Upload 2 files

app.py

@@ -0,0 +1,57 @@
+import gradio as gr
+import pandas as pd
+import sqlite3
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
+
+# Load NL2SQL model
+MODEL_NAME = "mrm8488/t5-base-finetuned-wikiSQL"
+tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
+model = AutoModelForSeq2SeqLM.from_pretrained(MODEL_NAME)
+
+def nl_to_sql(nl_query, uploaded_file):
+    try:
+        # Load CSV
+        df = pd.read_csv(uploaded_file.name)
+        table_name = "user_table"
+
+        # Save DataFrame to SQLite
+        conn = sqlite3.connect(":memory:")
+        df.to_sql(table_name, conn, index=False, if_exists='replace')
+
+        # Generate SQL
+        input_ids = tokenizer(nl_query, return_tensors="pt").input_ids
+        outputs = model.generate(input_ids)
+        sql_query = tokenizer.decode(outputs[0], skip_special_tokens=True)
+
+        # Adjust SQL to use uploaded table name if needed
+        sql_query = sql_query.replace("table", table_name)
+
+        # Execute SQL safely
+        try:
+            result_df = pd.read_sql_query(sql_query, conn)
+            result = result_df.to_string(index=False)
+        except Exception as e:
+            result = f"SQL execution failed:\n{e}\n\nGenerated SQL:\n{sql_query}"
+
+        return sql_query, result
+
+    except Exception as e:
+        return "", f"Error processing file: {e}"
+
+# Gradio UI
+iface = gr.Interface(
+    fn=nl_to_sql,
+    inputs=[
+        gr.Textbox(label="Enter your question in natural language"),
+        gr.File(label="Upload CSV", type="file")
+    ],
+    outputs=[
+        gr.Textbox(label="Generated SQL Query"),
+        gr.Textbox(label="Query Result")
+    ],
+    title="Natural Language to SQL",
+    description="Upload a CSV dataset and ask questions in English. SQL queries and results will be generated automatically."
+)
+
+if __name__ == "__main__":
+    iface.launch()

requirements.txt

@@ -0,0 +1,5 @@
+transformers==4.34.0
+torch
+pandas
+gradio
+sqlalchemy