Spaces:
Sleeping
Sleeping
HusnaManakkot
commited on
Commit
•
0c5ce5c
1
Parent(s):
4f13759
Update app.py
Browse files
app.py
CHANGED
@@ -5,35 +5,49 @@ from datasets import load_dataset
|
|
5 |
# Load the Spider dataset
|
6 |
spider_dataset = load_dataset("spider", split='train') # Load a subset of the dataset
|
7 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
8 |
# Load tokenizer and model
|
9 |
tokenizer = AutoTokenizer.from_pretrained("mrm8488/t5-base-finetuned-wikiSQL")
|
10 |
model = AutoModelForSeq2SeqLM.from_pretrained("mrm8488/t5-base-finetuned-wikiSQL")
|
11 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
12 |
def generate_sql_from_user_input(query):
|
13 |
# Generate SQL for the user's query
|
14 |
input_text = "translate English to SQL: " + query
|
15 |
inputs = tokenizer(input_text, return_tensors="pt", padding=True)
|
16 |
outputs = model.generate(**inputs, max_length=512)
|
17 |
sql_query = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
18 |
-
return sql_query
|
19 |
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
if item['question'].lower() == nl_query.lower():
|
24 |
-
return item['query']
|
25 |
-
return "No matching SQL query found in the Spider dataset."
|
26 |
|
27 |
# Create a Gradio interface
|
28 |
interface = gr.Interface(
|
29 |
-
fn=
|
30 |
-
"Generated SQL Query": generate_sql_from_user_input(query),
|
31 |
-
"Matching SQL Query from Spider Dataset": find_matching_sql(query)
|
32 |
-
},
|
33 |
inputs=gr.Textbox(label="Enter your natural language query"),
|
34 |
-
outputs=
|
35 |
-
title="NL to SQL
|
36 |
-
description="This
|
37 |
)
|
38 |
|
39 |
# Launch the app
|
|
|
5 |
# Load the Spider dataset
|
6 |
spider_dataset = load_dataset("spider", split='train') # Load a subset of the dataset
|
7 |
|
8 |
+
# Extract schema information from the dataset
|
9 |
+
db_table_names = set()
|
10 |
+
column_names = set()
|
11 |
+
for item in spider_dataset:
|
12 |
+
db_id = item['db_id']
|
13 |
+
for table in item['table_names']:
|
14 |
+
db_table_names.add((db_id, table))
|
15 |
+
for column in item['column_names']:
|
16 |
+
column_names.add(column[1])
|
17 |
+
|
18 |
# Load tokenizer and model
|
19 |
tokenizer = AutoTokenizer.from_pretrained("mrm8488/t5-base-finetuned-wikiSQL")
|
20 |
model = AutoModelForSeq2SeqLM.from_pretrained("mrm8488/t5-base-finetuned-wikiSQL")
|
21 |
|
22 |
+
def post_process_sql_query(sql_query):
|
23 |
+
# Modify the SQL query to match the dataset's schema
|
24 |
+
for db_id, table_name in db_table_names:
|
25 |
+
if "TABLE" in sql_query:
|
26 |
+
sql_query = sql_query.replace("TABLE", table_name)
|
27 |
+
break # Assuming only one table is referenced in the query
|
28 |
+
for column_name in column_names:
|
29 |
+
if "COLUMN" in sql_query:
|
30 |
+
sql_query = sql_query.replace("COLUMN", column_name, 1)
|
31 |
+
return sql_query
|
32 |
+
|
33 |
def generate_sql_from_user_input(query):
|
34 |
# Generate SQL for the user's query
|
35 |
input_text = "translate English to SQL: " + query
|
36 |
inputs = tokenizer(input_text, return_tensors="pt", padding=True)
|
37 |
outputs = model.generate(**inputs, max_length=512)
|
38 |
sql_query = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
|
|
39 |
|
40 |
+
# Post-process the SQL query to match the dataset's schema
|
41 |
+
sql_query = post_process_sql_query(sql_query)
|
42 |
+
return sql_query
|
|
|
|
|
|
|
43 |
|
44 |
# Create a Gradio interface
|
45 |
interface = gr.Interface(
|
46 |
+
fn=generate_sql_from_user_input,
|
|
|
|
|
|
|
47 |
inputs=gr.Textbox(label="Enter your natural language query"),
|
48 |
+
outputs=gr.Textbox(label="Generated SQL Query"),
|
49 |
+
title="NL to SQL using Spider Dataset",
|
50 |
+
description="This interface generates an SQL query from your natural language input based on the Spider dataset."
|
51 |
)
|
52 |
|
53 |
# Launch the app
|