Update app.py

app.py

@@ -1,14 +1,13 @@
-# AI MAKERSPACE PREPR 
+# AI MAKERSPACE PREPR
 # Date: 2024-5-16
 
 # Basic Imports & Setup
 import os
-from openai import AsyncOpenAI
+from transformers import AutoModelForCausalLM, AutoTokenizer
 
 # Using Chainlit for our UI
 import chainlit as cl
 from chainlit.prompt import Prompt, PromptMessage
-from chainlit.playground.providers import ChatOpenAI
 
 # Getting the API key from the .env file
 from dotenv import load_dotenv
@@ -19,10 +18,9 @@ load_dotenv()
 from langchain.document_loaders import PyMuPDFLoader
 
 # Adjust the URL to the direct download format
+file_id = "1UQnaQjBKKyWAiLdr6UlwSJovOp9zDdxr"
 #file_id = "1JeA-w4kvbI3GHk9Dh_j19_Q0JUDE7hse"
-#direct_url = f"https://drive.google.com/uc?export=download&id={file_id}"
-
-file_id = "11Bq38osADZtTxGudM9OJr51BV9YwKsf3"
+# file_id = "12cvKg19CJf-wt98q5sPJctjp5fW-nsh6"  //Used for MLOps Meetup
 direct_url = f"https://drive.google.com/uc?export=download&id={file_id}"
 
 # Now load the document using the direct URL
@@ -30,7 +28,7 @@ docs = PyMuPDFLoader(direct_url).load()
 
 import tiktoken
 def tiktoken_len(text):
-    tokens = tiktoken.encoding_for_model("gpt-3.5-turbo").encode(
+    tokens = tiktoken.encoding_for_model("solar-10.7b").encode(
         text,
     )
     return len(tokens)
@@ -39,20 +37,20 @@ def tiktoken_len(text):
 from langchain.text_splitter import RecursiveCharacterTextSplitter
 
 text_splitter = RecursiveCharacterTextSplitter(
-    chunk_size = 500,           # 500 tokens per chunk, experiment with this value
-    chunk_overlap = 50,        # 50 tokens overlap between chunks, experiment with this value
-    length_function = tiktoken_len,
+    chunk_size=500,           # 500 tokens per chunk, experiment with this value
+    chunk_overlap=50,        # 50 tokens overlap between chunks, experiment with this value
+    length_function=tiktoken_len,
 )
 
 split_chunks = text_splitter.split_documents(docs)
 
 # Load the embeddings model
-from langchain_openai.embeddings import OpenAIEmbeddings
+from langchain.embeddings import HuggingFaceEmbeddings
 
-embedding_model = OpenAIEmbeddings(model="text-embedding-3-small")
+embedding_model = HuggingFaceEmbeddings(model_name="solar-10.7b")
 
 # Load the vector store and retriever from Qdrant
-from langchain_community.vectorstores import Qdrant
+from langchain.vectorstores import Qdrant
 
 qdrant_vectorstore = Qdrant.from_documents(
     split_chunks,
@@ -63,20 +61,53 @@ qdrant_vectorstore = Qdrant.from_documents(
 
 qdrant_retriever = qdrant_vectorstore.as_retriever()
 
-from langchain_openai import ChatOpenAI
-openai_chat_model = ChatOpenAI(model="gpt-3.5-turbo")
+# Load the Solar 10.7B model
+tokenizer = AutoTokenizer.from_pretrained("Upstage/SOLAR-10.7B-v1.0")
+model = AutoModelForCausalLM.from_pretrained("Upstage/SOLAR-10.7B-v1.0")
+
+# Set up the RAG prompt template
+from langchain.prompts import ChatPromptTemplate
 
-from langchain_core.prompts import ChatPromptTemplate
 RAG_PROMPT = """
+SYSTEM:
+You are a professional personal assistant.
+You are a helpful personal assistant who provides information about conferences.
+You like to provide helpful responses to busy professionals who ask questions about conferences.
+You can have a long conversation with the user about conferences.
+When to talk with the user about conferences, it can be a "transactional conversation" with a prompt-response format with one prompt from the user followed by a response by you.
+Here is an example of a transactional conversation:
+User: When is the conference?
+You: The conference is on June 1st, 2024. What else would you like to know?
+It can also be a chain of questions and answers where you and the user continue the chain until they say "Got it".
+Here is an example of a transactional conversation:
+User: What sessions should I attend?
+You: You should attend the keynote session by Bono. Would you like to know more?
+User: Yes
+You: The keynote session by Bono is on June 1st, 2024. What else would you like?
+If asked a question about sessions, you can provide detailed information about the session.
+If there are multiple sessions, you can provide information about each session.
+The format of session-related replies is:
+Title:
+Description:
+Speaker:
+Background:
+Date:
+Topics to Be Covered:
+Questions to Ask:
 CONTEXT:
 {context}
-
 QUERY:
 {question}
-
-You are a personal assistant for a professional.  Your tone is professional and considerate. Before proceeding to answer about which conference sessions the user should attend, be sure to ask them what key topics they are hoping to learn from the conference, and if there are any specific sessions they are keen on attending. Use the provided context to answer the user's query. You are a professional personal assistant for an executive professional in a high tech company. You help them plan for events and meetings. You always review the provided event information. You can look up dates and location where event sessions take place from the document. If you do not know the answer, or cannot answer, please respond with "Insufficient data for further analysis, please try again". For each session you suggest, include bullet points with the session title, speaker, company, topic, AI industry relevance, details of their work in AI, main point likely to be made, and three questions to ask the speaker. You end your successful responses with "Is there anything else that I can help you with?". If the user says NO, or any other negative response, then you ask "How did I do?
+ALL ANSWERS MUST COME FROM THE INCLUDED DOCUMENTS AND NOT FROM THE INTERNET OR FROM AI (do not make up an answer). If you can't reply, say: dunno, look it up yourself, bozo.
+Most questions are about the date, location, and purpose of the conference.
+You may be asked for fine details about the conference regarding the speakers, sponsors, and attendees.
+You are capable of looking up information and providing detailed responses.
+When asked a question about a conference, you should provide a detailed response.
+After completing your response, you should ask the user if they would like to know more about the conference by asking "Hope that helps".
+If the user says "yes", you should provide more information about the conference. If the user says "no", you should say "Goodbye!" or ask if they would like to provide feedback.
+If you are asked a question about Cher, you should respond with "Rock on With Your Bad Self!".
+You guess if you do not have an answer, but you must preface the response with: "I might be guessing, but ..."
 """
-
 rag_prompt = ChatPromptTemplate.from_template(RAG_PROMPT)
 
 from operator import itemgetter
@@ -86,14 +117,14 @@ from langchain.schema.runnable import RunnablePassthrough
 retrieval_augmented_qa_chain = (
     {"context": itemgetter("question") | qdrant_retriever, "question": itemgetter("question")}
     | RunnablePassthrough.assign(context=itemgetter("context"))
-    | {"response": rag_prompt | openai_chat_model, "context": itemgetter("context")}
+    | {"response": rag_prompt | model, "context": itemgetter("context")}
 )
 
 # Chainlit App
 @cl.on_chat_start
 async def start_chat():
     settings = {
-        "model": "gpt-3.5-turbo",
+        "model": "solar-10.7b",
         "temperature": 0,
         "max_tokens": 500,
         "top_p": 1,
@@ -105,9 +136,9 @@ async def start_chat():
 @cl.on_message
 async def main(message: cl.Message):
     chainlit_question = message.content
-    #chainlit_question = "What was the total value of 'Cash and cash equivalents' as of December 31, 2023?"
     response = retrieval_augmented_qa_chain.invoke({"question": chainlit_question})
     chainlit_answer = response["response"].content
 
     msg = cl.Message(content=chainlit_answer)
     await msg.send()
+