Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -5,84 +5,68 @@ import duckdb
|
|
5 |
import openai
|
6 |
|
7 |
# 1) Load your OpenAI key from the Space’s Secrets
|
8 |
-
|
9 |
-
if not
|
10 |
raise RuntimeError("Missing OPENAI_API_KEY secret in your Space settings")
|
11 |
-
openai.api_key = OPENAI_KEY
|
12 |
|
13 |
-
# 2) Load your
|
14 |
-
df = pd.read_csv(
|
15 |
-
conn = duckdb.connect(
|
16 |
-
conn.register(
|
17 |
|
18 |
-
# 3) Build a one-line schema
|
19 |
schema = ", ".join(df.columns)
|
20 |
|
21 |
-
# 4) Function to
|
22 |
def generate_sql(question: str) -> str:
|
23 |
-
|
24 |
-
f"You are an expert SQL generator for
|
25 |
-
|
26 |
-
"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
27 |
)
|
28 |
-
try:
|
29 |
-
resp = openai.ChatCompletion.create(
|
30 |
-
model="gpt-3.5-turbo",
|
31 |
-
messages=[
|
32 |
-
{"role": "system", "content": system_prompt},
|
33 |
-
{"role": "user", "content": question},
|
34 |
-
],
|
35 |
-
temperature=0.0,
|
36 |
-
max_tokens=150,
|
37 |
-
)
|
38 |
-
except Exception as e:
|
39 |
-
# Catch network/auth errors
|
40 |
-
raise RuntimeError(f"OpenAI API error: {e}")
|
41 |
-
|
42 |
sql = resp.choices[0].message.content.strip()
|
43 |
-
#
|
44 |
if sql.startswith("```") and sql.endswith("```"):
|
45 |
sql = "\n".join(sql.splitlines()[1:-1])
|
46 |
return sql
|
47 |
|
48 |
-
# 5)
|
49 |
def answer_profitability(question: str) -> str:
|
50 |
-
# a)
|
51 |
try:
|
52 |
sql = generate_sql(question)
|
53 |
except Exception as e:
|
54 |
-
return f"❌
|
55 |
|
56 |
-
# b)
|
57 |
try:
|
58 |
result_df = conn.execute(sql).df()
|
59 |
except Exception as e:
|
60 |
-
return
|
61 |
-
f"❌ **SQL Execution Error**\n{e}\n\n"
|
62 |
-
f"**Generated SQL**\n```sql\n{sql}\n```"
|
63 |
-
)
|
64 |
|
65 |
-
# c)
|
66 |
if result_df.empty:
|
67 |
-
return f"No
|
68 |
-
|
69 |
-
# single-cell → scalar
|
70 |
if result_df.shape == (1,1):
|
71 |
return str(result_df.iat[0,0])
|
72 |
-
|
73 |
-
# multi-cell → markdown table
|
74 |
return result_df.to_markdown(index=False)
|
75 |
|
76 |
-
# 6) Gradio interface
|
77 |
iface = gr.Interface(
|
78 |
fn=answer_profitability,
|
79 |
-
inputs=gr.Textbox(lines=2, placeholder="Ask a question
|
80 |
-
outputs=gr.Textbox(lines=8, placeholder="Answer
|
81 |
-
title="SAP Profitability Q&A (OpenAI
|
82 |
-
description=
|
83 |
-
"Uses OpenAI’s GPT-3.5-Turbo to translate your question into SQL, "
|
84 |
-
"executes it against the `sap` table in DuckDB, and returns the result."
|
85 |
-
),
|
86 |
allow_flagging="never",
|
87 |
)
|
88 |
|
|
|
5 |
import openai
|
6 |
|
7 |
# 1) Load your OpenAI key from the Space’s Secrets
|
8 |
+
openai.api_key = os.getenv("OPENAI_API_KEY")
|
9 |
+
if not openai.api_key:
|
10 |
raise RuntimeError("Missing OPENAI_API_KEY secret in your Space settings")
|
|
|
11 |
|
12 |
+
# 2) Load your CSV into DuckDB
|
13 |
+
df = pd.read_csv("synthetic_profit.csv")
|
14 |
+
conn = duckdb.connect(":memory:")
|
15 |
+
conn.register("sap", df)
|
16 |
|
17 |
+
# 3) Build a one-line schema string for prompting
|
18 |
schema = ", ".join(df.columns)
|
19 |
|
20 |
+
# 4) Function to call OpenAI and get back a SQL string
|
21 |
def generate_sql(question: str) -> str:
|
22 |
+
system = (
|
23 |
+
f"You are an expert SQL generator for DuckDB table `sap` with columns: {schema}. "
|
24 |
+
"Translate the user’s question into a valid SQL query. "
|
25 |
+
"Return ONLY the SQL, no explanation."
|
26 |
+
)
|
27 |
+
resp = openai.ChatCompletion.create(
|
28 |
+
model="gpt-3.5-turbo",
|
29 |
+
messages=[
|
30 |
+
{"role": "system", "content": system},
|
31 |
+
{"role": "user", "content": question}
|
32 |
+
],
|
33 |
+
temperature=0.0,
|
34 |
+
max_tokens=150,
|
35 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
36 |
sql = resp.choices[0].message.content.strip()
|
37 |
+
# Strip ``` if present
|
38 |
if sql.startswith("```") and sql.endswith("```"):
|
39 |
sql = "\n".join(sql.splitlines()[1:-1])
|
40 |
return sql
|
41 |
|
42 |
+
# 5) The function Gradio will call
|
43 |
def answer_profitability(question: str) -> str:
|
44 |
+
# a) Generate the SQL
|
45 |
try:
|
46 |
sql = generate_sql(question)
|
47 |
except Exception as e:
|
48 |
+
return f"❌ OpenAI error:\n{e}"
|
49 |
|
50 |
+
# b) Execute it in DuckDB
|
51 |
try:
|
52 |
result_df = conn.execute(sql).df()
|
53 |
except Exception as e:
|
54 |
+
return f"❌ SQL error:\n{e}\n\nGenerated SQL:\n```sql\n{sql}\n```"
|
|
|
|
|
|
|
55 |
|
56 |
+
# c) Format the output
|
57 |
if result_df.empty:
|
58 |
+
return f"No results.\n\nSQL was:\n```sql\n{sql}\n```"
|
|
|
|
|
59 |
if result_df.shape == (1,1):
|
60 |
return str(result_df.iat[0,0])
|
|
|
|
|
61 |
return result_df.to_markdown(index=False)
|
62 |
|
63 |
+
# 6) Gradio interface—**note the explicit outputs=…**
|
64 |
iface = gr.Interface(
|
65 |
fn=answer_profitability,
|
66 |
+
inputs=gr.Textbox(lines=2, placeholder="Ask a question…", label="Question"),
|
67 |
+
outputs=gr.Textbox(lines=8, placeholder="Answer appears here", label="Answer"),
|
68 |
+
title="SAP Profitability Q&A (OpenAI→SQL→DuckDB)",
|
69 |
+
description="Enter a natural-language question and get back the numeric result or table.",
|
|
|
|
|
|
|
70 |
allow_flagging="never",
|
71 |
)
|
72 |
|