Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -33,14 +33,12 @@ processor = AutoProcessor.from_pretrained("Qwen/Qwen2-VL-2B-Instruct", trust_rem
|
|
33 |
|
34 |
@spaces.GPU()
|
35 |
def process_pdf_and_query(pdf_file, user_query):
|
36 |
-
|
37 |
-
images = convert_from_path(pdf_file.name) # pdf_file.name gives the file path
|
38 |
num_images = len(images)
|
39 |
|
40 |
-
# Indexing the PDF in RAG
|
41 |
RAG.index(
|
42 |
input_path=pdf_file.name,
|
43 |
-
index_name="image_index",
|
44 |
store_collection_with_index=False,
|
45 |
overwrite=True
|
46 |
)
|
@@ -65,7 +63,7 @@ def process_pdf_and_query(pdf_file, user_query):
|
|
65 |
}
|
66 |
]
|
67 |
|
68 |
-
|
69 |
text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
|
70 |
image_inputs, video_inputs = process_vision_info(messages)
|
71 |
inputs = processor(
|
@@ -77,7 +75,6 @@ def process_pdf_and_query(pdf_file, user_query):
|
|
77 |
)
|
78 |
inputs = inputs.to("cuda")
|
79 |
|
80 |
-
# Generate the output response
|
81 |
generated_ids = model.generate(**inputs, max_new_tokens=50)
|
82 |
generated_ids_trimmed = [
|
83 |
out_ids[len(in_ids):] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
|
@@ -88,18 +85,67 @@ def process_pdf_and_query(pdf_file, user_query):
|
|
88 |
|
89 |
return output_text[0], num_images
|
90 |
|
91 |
-
# Define the Gradio Interface
|
92 |
-
pdf_input = gr.File(label="Upload PDF") # Single PDF file input
|
93 |
-
query_input = gr.Textbox(label="Enter your query", placeholder="Ask a question about the PDF") # User query input
|
94 |
-
output_text = gr.Textbox(label="Model Answer") # Output for the model's answer
|
95 |
-
output_images = gr.Textbox(label="Number of Images in PDF") # Output for number of images
|
96 |
|
97 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
98 |
demo = gr.Interface(
|
99 |
fn=process_pdf_and_query,
|
100 |
-
inputs=[pdf_input, query_input],
|
101 |
-
outputs=[output_text, output_images],
|
102 |
-
title="Multimodal RAG with Image Query - By Pejman Ebrahimi"
|
|
|
|
|
|
|
|
|
103 |
)
|
104 |
|
105 |
-
demo
|
|
|
|
|
|
|
|
|
|
33 |
|
34 |
@spaces.GPU()
|
35 |
def process_pdf_and_query(pdf_file, user_query):
|
36 |
+
images = convert_from_path(pdf_file.name)
|
|
|
37 |
num_images = len(images)
|
38 |
|
|
|
39 |
RAG.index(
|
40 |
input_path=pdf_file.name,
|
41 |
+
index_name="image_index",
|
42 |
store_collection_with_index=False,
|
43 |
overwrite=True
|
44 |
)
|
|
|
63 |
}
|
64 |
]
|
65 |
|
66 |
+
|
67 |
text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
|
68 |
image_inputs, video_inputs = process_vision_info(messages)
|
69 |
inputs = processor(
|
|
|
75 |
)
|
76 |
inputs = inputs.to("cuda")
|
77 |
|
|
|
78 |
generated_ids = model.generate(**inputs, max_new_tokens=50)
|
79 |
generated_ids_trimmed = [
|
80 |
out_ids[len(in_ids):] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
|
|
|
85 |
|
86 |
return output_text[0], num_images
|
87 |
|
|
|
|
|
|
|
|
|
|
|
88 |
|
89 |
+
css = """
|
90 |
+
.duplicate-button {
|
91 |
+
background-color: #6272a4;
|
92 |
+
color: white;
|
93 |
+
font-weight: bold;
|
94 |
+
border-radius: 5px;
|
95 |
+
margin-top: 20px;
|
96 |
+
padding: 10px;
|
97 |
+
text-align: center;
|
98 |
+
}
|
99 |
+
.gradio-container {
|
100 |
+
background-color: #282a36;
|
101 |
+
color: #f8f8f2;
|
102 |
+
font-family: 'Courier New', Courier, monospace;
|
103 |
+
padding: 20px;
|
104 |
+
border-radius: 10px;
|
105 |
+
box-shadow: 0 4px 6px rgba(0, 0, 0, 0.1);
|
106 |
+
}
|
107 |
+
"""
|
108 |
+
|
109 |
+
explanation = """
|
110 |
+
### Multimodal RAG with Image Query
|
111 |
+
This demo showcases the **Multimodal RAG (Retriever-Augmented Generation)** model. The RAG system integrates retrieval and generation, allowing it to retrieve relevant information from a multimodal database (like PDFs with text and images) and then generate detailed responses.
|
112 |
+
|
113 |
+
We use **ColPali**, a state-of-the-art multimodal retriever, combined with the **Byaldi** library from **answer.ai**, which simplifies using ColPali. The language model used for generating answers is **Qwen/Qwen2-VL-2B-Instruct**, a powerful vision-language model capable of understanding both text and images.
|
114 |
+
"""
|
115 |
+
|
116 |
+
footer = """
|
117 |
+
<div style="text-align: center; margin-top: 20px;">
|
118 |
+
<a href="https://www.linkedin.com/in/pejman-ebrahimi-4a60151a7/" target="_blank">LinkedIn</a> |
|
119 |
+
<a href="https://github.com/arad1367" target="_blank">GitHub</a> |
|
120 |
+
<a href="https://arad1367.pythonanywhere.com/" target="_blank">Live demo of my PhD defense</a> |
|
121 |
+
<a href="https://huggingface.co/Qwen/Qwen2-VL-2B-Instruct" target="_blank">Qwen/Qwen2-VL-2B-Instruct</a> |
|
122 |
+
<a href="https://github.com/AnswerDotAI/byaldi" target="_blank">Byaldi</a> |
|
123 |
+
<a href="https://github.com/illuin-tech/colpali" target="_blank">ColPali</a>
|
124 |
+
<br>
|
125 |
+
Made with π by Pejman Ebrahimi
|
126 |
+
</div>
|
127 |
+
"""
|
128 |
+
|
129 |
+
pdf_input = gr.File(label="Upload PDF")
|
130 |
+
query_input = gr.Textbox(label="Enter your query", placeholder="Ask a question about the PDF")
|
131 |
+
output_text = gr.Textbox(label="Model Answer")
|
132 |
+
output_images = gr.Textbox(label="Number of Images in PDF")
|
133 |
+
duplicate_button = gr.DuplicateButton(value="Duplicate Space for private use", elem_classes="duplicate-button")
|
134 |
+
|
135 |
+
# Launch the Gradio app
|
136 |
demo = gr.Interface(
|
137 |
fn=process_pdf_and_query,
|
138 |
+
inputs=[pdf_input, query_input],
|
139 |
+
outputs=[output_text, output_images],
|
140 |
+
title="Multimodal RAG with Image Query - By Pejman Ebrahimi - Please like the space if it is useful",
|
141 |
+
theme='freddyaboulton/dracula_revamped',
|
142 |
+
css=css,
|
143 |
+
description=explanation,
|
144 |
+
allow_flagging="auto"
|
145 |
)
|
146 |
|
147 |
+
with demo:
|
148 |
+
gr.HTML(footer)
|
149 |
+
duplicate_button
|
150 |
+
|
151 |
+
demo.launch(debug=True)
|