implemented RAG

.gitignore

@@ -1 +1,4 @@
-.env
+.env
+*.ipynb
+*.csv
+*.json

app.py

@@ -1,13 +1,56 @@
 import streamlit as st
 from mistral7b import mistral
+import time
+import pandas as pd
+import pinecone
+import os
+from dotenv import load_dotenv
+from sentence_transformers import SentenceTransformer
+load_dotenv()
+
+PINECONE_TOKEN = os.getenv('PINECONE_TOKEN')
+
+pinecone.init(      
+	api_key=PINECONE_TOKEN,      
+	environment='gcp-starter'      
+)   
+
+pinecone_index = pinecone.Index('ikigai-chat')
+text_vectorizer = SentenceTransformer('all-distilroberta-v1')
+
+def gen_augmented_prompt(prompt, top_k) :
+    query_vector = text_vectorizer.encode(prompt).tolist()
+    res = pinecone_index.query(vector=query_vector, top_k=top_k, include_metadata=True)
+    matches = res['matches']
+
+    context = ""
+    links = []
+    for match in matches :
+        context+=match["metadata"]["chunk"] + "\n\n"
+        links.append(match["metadata"]["link"])
+    
+    
+
+    generated_prompt = f"""
+    FOR THIS GIVEN CONTEXT {context},
+
+
+    ANSWER THE FOLLOWING PROMPT {prompt}
+    """
+    return generated_prompt, links
+
+data = {
+    "Attribute": ["LLM", "Text Vectorizer", "Vector Database","CPU", "System RAM"],
+    "Information": ["Mistral-7B-Instruct-v0.1 (more models soon)","all-distilroberta-v1", "Hosted Pinecone" ,"2 vCPU", "16 GB"]
+}
+df = pd.DataFrame(data)
+
 
 st.set_page_config(
         page_title="Ikigai Chat",
+        page_icon="🤖",
 )
 
-import time
-
-
 if "messages" not in st.session_state:
     st.session_state.messages = []
 
@@ -17,7 +60,6 @@ if "tokens_used" not in st.session_state:
 if "inference_time" not in st.session_state:
     st.session_state.inference_time = [0.00]
 
-
 if "temp" not in st.session_state:
     st.session_state.temp = 0.8
 
@@ -25,24 +67,34 @@ if "history" not in st.session_state:
     st.session_state.history = [["""
                                   You are not mistral AI, but rather a chat bot trained at Ikigai Labs, when ever asked you need to answer as ikigai Labs' assistant. 
                                   Ikigai helps modern analysts and operations teams automate data-intensive business, finance, analytics, and supply-chain operations. 
-                                  The company's Inventory Ops automates inventory tracking and monitoring by creating a single, real-time view of inventory across all locations and channels.""", 
+                                  The company's Inventory Ops automates inventory tracking and monitoring by creating a single, real-time view of inventory across all locations and channels.
+                                 """, 
+                      
                                   """
                                 Yes, you are correct. Ikigai Labs is a company that specializes in helping 
                                 modern analysts and operations teams automate data-intensive business, finance, analytics, 
                                 and supply chain operations. One of their products is Inventory Ops, which automates inventory 
                                 tracking and monitoring by creating a single, real-time view of inventory across all locations and channels.
                                 This helps businesses optimize their inventory levels and reduce costs. 
-                                Is there anything else you would like to know about Ikigai Labs or their products?
-
-
-"""]]
+                                Is there anything else you would like to know about Ikigai Labs or their products?"""], 
+                                ["""You are ikigai chat from now on, so answer accordingly""", 
+                                 """Sure, I will do my best to answer your questions as Ikigai Chat. 
+                                 Let me know if you have any specific questions about Ikigai Labs or our products."""]]
 
 if "top_k" not in st.session_state:
-    st.session_state.top_k = 5
+    st.session_state.top_k = 3
 
 if "repetion_penalty" not in st.session_state :
     st.session_state.repetion_penalty = 1
+
+if "rag_enabled" not in st.session_state :
+    st.session_state.rag_enabled = True
 with st.sidebar:
+    st.markdown("# Retrieval Settings")
+    st.session_state.rag_enabled = st.toggle("Activate RAG")
+    st.session_state.top_k = st.slider(label="Documents to retrieve",
+              min_value=1, max_value=10, value=3, disabled=not st.session_state.rag_enabled)
+    st.markdown("---")
     st.markdown("# Model Analytics")
     
     st.write("Tokens used :", st.session_state['tokens_used'])
@@ -68,23 +120,20 @@ with st.sidebar:
         label="Repetion Penalty", min_value=0., max_value=1., step=0.1, value=1. 
     )
     
-    st.markdown("---")
+    st.markdown("""
+    > **2023 ©️ Pragnesh Barik**
+    """)
     
-    st.markdown("# Retrieval Settings")
-    st.slider(label="Documents to retrieve",
-              min_value=1, max_value=10, value=3)
-    
-    st.info("**2023 ©️ Pragnesh Barik**")
-
-
 st.image("ikigai.svg")
 st.title("Ikigai Chat")
+st.caption("Maintained and developed by Pragnesh Barik.")
 
 with st.expander("What is Ikigai Chat ?"):
     st.info("""Ikigai Chat is a vector database powered chat agent, it works on the principle of 
                 of Retrieval Augmented Generation (RAG), Its primary function revolves around maintaining an extensive repository of Ikigai Docs and providing users with answers that align with their queries. 
                 This approach ensures a more refined and tailored response to user inquiries.""")
-
+    
+    st.table(df)
 
 for message in st.session_state.messages:
     with st.chat_message(message["role"]):
@@ -95,8 +144,14 @@ if prompt := st.chat_input("Chat with Ikigai Docs..."):
     st.chat_message("user").markdown(prompt)
     st.session_state.messages.append({"role": "user", "content": prompt})
 
-    # st.write("ing")
+
     tick = time.time()
+    
+    links = []
+    if st.session_state.rag_enabled :
+        with st.spinner("Fetching relevent documents from Ikigai Docs...."):
+            prompt, links = gen_augmented_prompt(prompt=prompt, top_k=st.session_state.top_k)
+        
     with st.spinner("Generating response...") :
         response = mistral(prompt, st.session_state.history,
                        temperature=st.session_state.temp, max_new_tokens=st.session_state.max_tokens)
@@ -110,8 +165,16 @@ if prompt := st.chat_input("Chat with Ikigai Docs..."):
     st.session_state["tokens_used"] = len_response + \
         st.session_state["tokens_used"]
 
+    formatted_links = ", ".join(links) 
     with st.chat_message("assistant"):
-        st.markdown(response)
+        if st.session_state.rag_enabled :
+            st.markdown(response + f"""\n\nFetched from :\n {formatted_links}""")
+        else :
+            st.markdown(response)
     st.session_state.history.append([prompt, response])
-    st.session_state.messages.append(
-        {"role": "assistant", "content": response})
+    
+    if st.session_state.rag_enabled :
+        st.session_state.messages.append(
+            {"role": "assistant", "content": response + f"""\n\nFetched from :\n {formatted_links}"""})
+    else :
+        st.session_state.messages.append({"role": "assistant", "content": response})

requirements.txt

@@ -73,3 +73,5 @@ validators==0.22.0
 watchdog==3.0.0
 wcwidth==0.2.6
 zipp==3.17.0
+sentence-transformers==2.2.2
+pinecone-client==2.2.4