remove comments, add MAX_CHAT_TURNS, and other comments orienting user

app.py

@@ -1,5 +1,4 @@
 import os
-import time
 import threading
 import streamlit as st
 from itertools import tee
@@ -17,8 +16,8 @@ if DATABRICKS_TOKEN is None:
     raise ValueError("DATABRICKS_TOKEN environment variable must be set")
 
 MODEL_AVATAR_URL= "./VU.jpeg"
-
-# MSG_MAX_TURNS_EXCEEDED = f"Sorry! The Vanderbilt AI assistant playground is limited to {MAX_CHAT_TURNS} turns. Click the 'Clear Chat' button or refresh the page to start a new conversation."
+MAX_CHAT_TURNS = 5 # limit this for preliminary testing
+MSG_MAX_TURNS_EXCEEDED = f"Sorry! The Vanderbilt AI assistant playground is limited to {MAX_CHAT_TURNS} turns in a single history. Click the 'Clear Chat' button or refresh the page to start a new conversation."
 # MSG_CLIPPED_AT_MAX_OUT_TOKENS = "Reached maximum output tokens for DBRX Playground"
 
 EXAMPLE_PROMPTS = [
@@ -42,6 +41,15 @@ We hope to gradually improve this AI assistant to create a large-scale, all-incl
 
 GENERAL_ERROR_MSG = "An error occurred. Please refresh the page to start a new conversation."
 
+# # To prevent streaming too fast, chunk the output into TOKEN_CHUNK_SIZE chunks
+TOKEN_CHUNK_SIZE = 2 # test this number
+# if TOKEN_CHUNK_SIZE_ENV is not None:
+#     TOKEN_CHUNK_SIZE = int(TOKEN_CHUNK_SIZE_ENV)
+
+QUEUE_SIZE = 20 # maximize this value for adding enough places in the global queue?
+# if QUEUE_SIZE_ENV is not None:
+#     QUEUE_SIZE = int(QUEUE_SIZE_ENV)
+
 # @st.cache_resource
 # def get_global_semaphore():
 #     return threading.BoundedSemaphore(QUEUE_SIZE)
@@ -49,13 +57,8 @@ GENERAL_ERROR_MSG = "An error occurred. Please refresh the page to start a new c
 
 st.set_page_config(layout="wide")
 
-# # To prevent streaming to fast, chunk the output into TOKEN_CHUNK_SIZE chunks
-TOKEN_CHUNK_SIZE = 1
-# if TOKEN_CHUNK_SIZE_ENV is not None:
-#     TOKEN_CHUNK_SIZE = int(TOKEN_CHUNK_SIZE_ENV)
-
 st.title(TITLE)
-# st.image("sunrise.jpg", caption="Sunrise by the mountains") # add a Vanderbilt related picture to the head of our Space!
+# st.image("sunrise.jpg", caption="Sunrise by the mountains") # TODO add a Vanderbilt related picture to the head of our Space!
 st.markdown(DESCRIPTION)
 st.markdown("\n")
 
@@ -96,42 +99,8 @@ def get_stream_warning_error(stream):
     return warning, error
 
 # @retry(wait=wait_random_exponential(min=0.5, max=2), stop=stop_after_attempt(3))
-def chain_call(history):
-    # *** original code for instantiating the DBRX model through the OpenAI client *** skip this and introduce our chain eventually
-    # extra_body = {}
-    # if SAFETY_FILTER:
-    #     extra_body["enable_safety_filter"] = SAFETY_FILTER
-    # chat_completion = client.chat.completions.create(
-    #     messages=[
-    #         {"role": m["role"], "content": m["content"]}
-    #         for m in history
-    #     ],
-    #     model="databricks-dbrx-instruct",
-    #     stream=True,
-    #     max_tokens=MAX_TOKENS,
-    #     temperature=0.7,
-    #     extra_body= extra_body
-    # )
-
-    # *** can we stream the chain's response by incorporating the above OpenAI streaming functionality?
-    # *** Look back at the predict_stream function and see if we can incorporate that!
-    # *** looks like we want to use either chain.stream() or chain.astream()
-    # test first with invoke
-
-    input_example = {'messages': 
-                     [{'content': 'What does EDW stand for?', 'role': 'user'}, 
-                      {'content': 'Enterprise Data Warehouse.', 'role': 'assistant'}, 
-                      {'content': 'Thank you. What is the data lake?', 'role': 'user'}, 
-                      {'content': 'A data lake is a centralized repository of structured and unstructured data. It allows data to be stored in its native state, without the need for transformations, so that it can be consumed by other users later. It is not just a term for storage, but also covers functionalities required for a platform, including data analysis, machine learning, cataloging and data movement.', 'role': 'assistant'}, 
-                      {'content': 'Can you tell me more about how they are used?', 'role': 'user'}, 
-                      {'content': 'At Vanderbilt University Medical Center, a data lake is used as a centralized repository for storing and managing large amounts of data in its native format. This allows for the data to be easily accessed and analyzed by different teams and business units within the organization. The data lake also provides functionalities such as data analysis, machine learning, cataloging and data movement, making it a versatile tool for handling diverse data sets.\n\nAn Enterprise Data Warehouse (EDW) is used for executing analytic queries on structured data. It is optimized for this purpose, with data being stored in a way that allows for efficient querying and analysis. This makes it a useful tool for teams that need to perform complex analyses on large data sets.\n\nA data mart is a specific organizational structure or pattern used in the context of data warehouses. It is a layer that has specific subdivisions for each business unit or team, such as finance, marketing, and product. This allows users to consume data in a format that meets their specific needs.\n\nA data lakehouse is a term used to describe approaches that attempt to combine the data structure and management features of a data warehouse with the low cost of storage of a data lake. This includes a structured transactional layer, which allows for efficient querying and analysis of data. This approach aims to provide the benefits of both data lakes and data warehouses in a single platform.', 'role': 'assistant'}, 
-                      {'content': 'Nice answer. Can you tell me what the HCERA is?', 'role': 'user'}]}
-    
+def chain_call(history):    
     input = {'messages': [{"role": m["role"], "content": m["content"]} for m in history]}
-
-    # search_result = vector_store.similarity_search(query=st.session_state["messages"][-1]["content"], k=5)
-    # chat_completion = search_result # TODO update this after we implement our chain
-    # chat_completion = chain.invoke(input_example) # *** TODO here we will pass only the chat history, the chain handles the system prompt
     chat_completion = chain.stream(input)
     return chat_completion
 
@@ -150,18 +119,13 @@ def write_response():
     return response, stream_warning, stream_error
 
 def chat_completion(messages):
-    # history_dbrx_format = [
-    #     {"role": "system", "content": SYSTEM_PROMPT} # no longer need this because the chain handles all of this for us
-    # ]
-    
-    # history_dbrx_format = history_dbrx_format + messages
-    # if (len(history_dbrx_format)-1)//2 >= MAX_CHAT_TURNS:
-    #     yield {"content": None, "error": MSG_MAX_TURNS_EXCEEDED, "warning": None}
-    #     return
+    if (len(messages)-1)//2 >= MAX_CHAT_TURNS:
+        yield {"content": None, "error": MSG_MAX_TURNS_EXCEEDED, "warning": None}
+        return
 
     chat_completion = None
     error = None
-    # *** original code for querying DBRX through the OpenAI cleint for chat completion
+    # *** TODO add code for implementing a global queue with a bounded semaphore?  
     # wait to be in queue
     # with global_semaphore:
     #     try: 
@@ -169,7 +133,7 @@ def chat_completion(messages):
     #     except Exception as e:
     #         error = e    
     # chat_completion = chain_call(history_dbrx_format)
-    chat_completion = chain_call(messages) # simply pass the old messages, need not worry about the system prompt
+    chat_completion = chain_call(messages)
     if error is not None:
         yield {"content": None, "error": GENERAL_ERROR_MSG, "warning": None}
         print(error)
@@ -179,13 +143,8 @@ def chat_completion(messages):
     partial_message = ""
     chunk_counter = 0
     for chunk in chat_completion:
-        # if chunk.choices[0].delta.content is not None:
-        # TODO *** we need to refactor this logic to match what happens with the response from our chain - it should be strings or an iterator of strings
-        # if chunk.page_content is not None:
         if chunk is not None:
             chunk_counter += 1
-            # partial_message += chunk.choices[0].delta.content
-            # partial_message += f"* {chunk.page_content} [{chunk.metadata}]"
             partial_message += chunk
             if chunk_counter % TOKEN_CHUNK_SIZE == 0:
                 chunk_counter = 0
@@ -218,34 +177,16 @@ def handle_user_input(user_input):
 def feedback():
     with st.form("feedback_form"):
         st.title("Feedback Form")
-        # sentiment_mapping = [":material/thumb_down:", ":material/thumb_up:"]
-        # rating = None
-        # while not rating:
+        st.markdown("Please select either 👍 or 👎 before providing a reason for your review of the most recent response. Dont forget to click submit!")
         rating = st.feedback()
-        # feedback = st.text_input(f"Please detail your rationale for choosing {sentiment_mapping[rating]}: ", "")
         feedback = st.text_input("Please detail your feedback: ")
-        # feedback = ""
-        # review = {}
-        # if rating is not None:
-        #     # st.markdown(f"You selected: {sentiment_mapping[rating]}")
-        #     # rating = st.radio("Rate your experience:", ["👍", "Neutral", "👎"])
-        # review = {"rating": {rating}, "feedback": {feedback}}
+        # implement a method for writing these responses to storage!
         submitted = st.form_submit_button("Submit Feedback")
-        # if submitted:
-        #     st.write(f"The feedback was: {sentiment_mapping[rating]} : {feedback}")
-
-        # st.markdown(review)
-        # time.sleep(5)
-        #     # Save the feedback data
-        # if st.button("Submit"):
-        #     with open("feedback.json", "a") as f:
-        #         f.write()
-        #     st.write("Thank you for your feedback!")
 
 main = st.container()
 with main:
     if st.session_state["feedback"][-1] is not None: # TODO clean this up in a fn?
-            st.markdown("Thank you! Feedback received! Type a new message to continue.")
+            st.markdown("Thank you! Feedback received! Type a new message to continue your conversation.")
     history = st.container(height=400)
     with history:
         for message in st.session_state["messages"]:
@@ -255,9 +196,6 @@ with main:
             with st.chat_message(message["role"], avatar=avatar):
                 if message["content"] is not None:
                     st.markdown(message["content"])
-                    # receive feedback on AI outputs if the user feels inclined to give it
-                    # rating = st.radio("Rate your experience:", ["Very satisfied", "Somewhat satisfied", "Neutral", "Somewhat dissatisfied", "Very dissatisfied"])
-                    # st.button("Provide Feedback", on_click=feedback)
                 if message["error"] is not None:
                     st.error(message["error"],icon="🚨")
                 if message["warning"] is not None:
@@ -266,15 +204,11 @@ with main:
     if prompt := st.chat_input("Type a message!", max_chars=5000):
         handle_user_input(prompt)
     st.markdown("\n") #add some space for iphone users
-    # with st.container():
     gave_feedback = st.button('Give Feedback on Last Response', on_click=feedback)
     if gave_feedback: # TODO clean up the conditions here with a function
         st.session_state["feedback"].append("given") 
     else: 
         st.session_state["feedback"].append(None)
-        # st.markdown("Feedback received! Thank you for your insight.")
-    # time.sleep(3)
-    # add st.session_state["feedback"] var for keeping track of when the user gives feedback!
         
 
 with st.sidebar: