Spaces:
Sleeping
Sleeping
Kiddo
commited on
Commit
•
b6fc9a7
1
Parent(s):
e90ad78
Vision Processor
Browse files
app.py
CHANGED
@@ -1,4 +1,80 @@
|
|
1 |
import streamlit as st
|
|
|
|
|
|
|
2 |
|
3 |
-
|
4 |
-
st.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import streamlit as st
|
2 |
+
import torch
|
3 |
+
from PIL import Image
|
4 |
+
from transformers import AutoProcessor, AutoModelForImageTextToText
|
5 |
|
6 |
+
# Set page configuration
|
7 |
+
st.set_page_config(page_title="Llama 3.2 Vision Model", page_icon="???")
|
8 |
+
|
9 |
+
# Title and description
|
10 |
+
st.title("Llama 3.2 Vision Model Inference")
|
11 |
+
st.write("Upload an image and provide a prompt to get model insights!")
|
12 |
+
|
13 |
+
# Load model and processor (consider caching to improve performance)
|
14 |
+
@st.cache_resource
|
15 |
+
def load_model():
|
16 |
+
try:
|
17 |
+
processor = AutoProcessor.from_pretrained("meta-llama/Llama-3.2-90B-Vision-Instruct")
|
18 |
+
model = AutoModelForImageTextToText.from_pretrained("meta-llama/Llama-3.2-90B-Vision-Instruct")
|
19 |
+
return processor, model
|
20 |
+
except Exception as e:
|
21 |
+
st.error(f"Error loading model: {e}")
|
22 |
+
return None, None
|
23 |
+
|
24 |
+
# Inference function
|
25 |
+
def generate_response(image, prompt):
|
26 |
+
processor, model = load_model()
|
27 |
+
|
28 |
+
if not processor or not model:
|
29 |
+
return "Model could not be loaded."
|
30 |
+
|
31 |
+
try:
|
32 |
+
# Prepare inputs
|
33 |
+
inputs = processor(images=image, text=prompt, return_tensors="pt")
|
34 |
+
|
35 |
+
# Generate response
|
36 |
+
with torch.no_grad():
|
37 |
+
outputs = model.generate(**inputs)
|
38 |
+
|
39 |
+
# Decode the response
|
40 |
+
response = processor.decode(outputs[0], skip_special_tokens=True)
|
41 |
+
return response
|
42 |
+
|
43 |
+
except Exception as e:
|
44 |
+
st.error(f"Error during inference: {e}")
|
45 |
+
return "An error occurred during image processing."
|
46 |
+
|
47 |
+
# Sidebar for user inputs
|
48 |
+
st.sidebar.header("Image and Prompt")
|
49 |
+
|
50 |
+
# Image uploader
|
51 |
+
uploaded_file = st.sidebar.file_uploader("Choose an image...", type=["jpg", "jpeg", "png"])
|
52 |
+
|
53 |
+
# Prompt input
|
54 |
+
prompt = st.sidebar.text_input("Enter your prompt:",
|
55 |
+
placeholder="Describe what you want to know about the image")
|
56 |
+
|
57 |
+
# Main content area
|
58 |
+
if uploaded_file is not None:
|
59 |
+
# Display uploaded image
|
60 |
+
image = Image.open(uploaded_file)
|
61 |
+
st.image(image, caption="Uploaded Image", use_column_width=True)
|
62 |
+
|
63 |
+
# Generate button
|
64 |
+
if st.sidebar.button("Generate Response"):
|
65 |
+
if prompt:
|
66 |
+
# Show loading spinner
|
67 |
+
with st.spinner("Generating response..."):
|
68 |
+
response = generate_response(image, prompt)
|
69 |
+
|
70 |
+
# Display response
|
71 |
+
st.subheader("Model Response")
|
72 |
+
st.write(response)
|
73 |
+
else:
|
74 |
+
st.warning("Please enter a prompt!")
|
75 |
+
else:
|
76 |
+
st.info("Upload an image and enter a prompt to get started!")
|
77 |
+
|
78 |
+
# Additional error handling and information
|
79 |
+
st.sidebar.markdown("---")
|
80 |
+
st.sidebar.info("Note: Model performance depends on image quality and prompt specificity.")
|