Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,20 +1,24 @@
|
|
|
|
|
|
1 |
import gradio as gr
|
2 |
import pandas as pd
|
3 |
import torch
|
4 |
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline
|
5 |
|
6 |
-
# 1) Load
|
7 |
df = pd.read_csv('synthetic_profit.csv')
|
8 |
|
9 |
-
# 2)
|
10 |
MODEL_ID = "microsoft/tapex-base-finetuned-wikisql"
|
11 |
|
12 |
-
# 3)
|
13 |
device = 0 if torch.cuda.is_available() else -1
|
14 |
|
|
|
15 |
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
|
16 |
model = AutoModelForSeq2SeqLM.from_pretrained(MODEL_ID)
|
17 |
|
|
|
18 |
table_qa = pipeline(
|
19 |
"table-question-answering",
|
20 |
model=model,
|
@@ -23,23 +27,29 @@ table_qa = pipeline(
|
|
23 |
device=device,
|
24 |
)
|
25 |
|
26 |
-
#
|
27 |
def answer_profitability(question: str) -> str:
|
28 |
-
|
|
|
|
|
29 |
try:
|
30 |
out = table_qa(table=table, query=question)
|
31 |
return out.get("answer", "No answer found.")
|
32 |
except Exception as e:
|
33 |
return f"Error: {e}"
|
34 |
|
35 |
-
#
|
36 |
iface = gr.Interface(
|
37 |
fn=answer_profitability,
|
38 |
inputs=gr.Textbox(lines=2, placeholder="Ask a question about profitability…"),
|
39 |
outputs="text",
|
40 |
title="SAP Profitability Q&A (TAPEX-Base)",
|
41 |
-
description=
|
|
|
|
|
|
|
42 |
)
|
43 |
|
|
|
44 |
if __name__ == "__main__":
|
45 |
iface.launch(server_name="0.0.0.0", server_port=7860)
|
|
|
1 |
+
# app.py
|
2 |
+
|
3 |
import gradio as gr
|
4 |
import pandas as pd
|
5 |
import torch
|
6 |
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline
|
7 |
|
8 |
+
# 1) Load your synthetic profitability dataset
|
9 |
df = pd.read_csv('synthetic_profit.csv')
|
10 |
|
11 |
+
# 2) Choose the publicly available TAPEX WikiSQL model
|
12 |
MODEL_ID = "microsoft/tapex-base-finetuned-wikisql"
|
13 |
|
14 |
+
# 3) Set device: GPU if available, else CPU
|
15 |
device = 0 if torch.cuda.is_available() else -1
|
16 |
|
17 |
+
# 4) Load tokenizer and model
|
18 |
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
|
19 |
model = AutoModelForSeq2SeqLM.from_pretrained(MODEL_ID)
|
20 |
|
21 |
+
# 5) Build the table-question-answering pipeline
|
22 |
table_qa = pipeline(
|
23 |
"table-question-answering",
|
24 |
model=model,
|
|
|
27 |
device=device,
|
28 |
)
|
29 |
|
30 |
+
# 6) Define the QA function, casting all cells to strings to avoid float issues
|
31 |
def answer_profitability(question: str) -> str:
|
32 |
+
# Cast entire DataFrame to string
|
33 |
+
df_str = df.astype(str)
|
34 |
+
table = df_str.to_dict(orient="records")
|
35 |
try:
|
36 |
out = table_qa(table=table, query=question)
|
37 |
return out.get("answer", "No answer found.")
|
38 |
except Exception as e:
|
39 |
return f"Error: {e}"
|
40 |
|
41 |
+
# 7) Define Gradio interface
|
42 |
iface = gr.Interface(
|
43 |
fn=answer_profitability,
|
44 |
inputs=gr.Textbox(lines=2, placeholder="Ask a question about profitability…"),
|
45 |
outputs="text",
|
46 |
title="SAP Profitability Q&A (TAPEX-Base)",
|
47 |
+
description=(
|
48 |
+
"Free-form questions on the synthetic profitability dataset, "
|
49 |
+
"powered end-to-end by microsoft/tapex-base-finetuned-wikisql."
|
50 |
+
)
|
51 |
)
|
52 |
|
53 |
+
# 8) Launch the app
|
54 |
if __name__ == "__main__":
|
55 |
iface.launch(server_name="0.0.0.0", server_port=7860)
|