Upload app.py with huggingface_hub

app.py

@@ -0,0 +1,84 @@
+import re
+import gradio as gr
+from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
+import torch
+
+MODEL_ID = "RealMati/t2sql_v6_structured"
+
+print(f"Loading model: {MODEL_ID}")
+tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
+model = AutoModelForSeq2SeqLM.from_pretrained(MODEL_ID)
+model.eval()
+print("Model loaded.")
+
+SQL_KEYWORDS = [
+    "SELECT", "FROM", "WHERE", "AND", "OR", "NOT", "IN", "LIKE",
+    "JOIN", "LEFT", "RIGHT", "INNER", "OUTER", "ON", "AS",
+    "GROUP", "BY", "ORDER", "HAVING", "LIMIT", "OFFSET",
+    "INSERT", "INTO", "VALUES", "UPDATE", "SET", "DELETE",
+    "CREATE", "TABLE", "DROP", "ALTER", "INDEX", "VIEW",
+    "DISTINCT", "COUNT", "SUM", "AVG", "MIN", "MAX",
+    "BETWEEN", "EXISTS", "UNION", "ALL", "ANY", "CASE",
+    "WHEN", "THEN", "ELSE", "END", "IS", "NULL", "ASC", "DESC",
+]
+
+
+def postprocess_sql(sql: str) -> str:
+    sql = sql.strip()
+    sql = re.sub(r"<pad>|<unk>|<s>|</s>", "", sql)
+    sql = re.sub(r"\s+", " ", sql)
+    for kw in SQL_KEYWORDS:
+        sql = re.sub(rf"\b{re.escape(kw.lower())}\b", kw, sql, flags=re.IGNORECASE)
+    return sql.strip()
+
+
+def predict(question: str, schema: str, num_beams: int, max_length: int):
+    if not question.strip():
+        return "Please enter a question."
+
+    input_text = f"translate to SQL: {question}"
+    if schema.strip():
+        input_text += f" | schema: {schema.strip()}"
+
+    inputs = tokenizer(input_text, return_tensors="pt", max_length=512, truncation=True)
+
+    with torch.no_grad():
+        outputs = model.generate(
+            **inputs,
+            max_length=int(max_length),
+            num_beams=int(num_beams),
+            early_stopping=True,
+            do_sample=False,
+        )
+
+    raw_sql = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    return postprocess_sql(raw_sql)
+
+
+demo = gr.Interface(
+    fn=predict,
+    inputs=[
+        gr.Textbox(
+            label="Natural Language Question",
+            placeholder="e.g. Show all users older than 18",
+            lines=2,
+        ),
+        gr.Textbox(
+            label="Database Schema (optional)",
+            placeholder="e.g. users: id, name, age, email | orders: id, user_id, total",
+            lines=2,
+        ),
+        gr.Slider(minimum=1, maximum=10, value=5, step=1, label="Beam Size"),
+        gr.Slider(minimum=64, maximum=512, value=256, step=64, label="Max Output Length"),
+    ],
+    outputs=gr.Textbox(label="Generated SQL", lines=3),
+    title="Text-to-SQL (T5 Fine-tuned)",
+    description="Convert natural language questions to SQL queries. Provide a schema for better results.",
+    examples=[
+        ["Show all users older than 18", "users: id, name, age, email", 5, 256],
+        ["Count orders per customer", "customers: id, name | orders: id, customer_id, total", 5, 256],
+        ["Find the most expensive product", "products: id, name, price, category", 5, 256],
+    ],
+)
+
+demo.launch()