add csv ingestion and prompt templates
Browse files- .gitignore +3 -0
- app.py +33 -33
- employees +0 -0
- llm.py +93 -0
- sql.py +102 -0
.gitignore
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
__pycache__
|
2 |
+
./__pycache__
|
3 |
+
*/__pycache
|
app.py
CHANGED
@@ -1,35 +1,33 @@
|
|
1 |
import streamlit as st
|
2 |
-
from
|
3 |
-
|
4 |
-
|
5 |
-
repo_ir = "Qwen/Qwen2.5-Coder-1.5B-Instruct-GGUF"
|
6 |
-
llm = Llama.from_pretrained(
|
7 |
-
repo_id=repo_ir,
|
8 |
-
filename="qwen2.5-coder-1.5b-instruct-q8_0.gguf",
|
9 |
-
verbose=True,
|
10 |
-
use_mmap=True,
|
11 |
-
use_mlock=True,
|
12 |
-
n_threads=4,
|
13 |
-
n_threads_batch=4,
|
14 |
-
n_ctx=8000,
|
15 |
-
)
|
16 |
-
print(f"{repo_ir} loaded successfully. ✅")
|
17 |
-
|
18 |
-
|
19 |
-
# Streamed response emulator
|
20 |
-
def response_generator(messages):
|
21 |
-
completion = llm.create_chat_completion(
|
22 |
-
messages, max_tokens=2048, stream=True, temperature=0.7, top_p=0.95
|
23 |
-
)
|
24 |
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
|
|
|
30 |
|
31 |
st.title("CSV TO SQL")
|
32 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
33 |
# Initialize chat history
|
34 |
if "messages" not in st.session_state:
|
35 |
st.session_state.messages = []
|
@@ -47,14 +45,16 @@ if prompt := st.chat_input("What is up?"):
|
|
47 |
with st.chat_message("user"):
|
48 |
st.markdown(prompt)
|
49 |
|
50 |
-
messages = [{"role": "system", "content": "You are a helpful assistant"}]
|
51 |
-
|
52 |
-
for val in st.session_state.messages:
|
53 |
-
messages.append(val)
|
54 |
-
|
55 |
-
messages.append({"role": "user", "content": prompt})
|
56 |
# Display assistant response in chat message container
|
57 |
with st.chat_message("assistant"):
|
58 |
-
response = st.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
59 |
# Add assistant response to chat history
|
60 |
st.session_state.messages.append({"role": "assistant", "content": response})
|
|
|
1 |
import streamlit as st
|
2 |
+
from llm import load_llm, response_generator
|
3 |
+
from sql import csv_to_sqlite
|
4 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
5 |
|
6 |
+
# repo_id = "Qwen/Qwen2.5-Coder-1.5B-Instruct-GGUF"
|
7 |
+
repo_id = "Qwen/Qwen2.5-0.5B-Instruct-GGUF"
|
8 |
+
# filename="qwen2.5-coder-1.5b-instruct-q8_0.gguf"
|
9 |
+
filename = "qwen2.5-0.5b-instruct-q8_0.gguf"
|
10 |
|
11 |
+
llm = load_llm(repo_id, filename)
|
12 |
|
13 |
st.title("CSV TO SQL")
|
14 |
|
15 |
+
with st.expander("Upload CSV"):
|
16 |
+
csv_file = st.file_uploader(
|
17 |
+
"CSV",
|
18 |
+
)
|
19 |
+
db_name = st.text_input("DB Name")
|
20 |
+
table_name = st.text_input("Table Name")
|
21 |
+
if st.button("Save"):
|
22 |
+
if csv_file and db_name and table_name:
|
23 |
+
st.session_state.db_name = db_name
|
24 |
+
st.session_state.table_name = table_name
|
25 |
+
|
26 |
+
csv_to_sqlite(csv_file, db_name, table_name)
|
27 |
+
st.write("Saved ✅")
|
28 |
+
else:
|
29 |
+
st.write("Please enter all values")
|
30 |
+
|
31 |
# Initialize chat history
|
32 |
if "messages" not in st.session_state:
|
33 |
st.session_state.messages = []
|
|
|
45 |
with st.chat_message("user"):
|
46 |
st.markdown(prompt)
|
47 |
|
|
|
|
|
|
|
|
|
|
|
|
|
48 |
# Display assistant response in chat message container
|
49 |
with st.chat_message("assistant"):
|
50 |
+
response = st.write(
|
51 |
+
response_generator(
|
52 |
+
db_name=st.session_state.db_name,
|
53 |
+
table_name=st.session_state.table_name,
|
54 |
+
llm=llm,
|
55 |
+
messages=st.session_state.messages,
|
56 |
+
question=prompt,
|
57 |
+
)
|
58 |
+
)
|
59 |
# Add assistant response to chat history
|
60 |
st.session_state.messages.append({"role": "assistant", "content": response})
|
employees
ADDED
Binary file (8.19 kB). View file
|
|
llm.py
ADDED
@@ -0,0 +1,93 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
from llama_cpp import Llama
|
3 |
+
from sql import get_table_schema
|
4 |
+
|
5 |
+
|
6 |
+
@st.cache_resource()
|
7 |
+
def load_llm(repo_id, filename):
|
8 |
+
llm = Llama.from_pretrained(
|
9 |
+
repo_id=repo_id,
|
10 |
+
filename=filename,
|
11 |
+
verbose=True,
|
12 |
+
use_mmap=True,
|
13 |
+
use_mlock=True,
|
14 |
+
n_threads=4,
|
15 |
+
n_threads_batch=4,
|
16 |
+
n_ctx=8000,
|
17 |
+
)
|
18 |
+
print(f"{repo_id} loaded successfully. ✅")
|
19 |
+
return llm
|
20 |
+
|
21 |
+
|
22 |
+
def generate_llm_prompt(table_name, table_schema):
|
23 |
+
"""
|
24 |
+
Generates a prompt to provide context about a table's schema for LLM to convert natural language to SQL.
|
25 |
+
|
26 |
+
Args:
|
27 |
+
table_name (str): The name of the table.
|
28 |
+
table_schema (list): A list of tuples where each tuple contains information about the columns in the table.
|
29 |
+
|
30 |
+
Returns:
|
31 |
+
str: The generated prompt to be used by the LLM.
|
32 |
+
"""
|
33 |
+
prompt = f"""You are an expert in writing SQL queries for relational databases.
|
34 |
+
You will be provided with a database schema and a natural
|
35 |
+
language question, and your task is to generate an accurate SQL query.
|
36 |
+
|
37 |
+
The database has a table named '{table_name}' with the following schema:\n\n"""
|
38 |
+
|
39 |
+
prompt += "Columns:\n"
|
40 |
+
|
41 |
+
for col in table_schema:
|
42 |
+
column_name = col[1]
|
43 |
+
column_type = col[2]
|
44 |
+
prompt += f"- {column_name} ({column_type})\n"
|
45 |
+
|
46 |
+
prompt += "\nPlease generate a SQL query based on the following natural language question. ONLY return the SQL query."
|
47 |
+
|
48 |
+
return prompt
|
49 |
+
|
50 |
+
|
51 |
+
def generate_sql_query(question, table_name, db_name):
|
52 |
+
pass
|
53 |
+
# table_name = 'movies'
|
54 |
+
# db_name = 'movies_db.db'
|
55 |
+
# table_schema = get_table_schema(db_name, table_name)
|
56 |
+
# llm_prompt = generate_llm_prompt(table_name, table_schema)
|
57 |
+
# user_prompt = """Question: {question}"""
|
58 |
+
# response = completion(
|
59 |
+
# api_key=OPENAI_API_KEY,
|
60 |
+
# model="gpt-4o-mini",
|
61 |
+
# messages=[
|
62 |
+
# ,
|
63 |
+
# {"content": user_prompt.format(question=question),"role": "user"}],
|
64 |
+
# max_tokens=1000
|
65 |
+
# )
|
66 |
+
# answer = response.choices[0].message.content
|
67 |
+
|
68 |
+
# query = answer.replace("```sql", "").replace("```", "")
|
69 |
+
# query = query.strip()
|
70 |
+
# return query
|
71 |
+
|
72 |
+
|
73 |
+
# Streamed response emulator
|
74 |
+
def response_generator(llm, messages, question, table_name, db_name):
|
75 |
+
table_schema = get_table_schema(db_name, table_name)
|
76 |
+
llm_prompt = generate_llm_prompt(table_name, table_schema)
|
77 |
+
user_prompt = """Question: {question}"""
|
78 |
+
|
79 |
+
messages = [{"content": llm_prompt.format(table_name=table_name), "role": "system"}]
|
80 |
+
|
81 |
+
for val in st.session_state.messages:
|
82 |
+
messages.append(val)
|
83 |
+
|
84 |
+
messages.append({"role": "user", "content": user_prompt})
|
85 |
+
|
86 |
+
response = llm.create_chat_completion(
|
87 |
+
messages, max_tokens=2048, temperature=0.7, top_p=0.95
|
88 |
+
)
|
89 |
+
answer = response["choices"][0].message.content
|
90 |
+
|
91 |
+
query = answer.replace("```sql", "").replace("```", "")
|
92 |
+
query = query.strip()
|
93 |
+
return query
|
sql.py
ADDED
@@ -0,0 +1,102 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import pandas as pd
|
2 |
+
import sqlite3
|
3 |
+
|
4 |
+
|
5 |
+
def csv_to_sqlite(csv_file, db_name, table_name):
|
6 |
+
# Read the CSV file into a pandas DataFrame
|
7 |
+
df = pd.read_csv(csv_file)
|
8 |
+
|
9 |
+
# Connect to the SQLite database (it will create the database file if it doesn't exist)
|
10 |
+
conn = sqlite3.connect(db_name)
|
11 |
+
cursor = conn.cursor()
|
12 |
+
|
13 |
+
# Infer the schema based on the DataFrame columns and data types
|
14 |
+
def create_table_from_df(df, table_name):
|
15 |
+
# Get column names and types
|
16 |
+
col_types = []
|
17 |
+
for col in df.columns:
|
18 |
+
dtype = df[col].dtype
|
19 |
+
if dtype == "int64":
|
20 |
+
col_type = "INTEGER"
|
21 |
+
elif dtype == "float64":
|
22 |
+
col_type = "REAL"
|
23 |
+
else:
|
24 |
+
col_type = "TEXT"
|
25 |
+
col_types.append(f'"{col}" {col_type}')
|
26 |
+
|
27 |
+
# Create the table schema
|
28 |
+
col_definitions = ", ".join(col_types)
|
29 |
+
create_table_query = (
|
30 |
+
f"CREATE TABLE IF NOT EXISTS {table_name} ({col_definitions});"
|
31 |
+
)
|
32 |
+
# print(create_table_query)
|
33 |
+
|
34 |
+
# Execute the table creation query
|
35 |
+
cursor.execute(create_table_query)
|
36 |
+
print(f"Table '{table_name}' created with schema: {col_definitions}")
|
37 |
+
|
38 |
+
# Create table schema
|
39 |
+
create_table_from_df(df, table_name)
|
40 |
+
|
41 |
+
# Insert CSV data into the SQLite table
|
42 |
+
df.to_sql(table_name, conn, if_exists="replace", index=False)
|
43 |
+
|
44 |
+
# Commit and close the connection
|
45 |
+
conn.commit()
|
46 |
+
conn.close()
|
47 |
+
print(f"Data loaded into '{table_name}' table in '{db_name}' SQLite database.")
|
48 |
+
|
49 |
+
|
50 |
+
def run_sql_query(db_name, query):
|
51 |
+
"""
|
52 |
+
Executes a SQL query on a SQLite database and returns the results.
|
53 |
+
|
54 |
+
Args:
|
55 |
+
db_name (str): The name of the SQLite database file.
|
56 |
+
query (str): The SQL query to run.
|
57 |
+
|
58 |
+
Returns:
|
59 |
+
list: Query result as a list of tuples, or an empty list if no results or error occurred.
|
60 |
+
"""
|
61 |
+
try:
|
62 |
+
# Connect to the SQLite database
|
63 |
+
conn = sqlite3.connect(db_name)
|
64 |
+
cursor = conn.cursor()
|
65 |
+
|
66 |
+
# Execute the SQL query
|
67 |
+
cursor.execute(query)
|
68 |
+
|
69 |
+
# Fetch all results
|
70 |
+
results = cursor.fetchall()
|
71 |
+
|
72 |
+
# Close the connection
|
73 |
+
conn.close()
|
74 |
+
|
75 |
+
# Return results or an empty list if no results were found
|
76 |
+
return results if results else []
|
77 |
+
|
78 |
+
except sqlite3.Error as e:
|
79 |
+
print(f"An error occurred while executing the query: {e}")
|
80 |
+
return []
|
81 |
+
|
82 |
+
|
83 |
+
def get_table_schema(db_name, table_name):
|
84 |
+
"""
|
85 |
+
Retrieves the schema (columns and data types) for a given table in the SQLite database.
|
86 |
+
|
87 |
+
Args:
|
88 |
+
db_name (str): The name of the SQLite database file.
|
89 |
+
table_name (str): The name of the table.
|
90 |
+
|
91 |
+
Returns:
|
92 |
+
list: A list of tuples with column name, data type, and other info.
|
93 |
+
"""
|
94 |
+
conn = sqlite3.connect(db_name)
|
95 |
+
cursor = conn.cursor()
|
96 |
+
|
97 |
+
# Use PRAGMA to get the table schema
|
98 |
+
cursor.execute(f"PRAGMA table_info({table_name});")
|
99 |
+
schema = cursor.fetchall()
|
100 |
+
|
101 |
+
conn.close()
|
102 |
+
return schema
|