Spaces:
Running
Running
Abid Ali Awan
commited on
Commit
·
9b3bd46
1
Parent(s):
355b607
fix the issues with the app and optimized it.
Browse files
main.py
CHANGED
@@ -1,81 +1,47 @@
|
|
|
|
1 |
import os
|
|
|
2 |
import zipfile
|
3 |
-
from typing import
|
|
|
4 |
|
|
|
|
|
5 |
import gradio as gr
|
6 |
from groq import Groq
|
7 |
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
|
|
8 |
from langchain_core.output_parsers import StrOutputParser
|
9 |
from langchain_core.prompts import PromptTemplate
|
10 |
from langchain_core.runnables import RunnablePassthrough
|
|
|
11 |
from langchain_groq import ChatGroq
|
12 |
from langchain_huggingface import HuggingFaceEmbeddings
|
13 |
-
from langchain_core.vectorstores import InMemoryVectorStore
|
14 |
-
# Retrieve API key for Groq from the environment variables
|
15 |
-
GROQ_API_KEY = os.environ.get("GROQ_API_KEY")
|
16 |
-
|
17 |
-
# Initialize the Groq client
|
18 |
-
client = Groq(api_key=GROQ_API_KEY)
|
19 |
-
|
20 |
-
# Initialize the LLM
|
21 |
-
llm = ChatGroq(model="meta-llama/llama-4-scout-17b-16e-instruct", api_key=GROQ_API_KEY)
|
22 |
|
23 |
-
#
|
24 |
-
|
25 |
-
|
26 |
-
# General constants for the UI
|
27 |
-
TITLE = """<h1 align="center">✨ Llama 4 RAG Application</h1>"""
|
28 |
AVATAR_IMAGES = (
|
29 |
None,
|
30 |
"https://huggingface.co/datasets/huggingface/brand-assets/resolve/main/hf-logo.png",
|
31 |
)
|
32 |
|
33 |
-
#
|
34 |
-
TEXT_EXTENSIONS = [
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
".go",
|
44 |
-
".h",
|
45 |
-
".html",
|
46 |
-
".ini",
|
47 |
-
".java",
|
48 |
-
".js",
|
49 |
-
".json",
|
50 |
-
".jsx",
|
51 |
-
".md",
|
52 |
-
".php",
|
53 |
-
".ps1",
|
54 |
-
".py",
|
55 |
-
".rb",
|
56 |
-
".rs",
|
57 |
-
".sh",
|
58 |
-
".toml",
|
59 |
-
".ts",
|
60 |
-
".tsx",
|
61 |
-
".txt",
|
62 |
-
".xml",
|
63 |
-
".yaml",
|
64 |
-
".yml",
|
65 |
-
]
|
66 |
-
|
67 |
-
# Global variables
|
68 |
-
EXTRACTED_FILES = {}
|
69 |
-
VECTORSTORE = None
|
70 |
-
RAG_CHAIN = None
|
71 |
-
|
72 |
-
# Initialize the text splitter
|
73 |
text_splitter = RecursiveCharacterTextSplitter(
|
74 |
-
chunk_size=1000,
|
|
|
|
|
75 |
)
|
76 |
|
77 |
-
|
78 |
-
template = """You are an expert assistant tasked with answering questions based on the provided documents.
|
79 |
Use only the given context to generate your answer.
|
80 |
If the answer cannot be found in the context, clearly state that you do not know.
|
81 |
Be detailed and precise in your response, but avoid mentioning or referencing the context itself.
|
@@ -87,424 +53,285 @@ Question:
|
|
87 |
{question}
|
88 |
|
89 |
Answer:"""
|
|
|
90 |
|
91 |
-
# Create the PromptTemplate
|
92 |
-
rag_prompt = PromptTemplate.from_template(template)
|
93 |
|
|
|
|
|
|
|
|
|
94 |
|
95 |
-
def extract_text_from_zip(zip_file_path: str) -> Dict[str, str]:
|
96 |
-
"""
|
97 |
-
Extract text content from files in a ZIP archive.
|
98 |
|
99 |
-
|
100 |
-
zip_file_path (str): Path to the ZIP file.
|
101 |
|
102 |
-
|
103 |
-
Dict[str, str]: Dictionary mapping filenames to their text content.
|
104 |
-
"""
|
105 |
-
text_contents = {}
|
106 |
|
107 |
-
with zipfile.ZipFile(zip_file_path, "r") as zip_ref:
|
108 |
-
for file_info in zip_ref.infolist():
|
109 |
-
# Skip directories
|
110 |
-
if file_info.filename.endswith("/"):
|
111 |
-
continue
|
112 |
|
113 |
-
|
114 |
-
|
|
|
115 |
|
116 |
-
|
117 |
-
|
118 |
-
|
119 |
-
|
120 |
-
|
121 |
-
|
122 |
-
|
123 |
-
|
124 |
-
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
125 |
|
126 |
-
|
|
|
|
|
|
|
|
|
127 |
|
|
|
128 |
|
129 |
-
def extract_text_from_single_file(file_path: str) -> Dict[str, str]:
|
130 |
-
"""
|
131 |
-
Extract text content from a single file.
|
132 |
|
133 |
-
|
134 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
135 |
|
136 |
-
|
137 |
-
Dict[str, str]: Dictionary mapping filename to its text content.
|
138 |
-
"""
|
139 |
-
text_contents = {}
|
140 |
-
filename = os.path.basename(file_path)
|
141 |
-
file_ext = os.path.splitext(filename)[1].lower()
|
142 |
|
143 |
-
if file_ext in TEXT_EXTENSIONS:
|
144 |
-
try:
|
145 |
-
with open(file_path, "r", encoding="utf-8", errors="replace") as file:
|
146 |
-
content = file.read()
|
147 |
-
text_contents[filename] = content
|
148 |
-
except Exception as e:
|
149 |
-
text_contents[filename] = f"Error reading file: {str(e)}"
|
150 |
|
151 |
-
return text_contents
|
152 |
|
153 |
|
154 |
def upload_files(
|
155 |
-
files: Optional[List[str]], chatbot: List[Union[
|
156 |
):
|
157 |
-
"""
|
158 |
-
|
159 |
-
|
160 |
-
Parameters:
|
161 |
-
files (Optional[List[str]]): List of file paths.
|
162 |
-
chatbot (List[Union[dict, gr.ChatMessage]]): The conversation history.
|
163 |
-
|
164 |
-
Returns:
|
165 |
-
List[Union[dict, gr.ChatMessage]]: Updated conversation history.
|
166 |
-
"""
|
167 |
-
global EXTRACTED_FILES, VECTORSTORE, RAG_CHAIN
|
168 |
-
|
169 |
-
# Handle multiple file uploads
|
170 |
-
if len(files) > 1:
|
171 |
-
total_files_processed = 0
|
172 |
-
total_files_extracted = 0
|
173 |
-
file_types = set()
|
174 |
-
|
175 |
-
# Process each file
|
176 |
-
for file in files:
|
177 |
-
filename = os.path.basename(file)
|
178 |
-
file_ext = os.path.splitext(filename)[1].lower()
|
179 |
-
|
180 |
-
# Process based on file type
|
181 |
-
if file_ext == ".zip":
|
182 |
-
extracted_files = extract_text_from_zip(file)
|
183 |
-
file_types.add("zip")
|
184 |
-
else:
|
185 |
-
extracted_files = extract_text_from_single_file(file)
|
186 |
-
file_types.add("text")
|
187 |
|
188 |
-
|
189 |
-
|
190 |
-
# Store the extracted content in the global variable
|
191 |
-
EXTRACTED_FILES[filename] = extracted_files
|
192 |
|
193 |
-
|
|
|
|
|
|
|
194 |
|
195 |
-
|
196 |
-
|
197 |
-
|
198 |
-
|
199 |
-
|
200 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
201 |
|
202 |
-
|
203 |
-
|
204 |
|
|
|
205 |
chatbot.append(
|
206 |
gr.ChatMessage(
|
207 |
-
role="
|
208 |
-
content=f"<p>📚 Multiple {file_types_str} uploaded ({total_files_processed} files)</p><p>Extracted {total_files_extracted} text file(s) in total</p><p>Uploaded files:</p><pre>{file_list}</pre>",
|
209 |
-
)
|
210 |
-
)
|
211 |
-
|
212 |
-
# Handle single file upload
|
213 |
-
elif len(files) == 1:
|
214 |
-
file = files[0]
|
215 |
-
filename = os.path.basename(file)
|
216 |
-
file_ext = os.path.splitext(filename)[1].lower()
|
217 |
-
|
218 |
-
# Process based on file type
|
219 |
-
if file_ext == ".zip":
|
220 |
-
extracted_files = extract_text_from_zip(file)
|
221 |
-
file_type_msg = "📦 ZIP file"
|
222 |
-
else:
|
223 |
-
extracted_files = extract_text_from_single_file(file)
|
224 |
-
file_type_msg = "📄 File"
|
225 |
-
|
226 |
-
if not extracted_files:
|
227 |
-
chatbot.append(
|
228 |
-
gr.ChatMessage(
|
229 |
-
role="user",
|
230 |
-
content=f"<p>{file_type_msg} uploaded: {filename}, but no text content was found or the file format is not supported.</p>",
|
231 |
-
)
|
232 |
-
)
|
233 |
-
else:
|
234 |
-
file_list = "\n".join([f"- {name}" for name in extracted_files.keys()])
|
235 |
-
chatbot.append(
|
236 |
-
gr.ChatMessage(
|
237 |
-
role="user",
|
238 |
-
content=f"<p>{file_type_msg} uploaded: {filename}</p><p>Extracted {len(extracted_files)} text file(s):</p><pre>{file_list}</pre>",
|
239 |
-
)
|
240 |
)
|
241 |
-
|
242 |
-
# Store the extracted content in the global variable
|
243 |
-
EXTRACTED_FILES[filename] = extracted_files
|
244 |
-
|
245 |
-
# Process the extracted files and create vector embeddings
|
246 |
-
if EXTRACTED_FILES:
|
247 |
-
# Prepare documents for processing
|
248 |
-
all_texts = []
|
249 |
-
for filename, files in EXTRACTED_FILES.items():
|
250 |
-
for file_path, content in files.items():
|
251 |
-
all_texts.append(
|
252 |
-
{"page_content": content, "metadata": {"source": file_path}}
|
253 |
-
)
|
254 |
-
|
255 |
-
# Create document objects
|
256 |
-
from langchain_core.documents import Document
|
257 |
-
|
258 |
-
documents = [
|
259 |
-
Document(page_content=item["page_content"], metadata=item["metadata"])
|
260 |
-
for item in all_texts
|
261 |
-
]
|
262 |
-
|
263 |
-
# Split the documents into chunks
|
264 |
-
chunks = text_splitter.split_documents(documents)
|
265 |
-
|
266 |
-
# Create the vector store
|
267 |
-
VECTORSTORE = InMemoryVectorStore.from_documents(
|
268 |
-
documents=chunks,
|
269 |
-
embedding=embed_model,
|
270 |
-
)
|
271 |
-
|
272 |
-
# Create the retriever
|
273 |
-
retriever = VECTORSTORE.as_retriever()
|
274 |
-
|
275 |
-
# Create the RAG chain
|
276 |
-
RAG_CHAIN = (
|
277 |
-
{"context": retriever, "question": RunnablePassthrough()}
|
278 |
-
| rag_prompt
|
279 |
-
| llm
|
280 |
-
| StrOutputParser()
|
281 |
)
|
|
|
282 |
|
283 |
-
|
|
|
|
|
284 |
chatbot.append(
|
285 |
gr.ChatMessage(
|
286 |
-
role="assistant",
|
287 |
-
content="Documents processed and indexed. You can now ask questions about the content.",
|
288 |
)
|
289 |
)
|
|
|
290 |
|
291 |
-
|
292 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
293 |
|
294 |
-
|
295 |
-
|
296 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
297 |
|
298 |
-
Parameters:
|
299 |
-
text_prompt (str): The input text provided by the user.
|
300 |
-
chatbot (List[gr.ChatMessage]): The existing conversation history.
|
301 |
|
302 |
-
|
303 |
-
|
304 |
-
|
305 |
-
"""
|
306 |
-
if text_prompt:
|
307 |
chatbot.append(gr.ChatMessage(role="user", content=text_prompt))
|
308 |
return "", chatbot
|
309 |
|
310 |
|
311 |
-
def
|
312 |
-
|
313 |
-
|
314 |
-
|
315 |
-
|
316 |
-
|
317 |
-
|
318 |
-
Returns:
|
319 |
-
str: The textual content of the message.
|
320 |
-
"""
|
321 |
-
if isinstance(msg, dict):
|
322 |
-
return msg.get("content", "")
|
323 |
-
return msg.content
|
324 |
-
|
325 |
-
|
326 |
-
def process_query(chatbot: List[Union[dict, gr.ChatMessage]]):
|
327 |
-
"""
|
328 |
-
Process the user's query using the RAG pipeline.
|
329 |
-
|
330 |
-
Parameters:
|
331 |
-
chatbot (List[Union[dict, gr.ChatMessage]]): The conversation history.
|
332 |
-
|
333 |
-
Returns:
|
334 |
-
List[Union[dict, gr.ChatMessage]]: The updated conversation history with the response.
|
335 |
-
"""
|
336 |
-
global RAG_CHAIN
|
337 |
-
|
338 |
-
if len(chatbot) == 0:
|
339 |
-
chatbot.append(
|
340 |
-
gr.ChatMessage(
|
341 |
-
role="assistant",
|
342 |
-
content="Please enter a question or upload documents to start the conversation.",
|
343 |
-
)
|
344 |
-
)
|
345 |
-
return chatbot
|
346 |
-
|
347 |
-
# Get the last user message as the prompt
|
348 |
-
user_messages = [
|
349 |
-
msg
|
350 |
-
for msg in chatbot
|
351 |
-
if (isinstance(msg, dict) and msg.get("role") == "user")
|
352 |
-
or (hasattr(msg, "role") and msg.role == "user")
|
353 |
-
]
|
354 |
-
|
355 |
-
if not user_messages:
|
356 |
chatbot.append(
|
357 |
-
gr.ChatMessage(
|
358 |
-
role="assistant",
|
359 |
-
content="Please enter a question to start the conversation.",
|
360 |
-
)
|
361 |
)
|
362 |
return chatbot
|
363 |
|
364 |
-
|
365 |
-
prompt = get_message_content(last_user_msg)
|
366 |
-
|
367 |
-
# Skip if the last message was about uploading a file
|
368 |
-
if (
|
369 |
-
"📦 ZIP file uploaded:" in prompt
|
370 |
-
or "📄 File uploaded:" in prompt
|
371 |
-
or "📚 Multiple files uploaded" in prompt
|
372 |
-
):
|
373 |
-
return chatbot
|
374 |
-
|
375 |
-
# Check if RAG chain is available
|
376 |
-
if RAG_CHAIN is None:
|
377 |
chatbot.append(
|
378 |
-
gr.ChatMessage(
|
379 |
-
role="assistant",
|
380 |
-
content="Please upload documents first to enable question answering.",
|
381 |
-
)
|
382 |
)
|
383 |
return chatbot
|
384 |
|
385 |
-
# Append a placeholder for the assistant's response
|
386 |
chatbot.append(gr.ChatMessage(role="assistant", content="Thinking..."))
|
387 |
|
388 |
try:
|
389 |
-
|
390 |
-
response = RAG_CHAIN.invoke(prompt)
|
391 |
-
|
392 |
-
# Update the placeholder with the actual response
|
393 |
chatbot[-1].content = response
|
394 |
except Exception as e:
|
395 |
-
|
396 |
-
chatbot[-1].content = f"Error processing your query: {str(e)}"
|
397 |
|
398 |
return chatbot
|
399 |
|
400 |
|
401 |
-
def reset_app(
|
402 |
-
|
403 |
-
|
404 |
-
|
405 |
-
|
406 |
-
|
407 |
-
|
408 |
-
Returns:
|
409 |
-
List[Union[dict, gr.ChatMessage]]: A fresh conversation history.
|
410 |
-
"""
|
411 |
-
global EXTRACTED_FILES, VECTORSTORE, RAG_CHAIN
|
412 |
-
|
413 |
-
# Clear the global variables
|
414 |
-
EXTRACTED_FILES = {}
|
415 |
-
VECTORSTORE = None
|
416 |
-
RAG_CHAIN = None
|
417 |
-
|
418 |
-
# Reset the chatbot with a welcome message
|
419 |
return [
|
420 |
gr.ChatMessage(
|
421 |
-
role="assistant",
|
422 |
-
content="App has been reset. You can start a new conversation or upload new documents.",
|
423 |
)
|
424 |
]
|
425 |
|
426 |
|
427 |
-
#
|
428 |
-
chatbot_component = gr.Chatbot(
|
429 |
-
label="Llama 4 RAG",
|
430 |
-
type="messages",
|
431 |
-
bubble_full_width=False,
|
432 |
-
avatar_images=AVATAR_IMAGES,
|
433 |
-
scale=2,
|
434 |
-
height=350,
|
435 |
-
)
|
436 |
-
text_prompt_component = gr.Textbox(
|
437 |
-
placeholder="Ask a question about your documents...",
|
438 |
-
show_label=False,
|
439 |
-
autofocus=True,
|
440 |
-
scale=28,
|
441 |
-
)
|
442 |
-
upload_files_button_component = gr.UploadButton(
|
443 |
-
label="Upload",
|
444 |
-
file_count="multiple",
|
445 |
-
file_types=[".zip", ".docx"] + TEXT_EXTENSIONS,
|
446 |
-
scale=1,
|
447 |
-
min_width=80,
|
448 |
-
)
|
449 |
-
send_button_component = gr.Button(
|
450 |
-
value="Send", variant="primary", scale=1, min_width=80
|
451 |
-
)
|
452 |
-
reset_button_component = gr.Button(value="Reset", variant="stop", scale=1, min_width=80)
|
453 |
-
|
454 |
-
# Define input lists for button chaining
|
455 |
-
user_inputs = [text_prompt_component, chatbot_component]
|
456 |
|
457 |
-
with gr.Blocks(theme=gr.themes.
|
458 |
gr.HTML(TITLE)
|
459 |
-
|
460 |
-
|
461 |
-
|
462 |
-
|
463 |
-
|
464 |
-
|
465 |
-
|
466 |
-
|
467 |
-
# When the Send button is clicked, first process the user text then process the query
|
468 |
-
send_button_component.click(
|
469 |
-
fn=user,
|
470 |
-
inputs=user_inputs,
|
471 |
-
outputs=[text_prompt_component, chatbot_component],
|
472 |
-
queue=False,
|
473 |
-
).then(
|
474 |
-
fn=process_query,
|
475 |
-
inputs=[chatbot_component],
|
476 |
-
outputs=[chatbot_component],
|
477 |
-
api_name="process_query",
|
478 |
)
|
479 |
|
480 |
-
|
481 |
-
|
482 |
-
|
483 |
-
|
484 |
-
|
485 |
-
|
486 |
-
|
487 |
-
|
488 |
-
|
489 |
-
|
490 |
-
|
491 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
492 |
|
493 |
-
|
494 |
-
|
495 |
-
|
496 |
-
|
497 |
-
outputs=[chatbot_component],
|
498 |
queue=False,
|
499 |
-
)
|
500 |
|
501 |
-
|
502 |
-
|
503 |
-
|
504 |
-
|
505 |
-
outputs=[chatbot_component],
|
506 |
queue=False,
|
|
|
|
|
|
|
|
|
507 |
)
|
|
|
508 |
|
509 |
-
# Launch the demo interface
|
510 |
demo.queue().launch()
|
|
|
1 |
+
# ========== Standard Library ==========
|
2 |
import os
|
3 |
+
import tempfile
|
4 |
import zipfile
|
5 |
+
from typing import List, Optional, Tuple, Union
|
6 |
+
import collections
|
7 |
|
8 |
+
|
9 |
+
# ========== Third-Party Libraries ==========
|
10 |
import gradio as gr
|
11 |
from groq import Groq
|
12 |
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
13 |
+
from langchain_community.document_loaders import DirectoryLoader, UnstructuredFileLoader
|
14 |
from langchain_core.output_parsers import StrOutputParser
|
15 |
from langchain_core.prompts import PromptTemplate
|
16 |
from langchain_core.runnables import RunnablePassthrough
|
17 |
+
from langchain_core.vectorstores import InMemoryVectorStore
|
18 |
from langchain_groq import ChatGroq
|
19 |
from langchain_huggingface import HuggingFaceEmbeddings
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
20 |
|
21 |
+
# ========== Configs ==========
|
22 |
+
TITLE = """<h1 align="center">🗨️🦙 Llama 4 Docx Chatter</h1>"""
|
|
|
|
|
|
|
23 |
AVATAR_IMAGES = (
|
24 |
None,
|
25 |
"https://huggingface.co/datasets/huggingface/brand-assets/resolve/main/hf-logo.png",
|
26 |
)
|
27 |
|
28 |
+
# Acceptable file extensions
|
29 |
+
TEXT_EXTENSIONS = [".docx", ".zip"]
|
30 |
+
|
31 |
+
# ========== Models & Clients ==========
|
32 |
+
GROQ_API_KEY = os.getenv("GROQ_API_KEY")
|
33 |
+
client = Groq(api_key=GROQ_API_KEY)
|
34 |
+
llm = ChatGroq(model="meta-llama/llama-4-scout-17b-16e-instruct", api_key=GROQ_API_KEY)
|
35 |
+
embed_model = HuggingFaceEmbeddings(model_name="mixedbread-ai/mxbai-embed-large-v1")
|
36 |
+
|
37 |
+
# ========== Core Components ==========
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
38 |
text_splitter = RecursiveCharacterTextSplitter(
|
39 |
+
chunk_size=1000,
|
40 |
+
chunk_overlap=100,
|
41 |
+
separators=["\n\n", "\n"],
|
42 |
)
|
43 |
|
44 |
+
rag_template = """You are an expert assistant tasked with answering questions based on the provided documents.
|
|
|
45 |
Use only the given context to generate your answer.
|
46 |
If the answer cannot be found in the context, clearly state that you do not know.
|
47 |
Be detailed and precise in your response, but avoid mentioning or referencing the context itself.
|
|
|
53 |
{question}
|
54 |
|
55 |
Answer:"""
|
56 |
+
rag_prompt = PromptTemplate.from_template(rag_template)
|
57 |
|
|
|
|
|
58 |
|
59 |
+
# ========== App State ==========
|
60 |
+
class AppState:
|
61 |
+
vectorstore: Optional[InMemoryVectorStore] = None
|
62 |
+
rag_chain = None
|
63 |
|
|
|
|
|
|
|
64 |
|
65 |
+
state = AppState()
|
|
|
66 |
|
67 |
+
# ========== Utility Functions ==========
|
|
|
|
|
|
|
68 |
|
|
|
|
|
|
|
|
|
|
|
69 |
|
70 |
+
def load_documents_from_files(files: List[str]) -> List:
|
71 |
+
"""Load documents from uploaded files directly without moving."""
|
72 |
+
all_documents = []
|
73 |
|
74 |
+
# Temporary directory if ZIP needs extraction
|
75 |
+
with tempfile.TemporaryDirectory() as temp_dir:
|
76 |
+
for file_path in files:
|
77 |
+
ext = os.path.splitext(file_path)[1].lower()
|
78 |
+
|
79 |
+
if ext == ".zip":
|
80 |
+
# Extract ZIP inside temp_dir
|
81 |
+
with zipfile.ZipFile(file_path, "r") as zip_ref:
|
82 |
+
zip_ref.extractall(temp_dir)
|
83 |
+
|
84 |
+
# Load all docx from extracted zip
|
85 |
+
loader = DirectoryLoader(
|
86 |
+
path=temp_dir,
|
87 |
+
glob="**/*.docx",
|
88 |
+
use_multithreading=True,
|
89 |
+
)
|
90 |
+
docs = loader.load()
|
91 |
+
all_documents.extend(docs)
|
92 |
|
93 |
+
elif ext == ".docx":
|
94 |
+
# Load single docx directly
|
95 |
+
loader = UnstructuredFileLoader(file_path)
|
96 |
+
docs = loader.load()
|
97 |
+
all_documents.extend(docs)
|
98 |
|
99 |
+
return all_documents
|
100 |
|
|
|
|
|
|
|
101 |
|
102 |
+
def get_last_user_message(chatbot: List[Union[gr.ChatMessage, dict]]) -> Optional[str]:
|
103 |
+
"""Get last user prompt."""
|
104 |
+
for message in reversed(chatbot):
|
105 |
+
content = (
|
106 |
+
message.get("content") if isinstance(message, dict) else message.content
|
107 |
+
)
|
108 |
+
if (
|
109 |
+
message.get("role") if isinstance(message, dict) else message.role
|
110 |
+
) == "user":
|
111 |
+
return content
|
112 |
+
return None
|
113 |
+
|
114 |
|
115 |
+
# ========== Main Logic ==========
|
|
|
|
|
|
|
|
|
|
|
116 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
117 |
|
|
|
118 |
|
119 |
|
120 |
def upload_files(
|
121 |
+
files: Optional[List[str]], chatbot: List[Union[gr.ChatMessage, dict]]
|
122 |
):
|
123 |
+
"""Handle file upload - .docx or .zip containing docx."""
|
124 |
+
if not files:
|
125 |
+
return chatbot
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
126 |
|
127 |
+
file_summaries = [] # <-- Collect formatted file/folder info
|
128 |
+
documents = []
|
|
|
|
|
129 |
|
130 |
+
with tempfile.TemporaryDirectory() as temp_dir:
|
131 |
+
for file_path in files:
|
132 |
+
filename = os.path.basename(file_path)
|
133 |
+
ext = os.path.splitext(file_path)[1].lower()
|
134 |
|
135 |
+
if ext == ".zip":
|
136 |
+
file_summaries.append(f"📦 **{filename}** (ZIP file) contains:")
|
137 |
+
try:
|
138 |
+
with zipfile.ZipFile(file_path, "r") as zip_ref:
|
139 |
+
zip_ref.extractall(temp_dir)
|
140 |
+
zip_contents = zip_ref.namelist()
|
141 |
+
|
142 |
+
# Group files by folder
|
143 |
+
folder_map = collections.defaultdict(list)
|
144 |
+
for item in zip_contents:
|
145 |
+
if item.endswith("/"):
|
146 |
+
continue # skip folder entries themselves
|
147 |
+
folder = os.path.dirname(item)
|
148 |
+
file_name = os.path.basename(item)
|
149 |
+
folder_map[folder].append(file_name)
|
150 |
+
|
151 |
+
# Format nicely
|
152 |
+
for folder, files_in_folder in folder_map.items():
|
153 |
+
if folder:
|
154 |
+
file_summaries.append(f"📂 {folder}/")
|
155 |
+
else:
|
156 |
+
file_summaries.append(f"📄 (root)")
|
157 |
+
for f in files_in_folder:
|
158 |
+
file_summaries.append(f" - {f}")
|
159 |
+
|
160 |
+
# Load docx files extracted from ZIP
|
161 |
+
loader = DirectoryLoader(
|
162 |
+
path=temp_dir,
|
163 |
+
glob="**/*.docx",
|
164 |
+
use_multithreading=True,
|
165 |
+
)
|
166 |
+
docs = loader.load()
|
167 |
+
documents.extend(docs)
|
168 |
+
|
169 |
+
except zipfile.BadZipFile:
|
170 |
+
chatbot.append(
|
171 |
+
gr.ChatMessage(
|
172 |
+
role="assistant",
|
173 |
+
content=f"❌ Failed to open ZIP file: {filename}",
|
174 |
+
)
|
175 |
+
)
|
176 |
+
|
177 |
+
elif ext == ".docx":
|
178 |
+
file_summaries.append(f"📄 **{filename}**")
|
179 |
+
loader = UnstructuredFileLoader(file_path)
|
180 |
+
docs = loader.load()
|
181 |
+
documents.extend(docs)
|
182 |
|
183 |
+
else:
|
184 |
+
file_summaries.append(f"❌ Unsupported file type: {filename}")
|
185 |
|
186 |
+
if not documents:
|
187 |
chatbot.append(
|
188 |
gr.ChatMessage(
|
189 |
+
role="assistant", content="No valid .docx files found in upload."
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
190 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
191 |
)
|
192 |
+
return chatbot
|
193 |
|
194 |
+
# Split documents
|
195 |
+
chunks = text_splitter.split_documents(documents)
|
196 |
+
if not chunks:
|
197 |
chatbot.append(
|
198 |
gr.ChatMessage(
|
199 |
+
role="assistant", content="Failed to split documents into chunks."
|
|
|
200 |
)
|
201 |
)
|
202 |
+
return chatbot
|
203 |
|
204 |
+
# Create Vectorstore
|
205 |
+
state.vectorstore = InMemoryVectorStore.from_documents(
|
206 |
+
documents=chunks,
|
207 |
+
embedding=embed_model,
|
208 |
+
)
|
209 |
+
retriever = state.vectorstore.as_retriever()
|
210 |
+
|
211 |
+
# Build RAG Chain
|
212 |
+
state.rag_chain = (
|
213 |
+
{"context": retriever, "question": RunnablePassthrough()}
|
214 |
+
| rag_prompt
|
215 |
+
| llm
|
216 |
+
| StrOutputParser()
|
217 |
+
)
|
218 |
|
219 |
+
# Final display
|
220 |
+
chatbot.append(
|
221 |
+
gr.ChatMessage(
|
222 |
+
role="assistant",
|
223 |
+
content="**Uploaded Files:**\n"
|
224 |
+
+ "\n".join(file_summaries)
|
225 |
+
+ "\n\n✅ Ready to chat!",
|
226 |
+
)
|
227 |
+
)
|
228 |
+
return chatbot
|
229 |
|
|
|
|
|
|
|
230 |
|
231 |
+
def user_message(
|
232 |
+
text_prompt: str, chatbot: List[Union[gr.ChatMessage, dict]]
|
233 |
+
) -> Tuple[str, List[Union[gr.ChatMessage, dict]]]:
|
234 |
+
"""Add user's text input to conversation."""
|
235 |
+
if text_prompt.strip():
|
236 |
chatbot.append(gr.ChatMessage(role="user", content=text_prompt))
|
237 |
return "", chatbot
|
238 |
|
239 |
|
240 |
+
def process_query(
|
241 |
+
chatbot: List[Union[gr.ChatMessage, dict]],
|
242 |
+
) -> List[Union[gr.ChatMessage, dict]]:
|
243 |
+
"""Process user's query through RAG pipeline."""
|
244 |
+
prompt = get_last_user_message(chatbot)
|
245 |
+
if not prompt:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
246 |
chatbot.append(
|
247 |
+
gr.ChatMessage(role="assistant", content="Please type a question first.")
|
|
|
|
|
|
|
248 |
)
|
249 |
return chatbot
|
250 |
|
251 |
+
if state.rag_chain is None:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
252 |
chatbot.append(
|
253 |
+
gr.ChatMessage(role="assistant", content="Please upload documents first.")
|
|
|
|
|
|
|
254 |
)
|
255 |
return chatbot
|
256 |
|
|
|
257 |
chatbot.append(gr.ChatMessage(role="assistant", content="Thinking..."))
|
258 |
|
259 |
try:
|
260 |
+
response = state.rag_chain.invoke(prompt)
|
|
|
|
|
|
|
261 |
chatbot[-1].content = response
|
262 |
except Exception as e:
|
263 |
+
chatbot[-1].content = f"Error: {str(e)}"
|
|
|
264 |
|
265 |
return chatbot
|
266 |
|
267 |
|
268 |
+
def reset_app(
|
269 |
+
chatbot: List[Union[gr.ChatMessage, dict]],
|
270 |
+
) -> List[Union[gr.ChatMessage, dict]]:
|
271 |
+
"""Reset application state."""
|
272 |
+
state.vectorstore = None
|
273 |
+
state.rag_chain = None
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
274 |
return [
|
275 |
gr.ChatMessage(
|
276 |
+
role="assistant", content="App reset! Upload new documents to start."
|
|
|
277 |
)
|
278 |
]
|
279 |
|
280 |
|
281 |
+
# ========== UI Layout ==========
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
282 |
|
283 |
+
with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
284 |
gr.HTML(TITLE)
|
285 |
+
chatbot = gr.Chatbot(
|
286 |
+
label="Llama 4 RAG",
|
287 |
+
type="messages",
|
288 |
+
bubble_full_width=False,
|
289 |
+
avatar_images=AVATAR_IMAGES,
|
290 |
+
scale=2,
|
291 |
+
height=350,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
292 |
)
|
293 |
|
294 |
+
with gr.Row(equal_height=True):
|
295 |
+
text_prompt = gr.Textbox(
|
296 |
+
placeholder="Ask a question...", show_label=False, autofocus=True, scale=28
|
297 |
+
)
|
298 |
+
send_button = gr.Button(
|
299 |
+
value="Send",
|
300 |
+
variant="primary",
|
301 |
+
scale=1,
|
302 |
+
min_width=80,
|
303 |
+
)
|
304 |
+
upload_button = gr.UploadButton(
|
305 |
+
label="Upload",
|
306 |
+
file_count="multiple",
|
307 |
+
file_types=TEXT_EXTENSIONS,
|
308 |
+
scale=1,
|
309 |
+
min_width=80,
|
310 |
+
)
|
311 |
+
reset_button = gr.Button(
|
312 |
+
value="Reset",
|
313 |
+
variant="stop",
|
314 |
+
scale=1,
|
315 |
+
min_width=80,
|
316 |
+
)
|
317 |
|
318 |
+
send_button.click(
|
319 |
+
fn=user_message,
|
320 |
+
inputs=[text_prompt, chatbot],
|
321 |
+
outputs=[text_prompt, chatbot],
|
|
|
322 |
queue=False,
|
323 |
+
).then(fn=process_query, inputs=[chatbot], outputs=[chatbot])
|
324 |
|
325 |
+
text_prompt.submit(
|
326 |
+
fn=user_message,
|
327 |
+
inputs=[text_prompt, chatbot],
|
328 |
+
outputs=[text_prompt, chatbot],
|
|
|
329 |
queue=False,
|
330 |
+
).then(fn=process_query, inputs=[chatbot], outputs=[chatbot])
|
331 |
+
|
332 |
+
upload_button.upload(
|
333 |
+
fn=upload_files, inputs=[upload_button, chatbot], outputs=[chatbot], queue=False
|
334 |
)
|
335 |
+
reset_button.click(fn=reset_app, inputs=[chatbot], outputs=[chatbot], queue=False)
|
336 |
|
|
|
337 |
demo.queue().launch()
|