direct access to hf pipeline

app.py

@@ -13,6 +13,7 @@ from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
 # gpt_model = 'gpt-4-1106-preview'
 # embedding_model = 'text-embedding-3-small'
 default_model_id = "bigcode/starcoder2-3b"
+#default_model_id = "tiiuae/falcon-7b-instruct"
 
 def init():
     if "conversation" not in st.session_state:
@@ -38,7 +39,7 @@ def init_llm_pipeline(model_id):
         task="text-generation",
         max_new_tokens=1024
         )
-        st.session_state.llm = HuggingFacePipeline(pipeline=text_generation_pipeline)          
+        st.session_state.llm = text_generation_pipeline         
 
 def get_retriever(files):
     documents = [doc.getvalue().decode("utf-8") for doc in files]
@@ -58,17 +59,23 @@ def get_retriever(files):
     return retriever
     
 def get_conversation(retriever):
-    memory = ConversationBufferMemory(memory_key='chat_history', return_messages=True)
+    #memory = ConversationBufferMemory(memory_key='chat_history', return_messages=True)
+
     conversation_chain = ConversationalRetrievalChain.from_llm(
+        prompt=prompt,
         llm=st.session_state.llm,
-        retriever=retriever,
-        memory = memory   
+        retriever=retriever   
     )
     return conversation_chain
 
+def getprompt(user_input):
+    prompt = f"You are a helpful assistant. Please answer the user question. USER: {user_input} ASSISTANT:"
+    return prompt
+
 def handle_user_input(question):
-    response = st.session_state.conversation({'question':question})
-    st.session_state.chat_history = response['chat_history']
+    st.session_state.chat_history += {"role":"user","content":question}
+    response = st.session_state.llm(getprompt(question))
+    st.session_state.chat_history += {"role":"assistant","content":response}
     for i, message in enumerate(st.session_state.chat_history):
         if i % 2 == 0:
             with st.chat_message("user"):