Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -1,6 +1,6 @@
|
|
1 |
import streamlit as st
|
2 |
from langchain_google_genai import ChatGoogleGenerativeAI
|
3 |
-
from langchain_core.messages import
|
4 |
from langchain_core.output_parsers import PydanticOutputParser
|
5 |
from langchain_core.prompts import PromptTemplate
|
6 |
from langchain.chains import LLMChain
|
@@ -17,7 +17,6 @@ from langchain_community.embeddings import HuggingFaceInferenceAPIEmbeddings
|
|
17 |
from langchain_community.vectorstores import FAISS
|
18 |
from langchain_text_splitters import RecursiveCharacterTextSplitter
|
19 |
|
20 |
-
# Environment Variable Checks
|
21 |
GOOGLE_API_KEY = os.environ.get("GOOGLE_API_KEY")
|
22 |
HUGGINGFACE_ACCESS_TOKEN = os.environ.get("HUGGINGFACE_ACCESS_TOKEN")
|
23 |
|
@@ -29,10 +28,10 @@ if not HUGGINGFACE_ACCESS_TOKEN:
|
|
29 |
st.error("β HUGGINGFACE_ACCESS_TOKEN not found.")
|
30 |
st.stop()
|
31 |
|
32 |
-
# Initialize LLM and Embeddings
|
33 |
llm = ChatGoogleGenerativeAI(
|
34 |
model="gemini-1.5-pro",
|
35 |
-
google_api_key=GOOGLE_API_KEY
|
|
|
36 |
)
|
37 |
|
38 |
embeddings = HuggingFaceInferenceAPIEmbeddings(
|
@@ -40,7 +39,6 @@ embeddings = HuggingFaceInferenceAPIEmbeddings(
|
|
40 |
model_name="BAAI/bge-small-en-v1.5"
|
41 |
)
|
42 |
|
43 |
-
# Pydantic Models for Structured Output
|
44 |
class KeyPoint(BaseModel):
|
45 |
point: str = Field(description="A key point extracted from the document.")
|
46 |
|
@@ -51,10 +49,8 @@ class DocumentAnalysis(BaseModel):
|
|
51 |
key_points: List[KeyPoint] = Field(description="List of key points from the document.")
|
52 |
summary: Summary = Field(description="Summary of the document.")
|
53 |
|
54 |
-
# Output Parser
|
55 |
parser = PydanticOutputParser(pydantic_object=DocumentAnalysis)
|
56 |
|
57 |
-
# Prompt Template
|
58 |
prompt_template = """
|
59 |
Analyze the following text and extract key points and a summary.
|
60 |
{format_instructions}
|
@@ -66,19 +62,15 @@ prompt = PromptTemplate(
|
|
66 |
partial_variables={"format_instructions": parser.get_format_instructions()}
|
67 |
)
|
68 |
|
69 |
-
# LLM Chain
|
70 |
chain = LLMChain(llm=llm, prompt=prompt, output_parser=parser)
|
71 |
|
72 |
-
# Text Analysis Function
|
73 |
def analyze_text_structured(text):
|
74 |
return chain.run(text=text)
|
75 |
|
76 |
-
# PDF Text Extraction
|
77 |
def extract_text_from_pdf(pdf_file):
|
78 |
pdf_reader = PyPDF2.PdfReader(pdf_file)
|
79 |
return "".join(page.extract_text() for page in pdf_reader.pages)
|
80 |
|
81 |
-
# JSON to Readable Text
|
82 |
def json_to_text(analysis):
|
83 |
text_output = "=== Summary ===\n" + f"{analysis.summary.summary}\n\n"
|
84 |
text_output += "=== Key Points ===\n"
|
@@ -86,7 +78,6 @@ def json_to_text(analysis):
|
|
86 |
text_output += f"{i}. {key_point.point}\n"
|
87 |
return text_output
|
88 |
|
89 |
-
# PDF Report Generation (Updated)
|
90 |
def create_pdf_report(analysis):
|
91 |
pdf = FPDF()
|
92 |
pdf.add_page()
|
@@ -95,13 +86,11 @@ def create_pdf_report(analysis):
|
|
95 |
pdf.cell(200, 10, txt=f"Generated on: {datetime.now().strftime('%Y-%m-%d %H:%M:%S')}", ln=True, align='C')
|
96 |
pdf.multi_cell(0, 10, txt=json_to_text(analysis))
|
97 |
|
98 |
-
# Use BytesIO to create a bytes-like object
|
99 |
pdf_bytes = io.BytesIO()
|
100 |
pdf.output(pdf_bytes, dest='S')
|
101 |
pdf_bytes.seek(0)
|
102 |
return pdf_bytes.getvalue()
|
103 |
|
104 |
-
# Word Report Generation (Updated)
|
105 |
def create_word_report(analysis):
|
106 |
doc = Document()
|
107 |
doc.add_heading('PDF Analysis Report', 0)
|
@@ -109,16 +98,13 @@ def create_word_report(analysis):
|
|
109 |
doc.add_heading('Analysis', level=1)
|
110 |
doc.add_paragraph(json_to_text(analysis))
|
111 |
|
112 |
-
# Use BytesIO to create a bytes-like object
|
113 |
docx_bytes = io.BytesIO()
|
114 |
doc.save(docx_bytes)
|
115 |
docx_bytes.seek(0)
|
116 |
return docx_bytes.getvalue()
|
117 |
|
118 |
-
# Streamlit Page Configuration
|
119 |
st.set_page_config(page_title="Chat With PDF", page_icon="π")
|
120 |
|
121 |
-
# Custom CSS
|
122 |
def local_css():
|
123 |
st.markdown("""
|
124 |
<style>
|
@@ -203,7 +189,6 @@ def local_css():
|
|
203 |
|
204 |
local_css()
|
205 |
|
206 |
-
# Session State Initialization
|
207 |
if "current_file" not in st.session_state:
|
208 |
st.session_state.current_file = None
|
209 |
if "pdf_summary" not in st.session_state:
|
@@ -219,14 +204,12 @@ if "vectorstore" not in st.session_state:
|
|
219 |
if "messages" not in st.session_state:
|
220 |
st.session_state.messages = []
|
221 |
|
222 |
-
# Main App Layout
|
223 |
st.markdown('<div class="main-header">', unsafe_allow_html=True)
|
224 |
st.markdown('<div class="flag-stripe"></div>', unsafe_allow_html=True)
|
225 |
st.title("π Chat With PDF")
|
226 |
st.caption("Your AI-powered Document Analyzer")
|
227 |
st.markdown('</div>', unsafe_allow_html=True)
|
228 |
|
229 |
-
# PDF Upload and Analysis Section
|
230 |
with st.container():
|
231 |
st.markdown('<div class="card animate-fadeIn">', unsafe_allow_html=True)
|
232 |
uploaded_file = st.file_uploader("Upload a PDF file", type="pdf")
|
@@ -276,7 +259,6 @@ with st.container():
|
|
276 |
)
|
277 |
st.markdown('</div>', unsafe_allow_html=True)
|
278 |
|
279 |
-
# Document Chat Section
|
280 |
if "vectorstore" in st.session_state and st.session_state.vectorstore is not None:
|
281 |
st.subheader("Chat with the Document")
|
282 |
|
@@ -296,8 +278,7 @@ if "vectorstore" in st.session_state and st.session_state.vectorstore is not Non
|
|
296 |
context = "\n".join([doc.page_content for doc in docs])
|
297 |
|
298 |
messages = [
|
299 |
-
|
300 |
-
HumanMessage(content=f"Context: {context}\n\nQuestion: {prompt}")
|
301 |
]
|
302 |
|
303 |
response = llm.invoke(messages)
|
@@ -305,7 +286,6 @@ if "vectorstore" in st.session_state and st.session_state.vectorstore is not Non
|
|
305 |
|
306 |
st.session_state.messages.append({"role": "assistant", "content": response.content})
|
307 |
|
308 |
-
# Footer
|
309 |
st.markdown(
|
310 |
f'<div class="footer">Analysis Time: {st.session_state.analysis_time:.1f}s | Powered by Google Generative AI</div>',
|
311 |
unsafe_allow_html=True
|
|
|
1 |
import streamlit as st
|
2 |
from langchain_google_genai import ChatGoogleGenerativeAI
|
3 |
+
from langchain_core.messages import HumanMessage
|
4 |
from langchain_core.output_parsers import PydanticOutputParser
|
5 |
from langchain_core.prompts import PromptTemplate
|
6 |
from langchain.chains import LLMChain
|
|
|
17 |
from langchain_community.vectorstores import FAISS
|
18 |
from langchain_text_splitters import RecursiveCharacterTextSplitter
|
19 |
|
|
|
20 |
GOOGLE_API_KEY = os.environ.get("GOOGLE_API_KEY")
|
21 |
HUGGINGFACE_ACCESS_TOKEN = os.environ.get("HUGGINGFACE_ACCESS_TOKEN")
|
22 |
|
|
|
28 |
st.error("β HUGGINGFACE_ACCESS_TOKEN not found.")
|
29 |
st.stop()
|
30 |
|
|
|
31 |
llm = ChatGoogleGenerativeAI(
|
32 |
model="gemini-1.5-pro",
|
33 |
+
google_api_key=GOOGLE_API_KEY,
|
34 |
+
convert_system_message_to_human=True
|
35 |
)
|
36 |
|
37 |
embeddings = HuggingFaceInferenceAPIEmbeddings(
|
|
|
39 |
model_name="BAAI/bge-small-en-v1.5"
|
40 |
)
|
41 |
|
|
|
42 |
class KeyPoint(BaseModel):
|
43 |
point: str = Field(description="A key point extracted from the document.")
|
44 |
|
|
|
49 |
key_points: List[KeyPoint] = Field(description="List of key points from the document.")
|
50 |
summary: Summary = Field(description="Summary of the document.")
|
51 |
|
|
|
52 |
parser = PydanticOutputParser(pydantic_object=DocumentAnalysis)
|
53 |
|
|
|
54 |
prompt_template = """
|
55 |
Analyze the following text and extract key points and a summary.
|
56 |
{format_instructions}
|
|
|
62 |
partial_variables={"format_instructions": parser.get_format_instructions()}
|
63 |
)
|
64 |
|
|
|
65 |
chain = LLMChain(llm=llm, prompt=prompt, output_parser=parser)
|
66 |
|
|
|
67 |
def analyze_text_structured(text):
|
68 |
return chain.run(text=text)
|
69 |
|
|
|
70 |
def extract_text_from_pdf(pdf_file):
|
71 |
pdf_reader = PyPDF2.PdfReader(pdf_file)
|
72 |
return "".join(page.extract_text() for page in pdf_reader.pages)
|
73 |
|
|
|
74 |
def json_to_text(analysis):
|
75 |
text_output = "=== Summary ===\n" + f"{analysis.summary.summary}\n\n"
|
76 |
text_output += "=== Key Points ===\n"
|
|
|
78 |
text_output += f"{i}. {key_point.point}\n"
|
79 |
return text_output
|
80 |
|
|
|
81 |
def create_pdf_report(analysis):
|
82 |
pdf = FPDF()
|
83 |
pdf.add_page()
|
|
|
86 |
pdf.cell(200, 10, txt=f"Generated on: {datetime.now().strftime('%Y-%m-%d %H:%M:%S')}", ln=True, align='C')
|
87 |
pdf.multi_cell(0, 10, txt=json_to_text(analysis))
|
88 |
|
|
|
89 |
pdf_bytes = io.BytesIO()
|
90 |
pdf.output(pdf_bytes, dest='S')
|
91 |
pdf_bytes.seek(0)
|
92 |
return pdf_bytes.getvalue()
|
93 |
|
|
|
94 |
def create_word_report(analysis):
|
95 |
doc = Document()
|
96 |
doc.add_heading('PDF Analysis Report', 0)
|
|
|
98 |
doc.add_heading('Analysis', level=1)
|
99 |
doc.add_paragraph(json_to_text(analysis))
|
100 |
|
|
|
101 |
docx_bytes = io.BytesIO()
|
102 |
doc.save(docx_bytes)
|
103 |
docx_bytes.seek(0)
|
104 |
return docx_bytes.getvalue()
|
105 |
|
|
|
106 |
st.set_page_config(page_title="Chat With PDF", page_icon="π")
|
107 |
|
|
|
108 |
def local_css():
|
109 |
st.markdown("""
|
110 |
<style>
|
|
|
189 |
|
190 |
local_css()
|
191 |
|
|
|
192 |
if "current_file" not in st.session_state:
|
193 |
st.session_state.current_file = None
|
194 |
if "pdf_summary" not in st.session_state:
|
|
|
204 |
if "messages" not in st.session_state:
|
205 |
st.session_state.messages = []
|
206 |
|
|
|
207 |
st.markdown('<div class="main-header">', unsafe_allow_html=True)
|
208 |
st.markdown('<div class="flag-stripe"></div>', unsafe_allow_html=True)
|
209 |
st.title("π Chat With PDF")
|
210 |
st.caption("Your AI-powered Document Analyzer")
|
211 |
st.markdown('</div>', unsafe_allow_html=True)
|
212 |
|
|
|
213 |
with st.container():
|
214 |
st.markdown('<div class="card animate-fadeIn">', unsafe_allow_html=True)
|
215 |
uploaded_file = st.file_uploader("Upload a PDF file", type="pdf")
|
|
|
259 |
)
|
260 |
st.markdown('</div>', unsafe_allow_html=True)
|
261 |
|
|
|
262 |
if "vectorstore" in st.session_state and st.session_state.vectorstore is not None:
|
263 |
st.subheader("Chat with the Document")
|
264 |
|
|
|
278 |
context = "\n".join([doc.page_content for doc in docs])
|
279 |
|
280 |
messages = [
|
281 |
+
HumanMessage(content=f"You are a helpful assistant. Answer the question based on the provided document context.\n\nContext: {context}\n\nQuestion: {prompt}")
|
|
|
282 |
]
|
283 |
|
284 |
response = llm.invoke(messages)
|
|
|
286 |
|
287 |
st.session_state.messages.append({"role": "assistant", "content": response.content})
|
288 |
|
|
|
289 |
st.markdown(
|
290 |
f'<div class="footer">Analysis Time: {st.session_state.analysis_time:.1f}s | Powered by Google Generative AI</div>',
|
291 |
unsafe_allow_html=True
|