health model

.gitignore

@@ -0,0 +1,2 @@
+.env
+.env.example

app.py

@@ -3,7 +3,8 @@ from langchain.memory import ConversationBufferMemory
 from langchain.chains import LLMChain
 from langchain_community.llms import HuggingFaceHub
 from langchain.prompts import PromptTemplate
-
+from dotenv import load_dotenv
+import time
 
 def get_response(model, query):
     prompt_template = PromptTemplate(
@@ -12,27 +13,32 @@ def get_response(model, query):
         )
     # get the response
     memory = ConversationBufferMemory(memory_key="messages", return_messages=True)
+    print(memory)
     conversation_chain = LLMChain(
         llm=model,
         prompt=prompt_template,
         # retriever=vectorstore.as_retriever(),
         memory=memory)
-    # response = conversation_chain.invoke(query)
-    # response = response["result"]
-    # answer = response.split('\nHelpful Answer: ')[1]
     response = conversation_chain.invoke(query)
-    return response
+    answer = response["text"]
+    if "\n\n" in answer:
+        answer = answer.split("\n\n", 1)[1]
+    return answer
 
 def main():
     st.title("Health Chatbot")
-    
+    # load the environment variables    
+    load_dotenv()
     print("Loading LLM from HuggingFace")
     with st.spinner('Loading LLM from HuggingFace...'):
-        llm = HuggingFaceHub(repo_id="HuggingFaceH4/zephyr-7b-beta", model_kwargs={"temperature":0.7, "max_new_tokens":512, "top_p":0.95, "top_k":50})
+        # llm = HuggingFaceHub(repo_id="HuggingFaceH4/zephyr-7b-beta", model_kwargs={"temperature":0.7, "max_new_tokens":1028, "top_p":0.95})
+
+        llm = HuggingFaceHub(repo_id="epfl-llm/meditron-70b", model_kwargs={"temperature":0.7, "max_new_tokens":1028, "top_p":0.95})
     
     if "messages" not in st.session_state:
         st.session_state.messages = []
         
+        
     if st.button("Clear Chat"):
         st.session_state.messages = []
     
@@ -47,7 +53,9 @@ def main():
         st.chat_message("user").markdown(user_prompt)
         st.session_state.messages.append({"role": "user", "content": user_prompt})
         with st.spinner('Thinking...'):
+            start_time = time.time()
             response = get_response(llm, user_prompt)
+            st.write("Response Time: ", time.time() - start_time)
         st.chat_message("bot").markdown(response)
         st.session_state.messages.append({"role": "bot", "content": response})