NL-SQL / app.py
sidharth-pm's picture
Update app.py
581fba6 verified
import gradio as gr
import torch
import re
import tempfile
import pandas as pd
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
from database import init_database, get_schema, execute_query
# Model Setup
MODEL_ID = "microsoft/tapex-large-sql-execution"
tokenizer = None
sql_pipeline = None
def load_model():
global tokenizer, sql_pipeline
print("microsoft/tapex-large-sql-execution ...")
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
model = AutoModelForCausalLM.from_pretrained(
MODEL_ID,
torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
device_map="auto",
trust_remote_code=True,
)
sql_pipeline = pipeline(
"text-generation",
model=model,
tokenizer=tokenizer,
max_new_tokens=512,
do_sample=False,
return_full_text=False,
pad_token_id=tokenizer.eos_token_id,
)
print("Model loaded.")
PROMPT_TEMPLATE = """### Task
Generate a SQL query to answer [QUESTION]{question}[/QUESTION]
### Database Schema
The query will run on a database with the following schema:
{schema}
### Answer
Given the database schema, here is the SQL query that [QUESTION]{question}[/QUESTION]
[SQL]
"""
def build_prompt(question: str, schema: str) -> str:
return PROMPT_TEMPLATE.format(question=question, schema=schema)
def extract_sql(raw: str) -> str:
match = re.search(r"(SELECT[\s\S]+?);", raw, re.IGNORECASE)
if match:
return match.group(0).strip()
return raw.strip().split("[/SQL]")[0].strip()
def nl_to_sql_and_run(question: str, history: list):
if not question.strip():
yield history, "", gr.update(visible=False), gr.update(visible=False)
return
schema = get_schema()
prompt = build_prompt(question, schema)
yield history, "Generating SQL query...", gr.update(visible=False), gr.update(visible=False)
try:
output = sql_pipeline(prompt)[0]["generated_text"]
sql = extract_sql(output)
except Exception as e:
new_hist = history + [{"role": "user", "content": question},
{"role": "assistant", "content": f"Model error: {e}"}]
yield new_hist, "", gr.update(visible=False), gr.update(visible=False)
return
yield history, f"```sql\n{sql}\n```\n\nExecuting...", gr.update(visible=False), gr.update(visible=False)
try:
columns, rows = execute_query(sql)
except Exception as e:
answer = f"**Generated SQL:**\n```sql\n{sql}\n```\n\nExecution error: `{e}`"
new_hist = history + [{"role": "user", "content": question},
{"role": "assistant", "content": answer}]
yield new_hist, "", gr.update(visible=False), gr.update(visible=False)
return
if not rows:
result_md = "*(query returned no rows)*"
df = pd.DataFrame()
csv_path = None
else:
df = pd.DataFrame(rows, columns=columns)
result_md = df.to_markdown(index=False)
tmp = tempfile.NamedTemporaryFile(delete=False, suffix=".csv", mode="w", newline="")
df.to_csv(tmp.name, index=False)
tmp.close()
csv_path = tmp.name
row_label = "rows" if len(rows) != 1 else "row"
answer = f"**Generated SQL:**\n```sql\n{sql}\n```\n\n**Results ({len(rows)} {row_label}):**\n{result_md}"
new_hist = history + [{"role": "user", "content": question},
{"role": "assistant", "content": answer}]
yield (
new_hist,
"",
gr.update(value=df, visible=bool(rows)),
gr.update(value=csv_path, visible=bool(rows)),
)
def view_schema():
return f"```sql\n{get_schema()}\n```"
CSS = """
@import url('https://fonts.googleapis.com/css2?family=Space+Mono:wght@400;700&family=DM+Sans:wght@300;400;500&display=swap');
body, .gradio-container {
background: #0d0f14 !important;
font-family: 'DM Sans', sans-serif;
color: #e2e8f0;
}
.title-block {
text-align: center;
padding: 2rem 0 1rem;
}
.title-block h1 {
font-size: 2rem;
background: linear-gradient(135deg, #38bdf8, #818cf8);
-webkit-background-clip: text;
-webkit-text-fill-color: transparent;
font-family: 'Space Mono', monospace;
margin-bottom: 0.3rem;
}
.title-block p { color: #64748b; font-size: 0.95rem; }
.badge {
display: inline-block;
background: #1e2535;
border: 1px solid #2d3748;
border-radius: 20px;
padding: 2px 12px;
font-size: 0.75rem;
color: #94a3b8;
margin: 4px;
font-family: 'Space Mono', monospace;
}
"""
EXAMPLE_QUERIES = [
"Show me all employees in Engineering with salary above 120000",
"Which department has the highest total salary budget?",
"List all active projects with their budgets",
"Who are the top 3 sales performers by total amount?",
"How many employees are in each department?",
"Show me all sales made in the East region in 2024",
]
def create_app():
init_database()
with gr.Blocks(css=CSS, title="SQLCoder Studio") as demo:
gr.HTML("""
<div class="title-block">
<h1>SQLCoder Studio</h1>
<p>Natural language to SQL to Results &nbsp;|&nbsp; Powered by microsoft/tapex-large-sql-execution</p>
<div style="margin-top:0.8rem">
<span class="badge">employees</span>
<span class="badge">departments</span>
<span class="badge">projects</span>
<span class="badge">sales</span>
</div>
</div>
""")
with gr.Row():
with gr.Column(scale=3):
chatbot = gr.Chatbot(
label="Conversation",
height=460,
show_label=False,
render_markdown=True,
bubble_full_width=False,
type="messages",
)
with gr.Row():
question_input = gr.Textbox(
placeholder="Ask anything about the database...",
show_label=False,
scale=5,
lines=1,
)
submit_btn = gr.Button("RUN", variant="primary", scale=1)
with gr.Row():
clear_btn = gr.Button("Clear chat", variant="secondary", size="sm")
gr.HTML("<p style='color:#475569;font-size:0.78rem;margin-top:0.5rem'>Try an example:</p>")
example_btns = []
with gr.Row(wrap=True):
for eq in EXAMPLE_QUERIES:
b = gr.Button(eq, size="sm", variant="secondary")
example_btns.append(b)
with gr.Column(scale=2):
gr.HTML("<p style='color:#94a3b8;font-size:0.85rem;font-weight:500;margin-bottom:4px'>Result Table</p>")
result_table = gr.Dataframe(
visible=False,
wrap=True,
height=220,
)
download_file = gr.File(
label="Download CSV",
visible=False,
)
gr.HTML("<p style='color:#94a3b8;font-size:0.85rem;font-weight:500;margin:1rem 0 4px'>Database Schema</p>")
gr.Markdown(value=view_schema())
status_md = gr.Markdown(visible=False)
history_state = gr.State([])
def run(question, history):
gen = nl_to_sql_and_run(question, history)
for h, status, table_update, dl_update in gen:
yield h, h, status, table_update, dl_update
submit_btn.click(
fn=run,
inputs=[question_input, history_state],
outputs=[chatbot, history_state, status_md, result_table, download_file],
)
question_input.submit(
fn=run,
inputs=[question_input, history_state],
outputs=[chatbot, history_state, status_md, result_table, download_file],
)
clear_btn.click(
fn=lambda: ([], [], "", gr.update(visible=False), gr.update(visible=False)),
outputs=[chatbot, history_state, status_md, result_table, download_file],
)
for btn, eq in zip(example_btns, EXAMPLE_QUERIES):
btn.click(fn=lambda q=eq: q, outputs=[question_input])
return demo
if __name__ == "__main__":
load_model()
app = create_app()
app.launch()