Spaces:
Sleeping
Sleeping
CamiloVega
commited on
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,349 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import logging
|
3 |
+
from typing import List, Dict
|
4 |
+
import torch
|
5 |
+
import gradio as gr
|
6 |
+
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
7 |
+
from langchain.embeddings import HuggingFaceEmbeddings
|
8 |
+
from langchain.vectorstores import FAISS
|
9 |
+
from langchain.chains import RetrievalQA
|
10 |
+
from langchain.prompts import PromptTemplate
|
11 |
+
from langchain.llms import HuggingFacePipeline
|
12 |
+
from langchain_community.document_loaders import (
|
13 |
+
PyPDFLoader,
|
14 |
+
Docx2txtLoader,
|
15 |
+
CSVLoader,
|
16 |
+
UnstructuredFileLoader
|
17 |
+
)
|
18 |
+
from transformers import pipeline, AutoModelForCausalLM, AutoTokenizer
|
19 |
+
import spaces
|
20 |
+
import tempfile
|
21 |
+
|
22 |
+
# Configure logging
|
23 |
+
logging.basicConfig(
|
24 |
+
level=logging.INFO,
|
25 |
+
format='%(asctime)s - %(levelname)s - %(message)s'
|
26 |
+
)
|
27 |
+
logger = logging.getLogger(__name__)
|
28 |
+
|
29 |
+
# Constants
|
30 |
+
MODEL_NAME = "meta-llama/Llama-2-7b-chat-hf"
|
31 |
+
SUPPORTED_FORMATS = [".pdf", ".docx", ".doc", ".csv", ".txt"]
|
32 |
+
|
33 |
+
class DocumentLoader:
|
34 |
+
"""Enhanced document loader supporting multiple file formats."""
|
35 |
+
|
36 |
+
@staticmethod
|
37 |
+
def load_file(file_path: str) -> List:
|
38 |
+
"""Load a single file based on its extension."""
|
39 |
+
ext = os.path.splitext(file_path)[1].lower()
|
40 |
+
try:
|
41 |
+
if ext == '.pdf':
|
42 |
+
loader = PyPDFLoader(file_path)
|
43 |
+
elif ext in ['.docx', '.doc']:
|
44 |
+
loader = Docx2txtLoader(file_path)
|
45 |
+
elif ext == '.csv':
|
46 |
+
loader = CSVLoader(file_path)
|
47 |
+
else: # fallback for txt and other text files
|
48 |
+
loader = UnstructuredFileLoader(file_path)
|
49 |
+
|
50 |
+
documents = loader.load()
|
51 |
+
|
52 |
+
# Add metadata
|
53 |
+
for doc in documents:
|
54 |
+
doc.metadata.update({
|
55 |
+
'title': os.path.basename(file_path),
|
56 |
+
'type': 'document',
|
57 |
+
'format': ext[1:],
|
58 |
+
'language': 'auto'
|
59 |
+
})
|
60 |
+
|
61 |
+
logger.info(f"Successfully loaded {file_path}")
|
62 |
+
return documents
|
63 |
+
|
64 |
+
except Exception as e:
|
65 |
+
logger.error(f"Error loading {file_path}: {str(e)}")
|
66 |
+
raise
|
67 |
+
|
68 |
+
class RAGSystem:
|
69 |
+
"""Enhanced RAG system with dynamic document loading."""
|
70 |
+
|
71 |
+
def __init__(self, model_name: str = MODEL_NAME):
|
72 |
+
self.model_name = model_name
|
73 |
+
self.embeddings = None
|
74 |
+
self.vector_store = None
|
75 |
+
self.qa_chain = None
|
76 |
+
self.tokenizer = None
|
77 |
+
self.model = None
|
78 |
+
self.is_initialized = False
|
79 |
+
|
80 |
+
def initialize_model(self):
|
81 |
+
"""Initialize the base model and tokenizer."""
|
82 |
+
try:
|
83 |
+
logger.info("Initializing language model...")
|
84 |
+
|
85 |
+
# Initialize embeddings
|
86 |
+
self.embeddings = HuggingFaceEmbeddings(
|
87 |
+
model_name="intfloat/multilingual-e5-large",
|
88 |
+
model_kwargs={'device': 'cuda' if torch.cuda.is_available() else 'cpu'},
|
89 |
+
encode_kwargs={'normalize_embeddings': True}
|
90 |
+
)
|
91 |
+
|
92 |
+
# Initialize model and tokenizer
|
93 |
+
self.tokenizer = AutoTokenizer.from_pretrained(
|
94 |
+
self.model_name,
|
95 |
+
trust_remote_code=True
|
96 |
+
)
|
97 |
+
|
98 |
+
self.model = AutoModelForCausalLM.from_pretrained(
|
99 |
+
self.model_name,
|
100 |
+
torch_dtype=torch.float16,
|
101 |
+
trust_remote_code=True,
|
102 |
+
device_map="auto"
|
103 |
+
)
|
104 |
+
|
105 |
+
# Create generation pipeline
|
106 |
+
pipe = pipeline(
|
107 |
+
"text-generation",
|
108 |
+
model=self.model,
|
109 |
+
tokenizer=self.tokenizer,
|
110 |
+
max_new_tokens=512,
|
111 |
+
temperature=0.1,
|
112 |
+
top_p=0.95,
|
113 |
+
repetition_penalty=1.15,
|
114 |
+
device_map="auto"
|
115 |
+
)
|
116 |
+
|
117 |
+
self.llm = HuggingFacePipeline(pipeline=pipe)
|
118 |
+
self.is_initialized = True
|
119 |
+
|
120 |
+
logger.info("Model initialization completed")
|
121 |
+
|
122 |
+
except Exception as e:
|
123 |
+
logger.error(f"Error during model initialization: {str(e)}")
|
124 |
+
raise
|
125 |
+
|
126 |
+
def process_documents(self, files: List[tempfile._TemporaryFileWrapper]) -> None:
|
127 |
+
"""Process uploaded documents and update the vector store."""
|
128 |
+
try:
|
129 |
+
documents = []
|
130 |
+
for file in files:
|
131 |
+
docs = DocumentLoader.load_file(file.name)
|
132 |
+
documents.extend(docs)
|
133 |
+
|
134 |
+
if not documents:
|
135 |
+
raise ValueError("No documents were successfully loaded.")
|
136 |
+
|
137 |
+
# Process documents
|
138 |
+
text_splitter = RecursiveCharacterTextSplitter(
|
139 |
+
chunk_size=800,
|
140 |
+
chunk_overlap=200,
|
141 |
+
separators=["\n\n", "\n", ". ", " ", ""],
|
142 |
+
length_function=len
|
143 |
+
)
|
144 |
+
|
145 |
+
chunks = text_splitter.split_documents(documents)
|
146 |
+
|
147 |
+
# Create or update vector store
|
148 |
+
if self.vector_store is None:
|
149 |
+
self.vector_store = FAISS.from_documents(chunks, self.embeddings)
|
150 |
+
else:
|
151 |
+
self.vector_store.add_documents(chunks)
|
152 |
+
|
153 |
+
# Initialize QA chain
|
154 |
+
prompt_template = """
|
155 |
+
Context: {context}
|
156 |
+
|
157 |
+
Based on the provided context, please answer the following question clearly and concisely.
|
158 |
+
If the information is not in the context, please say so explicitly.
|
159 |
+
|
160 |
+
Question: {question}
|
161 |
+
"""
|
162 |
+
|
163 |
+
PROMPT = PromptTemplate(
|
164 |
+
template=prompt_template,
|
165 |
+
input_variables=["context", "question"]
|
166 |
+
)
|
167 |
+
|
168 |
+
self.qa_chain = RetrievalQA.from_chain_type(
|
169 |
+
llm=self.llm,
|
170 |
+
chain_type="stuff",
|
171 |
+
retriever=self.vector_store.as_retriever(
|
172 |
+
search_kwargs={"k": 6}
|
173 |
+
),
|
174 |
+
return_source_documents=True,
|
175 |
+
chain_type_kwargs={"prompt": PROMPT}
|
176 |
+
)
|
177 |
+
|
178 |
+
logger.info(f"Successfully processed {len(documents)} documents")
|
179 |
+
|
180 |
+
except Exception as e:
|
181 |
+
logger.error(f"Error processing documents: {str(e)}")
|
182 |
+
raise
|
183 |
+
|
184 |
+
def generate_response(self, question: str) -> Dict:
|
185 |
+
"""Generate response for a given question."""
|
186 |
+
if not self.is_initialized or self.qa_chain is None:
|
187 |
+
return {
|
188 |
+
'answer': "Please upload some documents first before asking questions.",
|
189 |
+
'sources': []
|
190 |
+
}
|
191 |
+
|
192 |
+
try:
|
193 |
+
result = self.qa_chain({"query": question})
|
194 |
+
|
195 |
+
response = {
|
196 |
+
'answer': result['result'],
|
197 |
+
'sources': []
|
198 |
+
}
|
199 |
+
|
200 |
+
for doc in result['source_documents']:
|
201 |
+
source = {
|
202 |
+
'title': doc.metadata.get('title', 'Unknown'),
|
203 |
+
'content': doc.page_content[:200] + "..." if len(doc.page_content) > 200 else doc.page_content,
|
204 |
+
'metadata': doc.metadata
|
205 |
+
}
|
206 |
+
response['sources'].append(source)
|
207 |
+
|
208 |
+
return response
|
209 |
+
|
210 |
+
except Exception as e:
|
211 |
+
logger.error(f"Error generating response: {str(e)}")
|
212 |
+
raise
|
213 |
+
|
214 |
+
@spaces.GPU(duration=60)
|
215 |
+
def process_response(user_input: str, chat_history: List, files: List) -> tuple:
|
216 |
+
"""Process user input and generate response."""
|
217 |
+
try:
|
218 |
+
if not rag_system.is_initialized:
|
219 |
+
rag_system.initialize_model()
|
220 |
+
|
221 |
+
if files and (rag_system.vector_store is None):
|
222 |
+
rag_system.process_documents(files)
|
223 |
+
|
224 |
+
response = rag_system.generate_response(user_input)
|
225 |
+
|
226 |
+
# Clean and format response
|
227 |
+
answer = response['answer']
|
228 |
+
if "Answer:" in answer:
|
229 |
+
answer = answer.split("Answer:")[-1].strip()
|
230 |
+
|
231 |
+
# Format sources
|
232 |
+
sources = set([source['title'] for source in response['sources'][:3]])
|
233 |
+
if sources:
|
234 |
+
answer += "\n\nπ Sources consulted:\n" + "\n".join([f"β’ {source}" for source in sources])
|
235 |
+
|
236 |
+
chat_history.append((user_input, answer))
|
237 |
+
return chat_history
|
238 |
+
|
239 |
+
except Exception as e:
|
240 |
+
logger.error(f"Error in process_response: {str(e)}")
|
241 |
+
error_message = f"Sorry, an error occurred: {str(e)}"
|
242 |
+
chat_history.append((user_input, error_message))
|
243 |
+
return chat_history
|
244 |
+
|
245 |
+
# Initialize RAG system
|
246 |
+
logger.info("Initializing RAG system...")
|
247 |
+
try:
|
248 |
+
rag_system = RAGSystem()
|
249 |
+
logger.info("RAG system created successfully")
|
250 |
+
except Exception as e:
|
251 |
+
logger.error(f"Failed to create RAG system: {str(e)}")
|
252 |
+
raise
|
253 |
+
|
254 |
+
# Create Gradio interface
|
255 |
+
try:
|
256 |
+
logger.info("Creating Gradio interface...")
|
257 |
+
with gr.Blocks(css="div.gradio-container {background-color: #f0f2f6}") as demo:
|
258 |
+
gr.HTML("""
|
259 |
+
<div style="text-align: center; max-width: 800px; margin: 0 auto; padding: 20px;">
|
260 |
+
<h1 style="color: #2d333a;">π DocumentGPT</h1>
|
261 |
+
<p style="color: #4a5568;">
|
262 |
+
Your AI Assistant for Document Analysis and Q&A
|
263 |
+
</p>
|
264 |
+
</div>
|
265 |
+
""")
|
266 |
+
|
267 |
+
with gr.Row():
|
268 |
+
with gr.Column(scale=1):
|
269 |
+
files = gr.Files(
|
270 |
+
label="Upload Your Documents",
|
271 |
+
file_types=SUPPORTED_FORMATS,
|
272 |
+
file_count="multiple"
|
273 |
+
)
|
274 |
+
gr.HTML("""
|
275 |
+
<div style="font-size: 0.9em; color: #666; margin-top: 0.5em;">
|
276 |
+
Supported formats: PDF, DOCX, CSV, TXT
|
277 |
+
</div>
|
278 |
+
""")
|
279 |
+
|
280 |
+
chatbot = gr.Chatbot(
|
281 |
+
show_label=False,
|
282 |
+
container=True,
|
283 |
+
height=500,
|
284 |
+
bubble_full_width=True,
|
285 |
+
show_copy_button=True,
|
286 |
+
scale=2
|
287 |
+
)
|
288 |
+
|
289 |
+
with gr.Row():
|
290 |
+
message = gr.Textbox(
|
291 |
+
placeholder="π Ask me anything about your documents...",
|
292 |
+
show_label=False,
|
293 |
+
container=False,
|
294 |
+
scale=8,
|
295 |
+
autofocus=True
|
296 |
+
)
|
297 |
+
clear = gr.Button("ποΈ Clear", size="sm", scale=1)
|
298 |
+
|
299 |
+
# Instructions
|
300 |
+
gr.HTML("""
|
301 |
+
<div style="background-color: #f8f9fa; padding: 15px; border-radius: 10px; margin: 20px 0;">
|
302 |
+
<h3 style="color: #2d333a; margin-bottom: 10px;">π How to use:</h3>
|
303 |
+
<ol style="color: #666; margin-left: 20px;">
|
304 |
+
<li>Upload one or more documents (PDF, DOCX, CSV, or TXT)</li>
|
305 |
+
<li>Wait for the documents to be processed</li>
|
306 |
+
<li>Ask questions about your documents</li>
|
307 |
+
<li>View sources used in the responses</li>
|
308 |
+
</ol>
|
309 |
+
</div>
|
310 |
+
""")
|
311 |
+
|
312 |
+
# Footer with credits
|
313 |
+
gr.HTML("""
|
314 |
+
<div style="text-align: center; max-width: 800px; margin: 20px auto; padding: 20px;
|
315 |
+
background-color: #f8f9fa; border-radius: 10px;">
|
316 |
+
<div style="margin-bottom: 15px;">
|
317 |
+
<h3 style="color: #2d333a;">β‘ About this assistant</h3>
|
318 |
+
<p style="color: #666; font-size: 14px;">
|
319 |
+
This application uses RAG (Retrieval Augmented Generation) technology combining:
|
320 |
+
</p>
|
321 |
+
<ul style="list-style: none; color: #666; font-size: 14px;">
|
322 |
+
<li>πΉ LLM Engine: Llama-2-7b-chat-hf</li>
|
323 |
+
<li>πΉ Embeddings: multilingual-e5-large</li>
|
324 |
+
<li>πΉ Vector Store: FAISS</li>
|
325 |
+
</ul>
|
326 |
+
</div>
|
327 |
+
<div style="border-top: 1px solid #ddd; padding-top: 15px;">
|
328 |
+
<p style="color: #666; font-size: 14px;">
|
329 |
+
Created by <a href="https://www.linkedin.com/in/camilo-vega-169084b1/"
|
330 |
+
target="_blank" style="color: #2196F3; text-decoration: none;">Camilo Vega</a>,
|
331 |
+
AI Professor and Solutions Consultant π€
|
332 |
+
</p>
|
333 |
+
</div>
|
334 |
+
</div>
|
335 |
+
""")
|
336 |
+
|
337 |
+
# Configure event handlers
|
338 |
+
def submit(user_input, chat_history, files):
|
339 |
+
return process_response(user_input, chat_history, files)
|
340 |
+
|
341 |
+
message.submit(submit, [message, chatbot, files], [chatbot])
|
342 |
+
clear.click(lambda: None, None, chatbot)
|
343 |
+
|
344 |
+
logger.info("Gradio interface created successfully")
|
345 |
+
demo.launch()
|
346 |
+
|
347 |
+
except Exception as e:
|
348 |
+
logger.error(f"Error in Gradio interface creation: {str(e)}")
|
349 |
+
raise
|