Create app.py

app.py

@@ -0,0 +1,61 @@
+import gradio as gr
+import pandas as pd
+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=2048)['dense_vecs']
+# 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=2048)['dense_vecs']
+    
+    # Compute similarity between the query and the CSV embeddings
+    similarity_matrix = embeddings_query @ embeddings_csv.T
+
+    # Rank records by similarity and select the top 'num_records'
+    similarity_scores = similarity_matrix.max(axis=0)
+    top_indices = similarity_scores.argsort()[-num_records:][::-1]
+
+    # Get the top results and return all columns
+    result_df = df.iloc[top_indices].drop_duplicates(subset=df.columns.difference(['embedding_context']), keep='first')
+
+    
+    return result_df
+
+
+# Gradio interface function
+def gradio_interface(query_text):
+    search_results = search_query(query_text)
+    return search_results
+
+# Gradio interface setup
+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)
+    
+    # Output table for displaying results
+    search_output = gr.DataFrame(label="Search Results")
+    
+    # Search button
+    search_button = gr.Button("Search")
+    
+    # Link button click to action
+    search_button.click(fn=gradio_interface, inputs=search_input, outputs=search_output)
+
+# Launch the Gradio app
+app.launch()