session_state.chat_id = output
+                print(f'OUTPUT : {output[0].outputs[0].text}')
+                st.session_state.chat_id = output[0].outputs[0].text
                 st.session_state.messages = []
 
                 torch.cuda.empty_cache()
@@ -761,7 +756,7 @@ if __name__ == '__main__':
             st.chat_message("user").markdown(prompt)
             st.session_state.messages.append({"role": "user", "content": prompt})
             
-            ai_response = main(prompt, client, collection_name, llm, dense_model, sparse_model, "\n".join([f'{msg["role"]}: {msg["content"]}' for msg in st.session_state.messages]))
+            ai_response = main(prompt, client, collection_name, model, dense_model, sparse_model, "\n".join([f'{msg["role"]}: {msg["content"]}' for msg in st.session_state.messages]))
             with st.chat_message("assistant"):
                 message_placeholder = st.empty()
                 full_response = ""