db.py

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

# Example schema
schema = {
    "products": {
        "columns": ["product_id", "name", "price", "category_id"],
        "relations": "category_id -> categories.id",
    },
    "categories": {
        "columns": ["id", "category_name"],
        "relations": None,
    },
    "orders": {
        "columns": ["order_id", "customer_name", "product_id", "order_date"],
        "relations": "product_id -> products.product_id",
    },
}

# Step 1: Generate context dynamically from schema
def generate_context(schema):
    context_lines = []
    for table, details in schema.items():
        # List table columns
        columns = ", ".join(details["columns"])
        context_lines.append(f"The {table} table has the following columns: {columns}.")
        
        # Add relationships if present
        if details["relations"]:
            context_lines.append(f"The {table} table has the following relationship: {details['relations']}.")
    
    return "\n".join(context_lines)

# Generate schema context
schema_context = generate_context(schema)

# Step 2: Load the T5-base-text-to-sql model
model_name = "suriya7/t5-base-text-to-sql"  # A model fine-tuned for SQL generation
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSeq2SeqLM.from_pretrained(model_name)

# Step 3: Define a natural language query
user_query = "List all orders where the product price is greater than 50."

# Prepare the input for the model
# Adjust the prompt to focus on SQL generation
input_text = f"Convert the following question into an SQL query:\nSchema:\n{schema_context}\n\nQuestion:\n{user_query}"
inputs = tokenizer.encode(input_text, return_tensors="pt")

# Step 4: Generate SQL query
outputs = model.generate(inputs, max_length=128, num_beams=4, early_stopping=True)
generated_sql = tokenizer.decode(outputs[0], skip_special_tokens=True)

# Step 5: Display the result
print("User Query:", user_query)
print("Generated SQL Query:", generated_sql)