Update app.py
Browse files
app.py
CHANGED
@@ -1,57 +1,93 @@
|
|
|
|
1 |
import torch
|
2 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
3 |
import sqlparse
|
4 |
-
import
|
|
|
|
|
|
|
|
|
|
|
5 |
|
6 |
model_name = "defog/llama-3-sqlcoder-8b"
|
7 |
-
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
8 |
|
9 |
-
#
|
10 |
-
|
11 |
-
model_name,
|
12 |
-
trust_remote_code=True,
|
13 |
-
device_map={"": "cpu"},
|
14 |
-
torch_dtype=torch.float32
|
15 |
-
)
|
16 |
|
17 |
-
#
|
18 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
19 |
|
20 |
Generate a SQL query to answer this question: `{question}`
|
21 |
|
22 |
DDL statements:
|
23 |
|
24 |
CREATE TABLE expenses (
|
25 |
-
id INTEGER PRIMARY KEY,
|
26 |
-
date DATE NOT NULL,
|
27 |
-
amount DECIMAL(10,2) NOT NULL,
|
28 |
-
category VARCHAR(50) NOT NULL,
|
29 |
-
description TEXT,
|
30 |
-
payment_method VARCHAR(20),
|
31 |
-
user_id INTEGER
|
32 |
);
|
33 |
|
34 |
CREATE TABLE categories (
|
35 |
-
id INTEGER PRIMARY KEY,
|
36 |
-
name VARCHAR(50) UNIQUE NOT NULL,
|
37 |
-
description TEXT
|
38 |
);
|
39 |
|
40 |
CREATE TABLE users (
|
41 |
-
id INTEGER PRIMARY KEY,
|
42 |
-
username VARCHAR(50) UNIQUE NOT NULL,
|
43 |
-
email VARCHAR(100) UNIQUE NOT NULL,
|
44 |
-
created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
|
45 |
);
|
46 |
|
47 |
CREATE TABLE budgets (
|
48 |
-
id INTEGER PRIMARY KEY,
|
49 |
-
user_id INTEGER,
|
50 |
-
category VARCHAR(50),
|
51 |
-
amount DECIMAL(10,2) NOT NULL,
|
52 |
-
period VARCHAR(20) DEFAULT 'monthly',
|
53 |
-
start_date DATE,
|
54 |
-
end_date DATE
|
55 |
);
|
56 |
|
57 |
-- expenses.user_id can be joined with users.id
|
@@ -63,38 +99,78 @@ The following SQL query best answers the question `{question}`:
|
|
63 |
```sql
|
64 |
"""
|
65 |
|
66 |
-
# Main function
|
67 |
def generate_query(question):
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
generated_ids = model.generate(
|
72 |
-
**inputs,
|
73 |
-
num_return_sequences=1,
|
74 |
-
eos_token_id=tokenizer.eos_token_id,
|
75 |
-
pad_token_id=tokenizer.eos_token_id,
|
76 |
-
max_new_tokens=400,
|
77 |
-
do_sample=False,
|
78 |
-
num_beams=1,
|
79 |
-
temperature=0.0,
|
80 |
-
top_p=1,
|
81 |
-
)
|
82 |
-
|
83 |
-
output = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
84 |
try:
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
91 |
iface = gr.Interface(
|
92 |
-
fn=
|
93 |
-
inputs=gr.Textbox(
|
94 |
-
|
95 |
-
|
96 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
97 |
)
|
98 |
|
99 |
if __name__ == "__main__":
|
100 |
-
iface.launch()
|
|
|
1 |
+
import gradio as gr
|
2 |
import torch
|
3 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
4 |
import sqlparse
|
5 |
+
import psutil
|
6 |
+
import os
|
7 |
+
|
8 |
+
# Check available memory
|
9 |
+
def get_available_memory():
|
10 |
+
return psutil.virtual_memory().available
|
11 |
|
12 |
model_name = "defog/llama-3-sqlcoder-8b"
|
|
|
13 |
|
14 |
+
# Initialize tokenizer
|
15 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
|
|
|
|
|
|
|
|
|
|
16 |
|
17 |
+
# CPU-compatible model loading
|
18 |
+
def load_model():
|
19 |
+
try:
|
20 |
+
available_memory = get_available_memory()
|
21 |
+
print(f"Available memory: {available_memory / 1e9:.1f} GB")
|
22 |
+
|
23 |
+
# For CPU deployment, we'll use float32 or float16 without quantization
|
24 |
+
if available_memory > 16e9: # 16GB+ RAM
|
25 |
+
print("Loading model in float16...")
|
26 |
+
model = AutoModelForCausalLM.from_pretrained(
|
27 |
+
model_name,
|
28 |
+
trust_remote_code=True,
|
29 |
+
torch_dtype=torch.float16,
|
30 |
+
device_map="cpu",
|
31 |
+
use_cache=True,
|
32 |
+
low_cpu_mem_usage=True
|
33 |
+
)
|
34 |
+
else:
|
35 |
+
print("Loading model in float32 with low memory usage...")
|
36 |
+
model = AutoModelForCausalLM.from_pretrained(
|
37 |
+
model_name,
|
38 |
+
trust_remote_code=True,
|
39 |
+
device_map="cpu",
|
40 |
+
use_cache=True,
|
41 |
+
low_cpu_mem_usage=True,
|
42 |
+
torch_dtype=torch.float32
|
43 |
+
)
|
44 |
+
|
45 |
+
return model
|
46 |
+
except Exception as e:
|
47 |
+
print(f"Error loading model: {e}")
|
48 |
+
return None
|
49 |
+
|
50 |
+
# Load model (this will take some time on first run)
|
51 |
+
print("Loading model... This may take a few minutes on CPU.")
|
52 |
+
model = load_model()
|
53 |
+
|
54 |
+
prompt_template = """<|begin_of_text|><|start_header_id|>user<|end_header_id|>
|
55 |
|
56 |
Generate a SQL query to answer this question: `{question}`
|
57 |
|
58 |
DDL statements:
|
59 |
|
60 |
CREATE TABLE expenses (
|
61 |
+
id INTEGER PRIMARY KEY, -- Unique ID for each expense
|
62 |
+
date DATE NOT NULL, -- Date when the expense occurred
|
63 |
+
amount DECIMAL(10,2) NOT NULL, -- Amount spent
|
64 |
+
category VARCHAR(50) NOT NULL, -- Category of expense (food, transport, utilities, etc.)
|
65 |
+
description TEXT, -- Optional description of the expense
|
66 |
+
payment_method VARCHAR(20), -- How the payment was made (cash, credit_card, debit_card, bank_transfer)
|
67 |
+
user_id INTEGER -- ID of the user who made the expense
|
68 |
);
|
69 |
|
70 |
CREATE TABLE categories (
|
71 |
+
id INTEGER PRIMARY KEY, -- Unique ID for each category
|
72 |
+
name VARCHAR(50) UNIQUE NOT NULL, -- Category name (food, transport, utilities, entertainment, etc.)
|
73 |
+
description TEXT -- Optional description of the category
|
74 |
);
|
75 |
|
76 |
CREATE TABLE users (
|
77 |
+
id INTEGER PRIMARY KEY, -- Unique ID for each user
|
78 |
+
username VARCHAR(50) UNIQUE NOT NULL, -- Username
|
79 |
+
email VARCHAR(100) UNIQUE NOT NULL, -- Email address
|
80 |
+
created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP -- When the user account was created
|
81 |
);
|
82 |
|
83 |
CREATE TABLE budgets (
|
84 |
+
id INTEGER PRIMARY KEY, -- Unique ID for each budget
|
85 |
+
user_id INTEGER, -- ID of the user who set the budget
|
86 |
+
category VARCHAR(50), -- Category for which budget is set
|
87 |
+
amount DECIMAL(10,2) NOT NULL, -- Budget amount
|
88 |
+
period VARCHAR(20) DEFAULT 'monthly', -- Budget period (daily, weekly, monthly, yearly)
|
89 |
+
start_date DATE, -- Budget start date
|
90 |
+
end_date DATE -- Budget end date
|
91 |
);
|
92 |
|
93 |
-- expenses.user_id can be joined with users.id
|
|
|
99 |
```sql
|
100 |
"""
|
101 |
|
|
|
102 |
def generate_query(question):
|
103 |
+
if model is None:
|
104 |
+
return "Error: Model not loaded properly"
|
105 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
106 |
try:
|
107 |
+
updated_prompt = prompt_template.format(question=question)
|
108 |
+
inputs = tokenizer(updated_prompt, return_tensors="pt")
|
109 |
+
|
110 |
+
# Generate on CPU
|
111 |
+
with torch.no_grad():
|
112 |
+
generated_ids = model.generate(
|
113 |
+
**inputs,
|
114 |
+
num_return_sequences=1,
|
115 |
+
eos_token_id=tokenizer.eos_token_id,
|
116 |
+
pad_token_id=tokenizer.eos_token_id,
|
117 |
+
max_new_tokens=400,
|
118 |
+
do_sample=False,
|
119 |
+
num_beams=1,
|
120 |
+
temperature=0.0,
|
121 |
+
top_p=1,
|
122 |
+
)
|
123 |
+
|
124 |
+
outputs = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)
|
125 |
+
|
126 |
+
# Extract SQL from output
|
127 |
+
if "```sql" in outputs[0]:
|
128 |
+
sql_part = outputs[0].split("```sql")[1].split("```")[0].strip()
|
129 |
+
else:
|
130 |
+
# Fallback extraction
|
131 |
+
sql_part = outputs[0].split("The following SQL query best answers the question")[1].strip()
|
132 |
+
if sql_part.startswith("`"):
|
133 |
+
sql_part = sql_part[1:]
|
134 |
+
if "```" in sql_part:
|
135 |
+
sql_part = sql_part.split("```")[0].strip()
|
136 |
+
|
137 |
+
# Clean up the SQL
|
138 |
+
if sql_part.endswith(";"):
|
139 |
+
sql_part = sql_part[:-1]
|
140 |
+
|
141 |
+
# Format the SQL
|
142 |
+
formatted_sql = sqlparse.format(sql_part, reindent=True, keyword_case='upper')
|
143 |
+
return formatted_sql
|
144 |
+
|
145 |
+
except Exception as e:
|
146 |
+
return f"Error generating query: {str(e)}"
|
147 |
+
|
148 |
+
def gradio_interface(question):
|
149 |
+
if not question.strip():
|
150 |
+
return "Please enter a question."
|
151 |
+
|
152 |
+
return generate_query(question)
|
153 |
+
|
154 |
+
# Create Gradio interface
|
155 |
iface = gr.Interface(
|
156 |
+
fn=gradio_interface,
|
157 |
+
inputs=gr.Textbox(
|
158 |
+
label="Question",
|
159 |
+
placeholder="Enter your question (e.g., 'Show me all expenses for food category')",
|
160 |
+
lines=3
|
161 |
+
),
|
162 |
+
outputs=gr.Code(label="Generated SQL Query", language="sql"),
|
163 |
+
title="SQL Query Generator",
|
164 |
+
description="Generate SQL queries from natural language questions about expense tracking database.",
|
165 |
+
examples=[
|
166 |
+
["Show me all expenses for food category"],
|
167 |
+
["What's the total amount spent on transport this month?"],
|
168 |
+
["Insert a new expense of 50 dollars for groceries on 2024-01-15"],
|
169 |
+
["Find users who spent more than 1000 dollars total"],
|
170 |
+
["Show me the budget vs actual spending for each category"]
|
171 |
+
],
|
172 |
+
cache_examples=False
|
173 |
)
|
174 |
|
175 |
if __name__ == "__main__":
|
176 |
+
iface.launch()
|