Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,171 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
import os
|
3 |
+
import nest_asyncio
|
4 |
+
import re
|
5 |
+
from pathlib import Path
|
6 |
+
import typing as t
|
7 |
+
import base64
|
8 |
+
from mimetypes import guess_type
|
9 |
+
from llama_parse import LlamaParse
|
10 |
+
from llama_index.core.schema import TextNode
|
11 |
+
from llama_index.core import VectorStoreIndex, StorageContext, load_index_from_storage, Settings
|
12 |
+
from llama_index.embeddings.openai import OpenAIEmbedding
|
13 |
+
from llama_index.llms.openai import OpenAI
|
14 |
+
from llama_index.core.query_engine import CustomQueryEngine
|
15 |
+
from llama_index.multi_modal_llms.openai import OpenAIMultiModal
|
16 |
+
from llama_index.core.prompts import PromptTemplate
|
17 |
+
from llama_index.core.schema import ImageNode
|
18 |
+
from llama_index.core.base.response.schema import Response
|
19 |
+
from typing import Optional, List
|
20 |
+
|
21 |
+
nest_asyncio.apply()
|
22 |
+
|
23 |
+
# Setting API keys
|
24 |
+
os.environ["OPENAI_API_KEY"] = os.getenv('OPENAI_API_KEY')
|
25 |
+
os.environ["LLAMA_CLOUD_API_KEY"] = os.getenv('LLAMA_CLOUD_API_KEY')
|
26 |
+
|
27 |
+
# Initialize Streamlit app
|
28 |
+
st.title("Medical Knowledge Base & Query System")
|
29 |
+
st.sidebar.title("Settings")
|
30 |
+
|
31 |
+
# User input for file upload
|
32 |
+
st.sidebar.subheader("Upload Knowledge Base")
|
33 |
+
uploaded_file = st.sidebar.file_uploader("Upload a medical text book (image)", type=["jpg", "png"])
|
34 |
+
|
35 |
+
# Initialize the parser
|
36 |
+
parser = LlamaParse(
|
37 |
+
result_type="markdown",
|
38 |
+
parsing_instruction="You are given medical text book on medicine",
|
39 |
+
use_vendor_multimodal_model=True,
|
40 |
+
vendor_multimodal_model_name="gpt-4o-mini-2024-07-18",
|
41 |
+
show_progress=True,
|
42 |
+
verbose=True,
|
43 |
+
invalidate_cache=True,
|
44 |
+
do_not_cache=True,
|
45 |
+
num_workers=8,
|
46 |
+
language="en"
|
47 |
+
)
|
48 |
+
|
49 |
+
# Function to encode image to data URL
|
50 |
+
def local_image_to_data_url(image_path):
|
51 |
+
mime_type, _ = guess_type(image_path)
|
52 |
+
if mime_type is None:
|
53 |
+
mime_type = 'image/png'
|
54 |
+
with open(image_path, "rb") as image_file:
|
55 |
+
base64_encoded_data = base64.b64encode(image_file.read()).decode('utf-8')
|
56 |
+
return f"data:{mime_type};base64,{base64_encoded_data}"
|
57 |
+
|
58 |
+
# Upload and process file
|
59 |
+
if uploaded_file:
|
60 |
+
st.sidebar.write("Processing file...")
|
61 |
+
file_path = f"files/{uploaded_file.name}"
|
62 |
+
with open(file_path, "wb") as f:
|
63 |
+
f.write(uploaded_file.read())
|
64 |
+
|
65 |
+
# Parse the uploaded image
|
66 |
+
md_json_objs = parser.get_json_result([file_path])
|
67 |
+
image_dicts = parser.get_images(md_json_objs, download_path="data_images")
|
68 |
+
|
69 |
+
# Extract and display parsed information
|
70 |
+
st.write("File successfully processed!")
|
71 |
+
st.write(f"Processed file: {uploaded_file.name}")
|
72 |
+
|
73 |
+
# Function to get sorted image files
|
74 |
+
def get_page_number(file_name):
|
75 |
+
match = re.search(r"-page-(\d+)\.jpg$", str(file_name))
|
76 |
+
if match:
|
77 |
+
return int(match.group(1))
|
78 |
+
return 0
|
79 |
+
|
80 |
+
def _get_sorted_image_files(image_dir):
|
81 |
+
raw_files = [f for f in list(Path(image_dir).iterdir()) if f.is_file()]
|
82 |
+
sorted_files = sorted(raw_files, key=get_page_number)
|
83 |
+
return sorted_files
|
84 |
+
|
85 |
+
def get_text_nodes(md_json_objs, image_dir) -> t.List[TextNode]:
|
86 |
+
nodes = []
|
87 |
+
for result in md_json_objs:
|
88 |
+
json_dicts = result["pages"]
|
89 |
+
document_name = result["file_path"].split('/')[-1]
|
90 |
+
docs = [doc["md"] for doc in json_dicts]
|
91 |
+
image_files = _get_sorted_image_files(image_dir)
|
92 |
+
for idx, doc in enumerate(docs):
|
93 |
+
node = TextNode(
|
94 |
+
text=doc,
|
95 |
+
metadata={"image_path": str(image_files[idx]), "page_num": idx + 1, "document_name": document_name},
|
96 |
+
)
|
97 |
+
nodes.append(node)
|
98 |
+
return nodes
|
99 |
+
|
100 |
+
# Load text nodes
|
101 |
+
text_nodes = get_text_nodes(md_json_objs, "data_images")
|
102 |
+
|
103 |
+
# Setup index and LLM
|
104 |
+
embed_model = OpenAIEmbedding(model="text-embedding-3-large")
|
105 |
+
llm = OpenAI("gpt-4o-mini-2024-07-18")
|
106 |
+
Settings.llm = llm
|
107 |
+
Settings.embed_model = embed_model
|
108 |
+
|
109 |
+
if not os.path.exists("storage_manuals"):
|
110 |
+
index = VectorStoreIndex(text_nodes, embed_model=embed_model)
|
111 |
+
index.storage_context.persist(persist_dir="./storage_manuals")
|
112 |
+
else:
|
113 |
+
ctx = StorageContext.from_defaults(persist_dir="./storage_manuals")
|
114 |
+
index = load_index_from_storage(ctx)
|
115 |
+
|
116 |
+
retriever = index.as_retriever()
|
117 |
+
|
118 |
+
# Query input
|
119 |
+
st.subheader("Ask a Question")
|
120 |
+
query_text = st.text_input("Enter your query:")
|
121 |
+
uploaded_query_image = st.file_uploader("Upload a query image (if any):", type=["jpg", "png"])
|
122 |
+
|
123 |
+
# Encode query image if provided
|
124 |
+
encoded_image_url = None
|
125 |
+
if uploaded_query_image:
|
126 |
+
query_image_path = f"query_images/{uploaded_query_image.name}"
|
127 |
+
with open(query_image_path, "wb") as img_file:
|
128 |
+
img_file.write(uploaded_query_image.read())
|
129 |
+
encoded_image_url = local_image_to_data_url(query_image_path)
|
130 |
+
|
131 |
+
# Setup query engine
|
132 |
+
QA_PROMPT_TMPL = """
|
133 |
+
You are a friendly medical chatbot designed to assist users by providing accurate and detailed responses to medical questions based on information from medical books.
|
134 |
+
|
135 |
+
### Context:
|
136 |
+
---------------------
|
137 |
+
{context_str}
|
138 |
+
---------------------
|
139 |
+
|
140 |
+
### Query Text:
|
141 |
+
{query_str}
|
142 |
+
|
143 |
+
### Query Image:
|
144 |
+
---------------------
|
145 |
+
{encoded_image_url}
|
146 |
+
---------------------
|
147 |
+
|
148 |
+
### Answer:
|
149 |
+
"""
|
150 |
+
QA_PROMPT = PromptTemplate(QA_PROMPT_TMPL)
|
151 |
+
gpt_4o_mm = OpenAIMultiModal(model="gpt-4o-mini-2024-07-18")
|
152 |
+
|
153 |
+
class MultimodalQueryEngine(CustomQueryEngine):
|
154 |
+
def __init__(self, qa_prompt, retriever, multi_modal_llm, node_postprocessors=[]):
|
155 |
+
super().__init__(qa_prompt=qa_prompt, retriever=retriever, multi_modal_llm=multi_modal_llm, node_postprocessors=node_postprocessors)
|
156 |
+
|
157 |
+
def custom_query(self, query_str):
|
158 |
+
nodes = self.retriever.retrieve(query_str)
|
159 |
+
image_nodes = [NodeWithScore(node=ImageNode(image_path=n.node.metadata["image_path"])) for n in nodes]
|
160 |
+
ctx_str = "\n\n".join([r.node.get_content().strip() for r in nodes])
|
161 |
+
fmt_prompt = self.qa_prompt.format(context_str=ctx_str, query_str=query_str, encoded_image_url=encoded_image_url)
|
162 |
+
llm_response = self.multi_modal_llm.complete(prompt=fmt_prompt, image_documents=[image_node.node for image_node in image_nodes])
|
163 |
+
return Response(response=str(llm_response), source_nodes=nodes, metadata={"text_nodes": text_nodes, "image_nodes": image_nodes})
|
164 |
+
|
165 |
+
query_engine = MultimodalQueryEngine(QA_PROMPT, retriever, gpt_4o_mm)
|
166 |
+
|
167 |
+
# Handle query
|
168 |
+
if query_text:
|
169 |
+
st.write("Querying...")
|
170 |
+
response = query_engine.custom_query(query_text)
|
171 |
+
st.markdown(response.response)
|