Rename 02JUL24app.py to app.py

02JUL24app.py → app.py

@@ -1,66 +1,62 @@
+
 import gradio as gr
 from sentence_transformers import SentenceTransformer
+from langchain_community.embeddings.sentence_transformer import (
+    SentenceTransformerEmbeddings
+)
+from langchain_community.vectorstores import Chroma
 import chromadb
 import pandas as pd
 import os
+import csv
 import json
 from pathlib import Path
 from llama_index.llms.anyscale import Anyscale
 
-# Load the sentence transformer model for embedding text
-model = SentenceTransformer('all-MiniLM-L6-v2')
+# Transformer model for embedding
+#model = SentenceTransformer('all-MiniLM-L6-v2')
+model = SentenceTransformerEmbeddings(model_name='thenlper/gte-large')
 
-# Initialize the ChromaDB client for managing the vector database
+# ChromaDB client for managing the vdb
 chroma_client = chromadb.Client()
 
-# Function to re-encode embeddings
-def reencode_embeddings(embeddings):
-    return [model.encode(eval(embedding.replace(',,', ','))).tolist() for embedding in embeddings]
-
-# Function to build the vector database from a CSV file
+# Function to build the vdb from csv
 def build_database():
-    # Read the CSV file containing document data
     df = pd.read_csv('vector_store.csv')
-
-    # Name of the collection to store the data
+    print(df.head())
     collection_name = 'Dataset-10k-companies'
-    
-    # Uncomment the line below to delete the existing collection if needed
-    # chroma_client.delete_collection(name=collection_name)
-    
-    # Create a new collection in ChromaDB
+
+
+    # Creating a new collection
     collection = chroma_client.create_collection(name=collection_name)
-    
-    # Re-encode the embeddings to match the model's dimensionality
-    embeddings = reencode_embeddings(df['embeddings'].tolist())
 
-    # Add data from the DataFrame to the collection
+    # Add data from the created DataFrame
     collection.add(
         documents=df['documents'].tolist(),
         ids=df['ids'].tolist(),
         metadatas=df['metadatas'].apply(eval).tolist(),
-        embeddings=embeddings
+        embeddings=df['embeddings'].apply(lambda x: eval(x.replace(',,', ','))).tolist()
     )
-    
+
     return collection
 
-# Build the database when the app starts
+# Build the database
 collection = build_database()
 
-# Access the Anyscale API key from environment variables
+# Get API key from hf environment variables
 anyscale_api_key = os.environ.get('anyscale_api_key')
 
-# Instantiate the Anyscale client for using the Llama language model
+# Anyscale client for using the Llama language model
 client = Anyscale(api_key=anyscale_api_key, model="meta-llama/Llama-2-70b-chat-hf")
 
 # Function to get relevant chunks from the database based on the query
 def get_relevant_chunks(query, collection, top_n=3):
     # Encode the query into an embedding
     query_embedding = model.encode(query).tolist()
-    
+
     # Query the collection to get the top_n most relevant results
     results = collection.query(query_embeddings=[query_embedding], n_results=top_n)
-    
+
     relevant_chunks = []
     # Extract relevant chunks and their metadata
     for i in range(len(results['documents'][0])):
@@ -68,7 +64,7 @@ def get_relevant_chunks(query, collection, top_n=3):
         source = results['metadatas'][0][i]['source']
         page = results['metadatas'][0][i]['page']
         relevant_chunks.append((chunk, source, page))
-    
+
     return relevant_chunks
 
 # System message template for the LLM to provide structured responses
@@ -128,10 +124,10 @@ def predict(company, user_query):
     try:
         # Modify the query to include the company name
         modified_query = f"{user_query} for {company}"
-        
+
         # Get relevant chunks from the database
         relevant_chunks = get_relevant_chunks(modified_query, collection)
-        
+
         # Prepare the context string from the relevant chunks
         context = ""
         for chunk, source, page in relevant_chunks:
@@ -149,7 +145,7 @@ def predict(company, user_query):
 
         # Log the interaction for future reference
         log_interaction(company, user_query, context, answer)
-        
+
         return answer
     except Exception as e:
         return f"An error occurred: {str(e)}"
@@ -167,8 +163,8 @@ def log_interaction(company, user_query, context, answer):
         f.write("\n")
 
 # Create Gradio interface for user interaction
-company_list = ["MSFT", "AWS", "Meta", "Google", "IBM"]
-iface = gr.Interface(
+company_list = ["Meta", "IBM", "MSFT", "Google", "AWS"]
+interface = gr.Interface(
     fn=predict,
     inputs=[
         gr.Radio(company_list, label="Select Company"),
@@ -179,5 +175,5 @@ iface = gr.Interface(
     description="Query the vector database and get an LLM response based on the documents in the collection."
 )
 
-# Launch the Gradio interface
-iface.launch(share=True)
+# Launch the Gradio interface with public sharing enabled
+interface.launch()