steamlit app

chatbot.py

@@ -1,6 +1,5 @@
 from langchain.prompts import PromptTemplate
-from langchain.llms import OpenAI
-from langchain.embeddings import OpenAIEmbeddings
+from langchain import HuggingFaceHub
 from langchain.chains import LLMChain
 from langchain.memory import ConversationBufferMemory
 from redis.commands.search.query import Query
@@ -8,46 +7,25 @@ import time
 import os
 from dotenv import load_dotenv
 import numpy as np
-from database import redis_conn
-
 load_dotenv()
+HUGGINGFACEHUB_API_TOKEN = os.getenv('HUGGINGFACEHUB_API_TOKEN')
+repo_id = 'tiiuae/falcon-7b-instruct'
+
+falcon_llm_1 = HuggingFaceHub(repo_id = repo_id, model_kwargs={'temperature':0.1,'max_new_tokens':500},huggingfacehub_api_token=HUGGINGFACEHUB_API_TOKEN)
 
-llm = OpenAI(model_name="gpt-3.5-turbo", temperature=0.3, openai_api_key=os.getenv('OPENAI_API_KEY'))
 prompt = PromptTemplate(
     input_variables=["product_description"],
     template="Create comma seperated product keywords to perform a query on a amazon dataset for this user input: {product_description}",
 )
 
-chain = LLMChain(llm=llm, prompt=prompt)
-
-userinput = input("Hey im a E-commerce Chatbot, how can i help you today? ")
-print("User:", userinput)
-# Run the chain only specifying the input variable.
-keywords = chain.run(userinput)
-
-embedding_model = OpenAIEmbeddings(openai_api_key=os.getenv('OPENAI_API_KEY'))
-#vectorize the query
-query_vector = embedding_model.embed_query(keywords)
-query_vector = np.array(query_vector).astype(np.float32).tobytes()
-
-
-#prepare the query
-ITEM_KEYWORD_EMBEDDING_FIELD = 'item_vector'
-topK=5
-q = Query(f'*=>[KNN {topK} @{ITEM_KEYWORD_EMBEDDING_FIELD} $vec_param AS vector_score]').sort_by('vector_score').paging(0,topK).return_fields('vector_score','item_name','item_id','item_keywords').dialect(2)
-params_dict = {"vec_param": query_vector}
-#Execute the query
-results = redis_conn.ft().search(q, query_params = params_dict)
-
-full_result_string = ''
-for product in results.docs:
-    full_result_string += product.item_name + ' ' + product.item_keywords + ' ' + product.item_id + "\n\n\n"
+chain = LLMChain(llm=falcon_llm_1, prompt=prompt)
 
 # code The response
-template = """You are a chatbot. Be kind, detailed and nice. Present the given queried search result in a nice way as answer to the user input. dont ask questions back! just take the given context
+repo_id_2 = 'tiiuae/falcon-7b' 
+template = """You are a salesman. Be kind, detailed and nice.  take the given context and Present the given queried search result in a nice way as answer to the user_msg. dont ask questions back or freestyle and invent followup conversation! just 
 
 {chat_history}
-Human: {user_msg}
+{user_msg}
 Chatbot:"""
 
 prompt = PromptTemplate(
@@ -55,22 +33,15 @@ prompt = PromptTemplate(
     template=template
 )
 memory = ConversationBufferMemory(memory_key="chat_history")
+
 llm_chain = LLMChain(
-    llm=OpenAI(model_name="gpt-3.5-turbo", temperature=0.8, openai_api_key=os.getenv('OPENAI_API_KEY')),
+    llm = HuggingFaceHub(repo_id = repo_id_2, model_kwargs={'temperature':0.8,'max_new_tokens':500}),
     prompt=prompt,
     verbose=False,
     memory=memory,
 )
 
-answer = llm_chain.predict(user_msg=f"{full_result_string} ---\n\n {userinput}")
-print("Bot:", answer)
-time.sleep(0.5)
 
-while True:
-    follow_up = input("Anything else you want to ask about this topic?")
-    print("User:", follow_up)
-    answer = llm_chain.predict(
-        user_msg=follow_up
-    )
-    print("Bot:", answer)
-    time.sleep(0.5)
+
+
+

database.py

@@ -8,8 +8,8 @@ redis_key = os.getenv('REDIS_KEY')
 
 
 redis_conn = redis.Redis(
-  host='redis-10923.c10.us-east-1-4.ec2.cloud.redislabs.com',
-  port=10923,
+  host='redis-12882.c259.us-central1-2.gce.cloud.redislabs.com',
+  port=12882,
   password=redis_key)
 
 print('connected to redis')

main.py

@@ -0,0 +1,47 @@
+import streamlit as st
+from chatbot import llm_chain, chain
+from sentence_transformers import SentenceTransformer
+from redis.commands.search.query import Query
+from database import redis_conn
+import numpy as np
+
+
+
+st.title('My Amazon shopping buddy 🏷️')
+st.caption('🤖 Powered by Falcon Open Source AI model')
+st.session_state['disabled']= False
+
+if "messages" not in st.session_state:
+    st.session_state["messages"] = [{"role": "assistant", "content": "Hey im your online shopping buddy, how can i help you today?"}]
+for msg in st.session_state["messages"]:
+    st.chat_message(msg["role"]).write(msg["content"])
+
+prompt = st.chat_input(key="user_input",disabled=st.session_state.disabled )
+embedding_model = SentenceTransformer('sentence-transformers/all-distilroberta-v1')
+if prompt:
+    st.session_state["messages"].append({"role": "user", "content": prompt})
+    st.chat_message('user').write(prompt)
+    st.session_state.disabled = True
+    keywords = chain.run(prompt)
+
+    #vectorize the query
+    query_vector = embedding_model.encode(keywords)
+    query_vector = np.array(query_vector).astype(np.float32).tobytes()
+    #prepare the query
+    ITEM_KEYWORD_EMBEDDING_FIELD = 'item_vector'
+    topK=5
+    q = Query(f'*=>[KNN {topK} @{ITEM_KEYWORD_EMBEDDING_FIELD} $vec_param AS vector_score]').sort_by('vector_score').paging(0,topK).return_fields('vector_score','item_name','item_id','item_keywords').dialect(2)
+    params_dict = {"vec_param": query_vector}
+    #Execute the query
+    results = redis_conn.ft().search(q, query_params = params_dict)
+
+    full_result_string = ''
+    for product in results.docs:
+        full_result_string += product.item_name + ' ' + product.item_keywords  + "\n\n\n"
+
+    result = llm_chain.predict(user_msg=f"{full_result_string} ---\n\n {prompt}")
+    st.session_state.messages.append({"role": "assistant", "content": result})
+    st.chat_message('assistant').write(result)
+
+
+

preprocess.py

@@ -1,4 +1,5 @@
 from langchain.embeddings import OpenAIEmbeddings
+from sentence_transformers import SentenceTransformer
 import os
 import pandas as pd
 import numpy as np
@@ -6,8 +7,6 @@ from dotenv import load_dotenv
 from database import redis_conn
 from utilities import create_flat_index, load_vectors
 
-load_dotenv()
-openai_api_key = os.getenv("OPENAI_API_KEY")
 
 #set maximum length for text fields
 MAX_TEXT_LENGTH = 512
@@ -24,12 +23,12 @@ data.reset_index(drop=True, inplace=True)
 data_metadata = data.head(500).to_dict(orient='index')
 
 #generating embeddings (vectors) for the item keywords
-# embedding_model = SentenceTransformer('sentence-transformers/all-distilroberta-v1')
-embedding_model = OpenAIEmbeddings(openai_api_key=openai_api_key)
+embedding_model = SentenceTransformer('sentence-transformers/all-distilroberta-v1')
+# embedding_model = OpenAIEmbeddings(openai_api_key=openai_api_key)
 
 #get the item keywords attribute for each product and encode them into vector embeddings
 item_keywords = [data_metadata[i]['item_keywords'] for i in data_metadata.keys()]
-item_keywords_vectors = [embedding_model.embed_query(item) for item in item_keywords]
+item_keywords_vectors = [embedding_model.encode(item) for item in item_keywords]
 
 TEXT_EMBEDDING_DIMENSION=768
 NUMBER_PRODUCTS=500

requirements.txt

@@ -3,3 +3,6 @@ openai == 0.27.8
 redis == 5.0.1
 pandas == 2.0.3
 sentence-transformers == 2.2.2
+tiktoken == 0.5.1
+streamlit == 1.27.2
+python-dotenv = 1.0.0