Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -0,0 +1,71 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
from transformers import Pix2StructForConditionalGeneration, Pix2StructProcessor
|
3 |
+
import gradio as gr
|
4 |
+
from PIL import Image
|
5 |
+
|
6 |
+
# Use a publicly available high-capacity model.
|
7 |
+
# For instance, we use "google/pix2struct-docvqa-large".
|
8 |
+
# (If you need a different model or a private one, adjust accordingly and add authentication if necessary.)
|
9 |
+
model_name = "google/pix2struct-docvqa-large"
|
10 |
+
|
11 |
+
model = Pix2StructForConditionalGeneration.from_pretrained(model_name)
|
12 |
+
processor = Pix2StructProcessor.from_pretrained(model_name)
|
13 |
+
|
14 |
+
def solve_problem(image):
|
15 |
+
try:
|
16 |
+
# Ensure the image is in RGB.
|
17 |
+
image = image.convert("RGB")
|
18 |
+
|
19 |
+
# Preprocess image and text prompt.
|
20 |
+
inputs = processor(
|
21 |
+
images=[image],
|
22 |
+
text="Solve the following problem:",
|
23 |
+
return_tensors="pt",
|
24 |
+
max_patches=2048
|
25 |
+
)
|
26 |
+
|
27 |
+
# Generate prediction.
|
28 |
+
predictions = model.generate(
|
29 |
+
**inputs,
|
30 |
+
max_new_tokens=200,
|
31 |
+
early_stopping=True,
|
32 |
+
num_beams=4,
|
33 |
+
temperature=0.2
|
34 |
+
)
|
35 |
+
|
36 |
+
# Decode the prompt (input IDs) and the generated output.
|
37 |
+
problem_text = processor.decode(
|
38 |
+
inputs["input_ids"][0],
|
39 |
+
skip_special_tokens=True,
|
40 |
+
clean_up_tokenization_spaces=True
|
41 |
+
)
|
42 |
+
solution = processor.decode(
|
43 |
+
predictions[0],
|
44 |
+
skip_special_tokens=True,
|
45 |
+
clean_up_tokenization_spaces=True
|
46 |
+
)
|
47 |
+
return f"Problem: {problem_text}\nSolution: {solution}"
|
48 |
+
except Exception as e:
|
49 |
+
return f"Error processing image: {str(e)}"
|
50 |
+
|
51 |
+
# Set up the Gradio interface.
|
52 |
+
iface = gr.Interface(
|
53 |
+
fn=solve_problem,
|
54 |
+
inputs=gr.Image(type="pil", label="Upload Your Problem Image", image_mode="RGB"),
|
55 |
+
outputs=gr.Textbox(label="Solution", show_copy_button=True),
|
56 |
+
title="Problem Solver with Pix2Struct",
|
57 |
+
description=(
|
58 |
+
"Upload an image (for example, a handwritten math or logic problem) "
|
59 |
+
"and get a solution generated by a high-capacity Pix2Struct model.\n\n"
|
60 |
+
"Note: For best results on domain-specific tasks, consider fine-tuning on your own dataset."
|
61 |
+
),
|
62 |
+
examples=[
|
63 |
+
["example_problem1.png"],
|
64 |
+
["example_problem2.jpg"]
|
65 |
+
],
|
66 |
+
theme="soft",
|
67 |
+
allow_flagging="never"
|
68 |
+
)
|
69 |
+
|
70 |
+
if __name__ == "__main__":
|
71 |
+
iface.launch()
|