Update app.py

app.py

@@ -1,35 +1,52 @@
 import gradio as gr
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
 from datasets import load_dataset
-from difflib import get_close_matches
 
 # Load the WikiSQL dataset
-wikisql_dataset = load_dataset("wikisql", split='train[:100]')
+wikisql_dataset = load_dataset("wikisql", split='train[:100]')  # Load a subset of the dataset
 
-# Create a mapping between natural language queries and SQL queries
-query_sql_mapping = {item['question']: item['sql']['human_readable'] for item in wikisql_dataset}
+# Extract schema information from the dataset
+table_names = set()
+column_names = set()
+for item in wikisql_dataset:
+    table_names.add(item['table']['name'])
+    for column in item['table']['header']:
+        column_names.add(column)
 
-def find_closest_match(query, dataset):
-    questions = [item['question'] for item in dataset]
-    matches = get_close_matches(query, questions, n=1)
-    return matches[0] if matches else None
+# Load tokenizer and model
+tokenizer = AutoTokenizer.from_pretrained("mrm8488/t5-base-finetuned-wikiSQL")
+model = AutoModelForSeq2SeqLM.from_pretrained("mrm8488/t5-base-finetuned-wikiSQL")
+
+def post_process_sql_query(sql_query):
+    # Modify the SQL query to match the dataset's schema
+    # This is just an example and might need to be adapted based on the dataset and model output
+    for table_name in table_names:
+        if "TABLE" in sql_query:
+            sql_query = sql_query.replace("TABLE", table_name)
+            break  # Assuming only one table is referenced in the query
+    for column_name in column_names:
+        if "COLUMN" in sql_query:
+            sql_query = sql_query.replace("COLUMN", column_name, 1)
+    return sql_query
 
 def generate_sql_from_user_input(query):
-    # Find the closest match in the dataset
-    matched_query = find_closest_match(query, wikisql_dataset)
-    if not matched_query:
-        return "No close match found in the dataset."
+    # Generate SQL for the user's query
+    input_text = "translate English to SQL: " + query
+    inputs = tokenizer(input_text, return_tensors="pt", padding=True)
+    outputs = model.generate(**inputs, max_length=512)
+    sql_query = tokenizer.decode(outputs[0], skip_special_tokens=True)
 
-    # Retrieve the corresponding SQL query from the dataset
-    sql_query = query_sql_mapping.get(matched_query, "SQL query not found.")
+    # Post-process the SQL query to match the dataset's schema
+    sql_query = post_process_sql_query(sql_query)
     return sql_query
 
 # Create a Gradio interface
 interface = gr.Interface(
     fn=generate_sql_from_user_input,
     inputs=gr.Textbox(label="Enter your natural language query"),
-    outputs=gr.Textbox(label="SQL Query from Dataset"),
-    title="NL to SQL using WikiSQL Dataset",
-    description="Enter a natural language query and get the corresponding SQL query from the WikiSQL dataset."
+    outputs=gr.Textbox(label="Generated SQL Query"),
+    title="NL to SQL with T5 using WikiSQL Dataset",
+    description="This model generates an SQL query for your natural language input based on the WikiSQL dataset."
 )
 
 # Launch the app