app.py

import gradio as gr
import pandas as pd
import faiss
import numpy as np
import os
from FlagEmbedding import BGEM3FlagModel

# Load the pre-trained embedding model
model = BGEM3FlagModel('BAAI/bge-m3', use_fp16=True)

# Load the JSON data into a DataFrame
df = pd.read_json('White-Stride-Red-68.json')
df['embeding_context'] = df['embeding_context'].astype(str).fillna('')

# Filter out any rows where 'embeding_context' might be empty or invalid
df = df[df['embeding_context'] != '']

# Encode the 'embeding_context' column
embedding_contexts = df['embeding_context'].tolist()
embeddings_csv = model.encode(embedding_contexts, batch_size=12, max_length=1024)['dense_vecs']

# Convert embeddings to numpy array
embeddings_np = np.array(embeddings_csv).astype('float32')

# FAISS index file path
index_file_path = 'vector_store_bge_m3.index'

# Check if FAISS index file already exists
if os.path.exists(index_file_path):
    # Load the existing FAISS index from file
    index = faiss.read_index(index_file_path)
    print("FAISS index loaded from file.")
else:
    # Initialize FAISS index (for L2 similarity)
    dim = embeddings_np.shape[1]
    index = faiss.IndexFlatL2(dim)

    # Add embeddings to the FAISS index
    index.add(embeddings_np)

    # Save the FAISS index to a file for future use
    faiss.write_index(index, index_file_path)
    print("FAISS index created and saved to file.")

index = faiss.read_index(index_file_path)


# Function to perform search and return all columns
def search_query(query_text):
    num_records = 50

    # Encode the input query text
    embeddings_query = model.encode([query_text], batch_size=12, max_length=1024)['dense_vecs']
    embeddings_query_np = np.array(embeddings_query).astype('float32')

    # Search in FAISS index for nearest neighbors
    distances, indices = index.search(embeddings_query_np, num_records)

    # Get the top results based on FAISS indices
    result_df = df.iloc[indices[0]].drop(columns=['embeding_context']).drop_duplicates().reset_index(drop=True)

    return result_df

# Gradio interface function
def gradio_interface(query_text):
    search_results = search_query(query_text)
    return search_results

with gr.Blocks() as app:
    gr.Markdown("<h1>White Stride Red Search (BEG-M3)</h1>")

    # Input text box for the search query
    search_input = gr.Textbox(label="Search Query", placeholder="Enter search text", interactive=True)

    # Search button below the text box
    search_button = gr.Button("Search")

    # Output table for displaying results
    search_output = gr.DataFrame(label="Search Results")

    # Link button click to action
    search_button.click(fn=gradio_interface, inputs=search_input, outputs=search_output)



# Launch the Gradio app
app.launch()