Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,246 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
3 |
+
import torch
|
4 |
+
import re
|
5 |
+
import sqlparse
|
6 |
+
|
7 |
+
# Load model and tokenizer
|
8 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
9 |
+
model = AutoModelForCausalLM.from_pretrained(
|
10 |
+
"onkolahmet/Qwen2-0.5B-Instruct-SQL-generator",
|
11 |
+
torch_dtype="auto",
|
12 |
+
device_map="auto"
|
13 |
+
)
|
14 |
+
tokenizer = AutoTokenizer.from_pretrained("onkolahmet/Qwen2-0.5B-Instruct-SQL-generator")
|
15 |
+
|
16 |
+
# # Few-shot examples to include in each prompt
|
17 |
+
# examples = [
|
18 |
+
# {
|
19 |
+
# "question": "Get the names and emails of customers who placed an order in the last 30 days.",
|
20 |
+
# "sql": "SELECT name, email FROM customers WHERE order_date >= DATE_SUB(CURDATE(), INTERVAL 30 DAY);"
|
21 |
+
# },
|
22 |
+
# {
|
23 |
+
# "question": "Find all employees with a salary greater than 50000.",
|
24 |
+
# "sql": "SELECT * FROM employees WHERE salary > 50000;"
|
25 |
+
# },
|
26 |
+
# {
|
27 |
+
# "question": "List all product names and their categories where the price is below 50.",
|
28 |
+
# "sql": "SELECT name, category FROM products WHERE price < 50;"
|
29 |
+
# },
|
30 |
+
# {
|
31 |
+
# "question": "How many users registered in the year 2022?",
|
32 |
+
# "sql": "SELECT COUNT(*) FROM users WHERE YEAR(registration_date) = 2022;"
|
33 |
+
# }
|
34 |
+
# ]
|
35 |
+
|
36 |
+
def generate_sql(question, context=None):
|
37 |
+
# Construct prompt with few-shot examples and context if available
|
38 |
+
prompt = "Translate natural language questions to SQL queries.\n\n"
|
39 |
+
|
40 |
+
# Add table context if available
|
41 |
+
if context and context.strip():
|
42 |
+
prompt += f"Table Context:\n{context}\n\n"
|
43 |
+
|
44 |
+
# # Add few-shot examples
|
45 |
+
# for ex in examples:
|
46 |
+
# prompt += f"Q: {ex['question']}\nSQL: {ex['sql']}\n\n"
|
47 |
+
|
48 |
+
# Add the current question
|
49 |
+
prompt += f"Q: {question}\nSQL:"
|
50 |
+
|
51 |
+
# Tokenize and generate
|
52 |
+
inputs = tokenizer(prompt, return_tensors="pt").to(device)
|
53 |
+
|
54 |
+
# Generate SQL query
|
55 |
+
outputs = model.generate(
|
56 |
+
inputs.input_ids,
|
57 |
+
max_new_tokens=128,
|
58 |
+
do_sample=True,
|
59 |
+
eos_token_id=tokenizer.eos_token_id
|
60 |
+
)
|
61 |
+
|
62 |
+
# Extract and decode only the new generation
|
63 |
+
sql_query = tokenizer.decode(outputs[0][inputs.input_ids.shape[-1]:], skip_special_tokens=True)
|
64 |
+
return sql_query.strip()
|
65 |
+
|
66 |
+
def clean_sql_output(sql_text):
|
67 |
+
"""
|
68 |
+
Clean and deduplicate SQL queries:
|
69 |
+
1. Remove comments
|
70 |
+
2. Remove duplicate queries
|
71 |
+
3. Extract only the most relevant query
|
72 |
+
4. Format properly
|
73 |
+
"""
|
74 |
+
# Remove SQL comments (both single line and multi-line)
|
75 |
+
sql_text = re.sub(r'--.*?$', '', sql_text, flags=re.MULTILINE)
|
76 |
+
sql_text = re.sub(r'/\*.*?\*/', '', sql_text, flags=re.DOTALL)
|
77 |
+
|
78 |
+
# Remove markdown code block syntax if present
|
79 |
+
sql_text = re.sub(r'```sql|```', '', sql_text)
|
80 |
+
|
81 |
+
# Split into individual queries if multiple exist
|
82 |
+
if ';' in sql_text:
|
83 |
+
queries = [q.strip() for q in sql_text.split(';') if q.strip()]
|
84 |
+
else:
|
85 |
+
# If no semicolons, try to identify separate queries by SELECT statements
|
86 |
+
sql_text_cleaned = re.sub(r'\s+', ' ', sql_text)
|
87 |
+
select_matches = list(re.finditer(r'SELECT\s+', sql_text_cleaned, re.IGNORECASE))
|
88 |
+
|
89 |
+
if len(select_matches) > 1:
|
90 |
+
queries = []
|
91 |
+
for i in range(len(select_matches)):
|
92 |
+
start = select_matches[i].start()
|
93 |
+
end = select_matches[i+1].start() if i < len(select_matches) - 1 else len(sql_text_cleaned)
|
94 |
+
queries.append(sql_text_cleaned[start:end].strip())
|
95 |
+
else:
|
96 |
+
queries = [sql_text]
|
97 |
+
|
98 |
+
# Remove empty queries
|
99 |
+
queries = [q for q in queries if q.strip()]
|
100 |
+
|
101 |
+
if not queries:
|
102 |
+
return ""
|
103 |
+
|
104 |
+
# If we have multiple queries, need to deduplicate
|
105 |
+
if len(queries) > 1:
|
106 |
+
# Normalize queries for comparison (lowercase, remove extra spaces)
|
107 |
+
normalized_queries = []
|
108 |
+
for q in queries:
|
109 |
+
# Use sqlparse to format and normalize
|
110 |
+
try:
|
111 |
+
formatted = sqlparse.format(
|
112 |
+
q + ('' if q.strip().endswith(';') else ';'),
|
113 |
+
keyword_case='lower',
|
114 |
+
identifier_case='lower',
|
115 |
+
strip_comments=True,
|
116 |
+
reindent=True
|
117 |
+
)
|
118 |
+
normalized_queries.append(formatted)
|
119 |
+
except:
|
120 |
+
# If sqlparse fails, just do basic normalization
|
121 |
+
normalized = re.sub(r'\s+', ' ', q.lower().strip())
|
122 |
+
normalized_queries.append(normalized)
|
123 |
+
|
124 |
+
# Find unique queries
|
125 |
+
unique_queries = []
|
126 |
+
unique_normalized = []
|
127 |
+
|
128 |
+
for i, norm_q in enumerate(normalized_queries):
|
129 |
+
if norm_q not in unique_normalized:
|
130 |
+
unique_normalized.append(norm_q)
|
131 |
+
unique_queries.append(queries[i])
|
132 |
+
|
133 |
+
# Choose the most likely correct query:
|
134 |
+
# 1. Prefer queries with SELECT
|
135 |
+
# 2. Prefer longer queries (often more detailed)
|
136 |
+
# 3. Prefer first query if all else equal
|
137 |
+
select_queries = [q for q in unique_queries if re.search(r'SELECT\s+', q, re.IGNORECASE)]
|
138 |
+
|
139 |
+
if select_queries:
|
140 |
+
# Choose the longest SELECT query (likely most detailed)
|
141 |
+
best_query = max(select_queries, key=len)
|
142 |
+
elif unique_queries:
|
143 |
+
# If no SELECT queries, choose the longest query
|
144 |
+
best_query = max(unique_queries, key=len)
|
145 |
+
else:
|
146 |
+
# Fallback to the first query
|
147 |
+
best_query = queries[0]
|
148 |
+
else:
|
149 |
+
best_query = queries[0]
|
150 |
+
|
151 |
+
# Clean up the chosen query
|
152 |
+
best_query = best_query.strip()
|
153 |
+
if not best_query.endswith(';'):
|
154 |
+
best_query += ';'
|
155 |
+
|
156 |
+
# Final formatting to ensure consistent spacing
|
157 |
+
best_query = re.sub(r'\s+', ' ', best_query)
|
158 |
+
|
159 |
+
try:
|
160 |
+
# Use sqlparse to nicely format the SQL for display
|
161 |
+
formatted_sql = sqlparse.format(
|
162 |
+
best_query,
|
163 |
+
keyword_case='upper',
|
164 |
+
identifier_case='lower',
|
165 |
+
reindent=True,
|
166 |
+
indent_width=2
|
167 |
+
)
|
168 |
+
return formatted_sql
|
169 |
+
except:
|
170 |
+
return best_query
|
171 |
+
|
172 |
+
def process_input(question, table_context):
|
173 |
+
"""Function to process user input through the model and return formatted results"""
|
174 |
+
if not question.strip():
|
175 |
+
return "Please enter a question."
|
176 |
+
|
177 |
+
# Generate SQL from the question and context
|
178 |
+
raw_sql = generate_sql(question, table_context)
|
179 |
+
|
180 |
+
# Clean the SQL output
|
181 |
+
cleaned_sql = clean_sql_output(raw_sql)
|
182 |
+
|
183 |
+
if not cleaned_sql:
|
184 |
+
return "Sorry, I couldn't generate a valid SQL query. Please try rephrasing your question."
|
185 |
+
|
186 |
+
return cleaned_sql
|
187 |
+
|
188 |
+
# Sample table context examples for the example selector
|
189 |
+
example_contexts = [
|
190 |
+
# Example 1
|
191 |
+
"""
|
192 |
+
CREATE TABLE customers (
|
193 |
+
id INT PRIMARY KEY,
|
194 |
+
name VARCHAR(100),
|
195 |
+
email VARCHAR(100),
|
196 |
+
order_date DATE
|
197 |
+
);
|
198 |
+
""",
|
199 |
+
|
200 |
+
# Example 2
|
201 |
+
"""
|
202 |
+
CREATE TABLE products (
|
203 |
+
id INT PRIMARY KEY,
|
204 |
+
name VARCHAR(100),
|
205 |
+
category VARCHAR(50),
|
206 |
+
price DECIMAL(10,2),
|
207 |
+
stock_quantity INT
|
208 |
+
);
|
209 |
+
""",
|
210 |
+
|
211 |
+
# Example 3
|
212 |
+
"""
|
213 |
+
CREATE TABLE employees (
|
214 |
+
id INT PRIMARY KEY,
|
215 |
+
name VARCHAR(100),
|
216 |
+
department VARCHAR(50),
|
217 |
+
salary DECIMAL(10,2),
|
218 |
+
hire_date DATE
|
219 |
+
);
|
220 |
+
CREATE TABLE departments (
|
221 |
+
id INT PRIMARY KEY,
|
222 |
+
name VARCHAR(50),
|
223 |
+
manager_id INT,
|
224 |
+
budget DECIMAL(15,2)
|
225 |
+
);
|
226 |
+
"""
|
227 |
+
]
|
228 |
+
|
229 |
+
# Sample question examples
|
230 |
+
example_questions = [
|
231 |
+
"Get the names and emails of customers who placed an order in the last 30 days.",
|
232 |
+
"Find all products with less than 10 items in stock.",
|
233 |
+
"List all employees in the Sales department with a salary greater than 50000.",
|
234 |
+
"What is the total budget for departments with more than 5 employees?",
|
235 |
+
"Count how many products are in each category where the price is greater than 100."
|
236 |
+
]
|
237 |
+
|
238 |
+
# Create the Gradio interface
|
239 |
+
with gr.Blocks(title="Text to SQL Converter") as demo:
|
240 |
+
gr.Markdown("# Text to SQL Query Converter")
|
241 |
+
gr.Markdown("Enter your question and optional table context to generate an SQL query.")
|
242 |
+
|
243 |
+
|
244 |
+
|
245 |
+
# Launch the app
|
246 |
+
demo.launch()
|