Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,343 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
import gradio as gr
|
3 |
+
import spaces
|
4 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
|
5 |
+
import os, gc, logging
|
6 |
+
from threading import Thread
|
7 |
+
import random
|
8 |
+
from datasets import load_dataset
|
9 |
+
import numpy as np
|
10 |
+
from sklearn.feature_extraction.text import TfidfVectorizer
|
11 |
+
import pandas as pd
|
12 |
+
from typing import List, Tuple, Iterator
|
13 |
+
import json
|
14 |
+
from datetime import datetime
|
15 |
+
from concurrent.futures import ThreadPoolExecutor
|
16 |
+
from functools import lru_cache
|
17 |
+
import pyarrow.parquet as pq
|
18 |
+
import pypdf
|
19 |
+
from pdfminer.high_level import extract_text
|
20 |
+
from pdfminer.layout import LAParams
|
21 |
+
from tabulate import tabulate
|
22 |
+
from pydantic import BaseModel
|
23 |
+
import unittest
|
24 |
+
|
25 |
+
# ๋ก๊น
์ค์
|
26 |
+
logging.basicConfig(
|
27 |
+
level=logging.INFO,
|
28 |
+
format='%(asctime)s - %(name)s - %(levelname)s - %(message)s',
|
29 |
+
handlers=[
|
30 |
+
logging.FileHandler('app.log'),
|
31 |
+
logging.StreamHandler()
|
32 |
+
]
|
33 |
+
)
|
34 |
+
logger = logging.getLogger(__name__)
|
35 |
+
|
36 |
+
# ์ค์ ํด๋์ค
|
37 |
+
class Config:
|
38 |
+
def __init__(self):
|
39 |
+
self.MODEL_ID = "CohereForAI/c4ai-command-r7b-12-2024"
|
40 |
+
self.MAX_HISTORY = 10
|
41 |
+
self.MAX_TOKENS = 4096
|
42 |
+
self.DEFAULT_TEMPERATURE = 0.8
|
43 |
+
self.HF_TOKEN = os.environ.get("HF_TOKEN", None)
|
44 |
+
self.MODELS = os.environ.get("MODELS")
|
45 |
+
|
46 |
+
config = Config()
|
47 |
+
|
48 |
+
# ์ปค์คํ
์์ธ ํด๋์ค
|
49 |
+
class FileProcessingError(Exception):
|
50 |
+
pass
|
51 |
+
|
52 |
+
# ์๋ต ๋ชจ๋ธ
|
53 |
+
class ChatResponse(BaseModel):
|
54 |
+
message: str
|
55 |
+
status: str
|
56 |
+
timestamp: datetime
|
57 |
+
|
58 |
+
# ํ์ผ ์ฒ๋ฆฌ ํด๋์ค
|
59 |
+
class FileProcessor:
|
60 |
+
@staticmethod
|
61 |
+
def process_pdf(file_path):
|
62 |
+
try:
|
63 |
+
with ThreadPoolExecutor() as executor:
|
64 |
+
pdf_reader = pypdf.PdfReader(file_path)
|
65 |
+
text = extract_text(
|
66 |
+
file_path,
|
67 |
+
laparams=LAParams(
|
68 |
+
line_margin=0.5,
|
69 |
+
word_margin=0.1,
|
70 |
+
char_margin=2.0,
|
71 |
+
all_texts=True
|
72 |
+
)
|
73 |
+
)
|
74 |
+
return text
|
75 |
+
except Exception as e:
|
76 |
+
raise FileProcessingError(f"PDF processing error: {str(e)}")
|
77 |
+
|
78 |
+
@staticmethod
|
79 |
+
def process_csv(file_path):
|
80 |
+
try:
|
81 |
+
encodings = ['utf-8', 'cp949', 'euc-kr', 'latin1']
|
82 |
+
for encoding in encodings:
|
83 |
+
try:
|
84 |
+
return pd.read_csv(file_path, encoding=encoding)
|
85 |
+
except UnicodeDecodeError:
|
86 |
+
continue
|
87 |
+
raise FileProcessingError("Unable to read CSV with supported encodings")
|
88 |
+
except Exception as e:
|
89 |
+
raise FileProcessingError(f"CSV processing error: {str(e)}")
|
90 |
+
|
91 |
+
# ๋ฉ๋ชจ๋ฆฌ ๊ด๋ฆฌ
|
92 |
+
@torch.no_grad()
|
93 |
+
def clear_cuda_memory():
|
94 |
+
if torch.cuda.is_available():
|
95 |
+
torch.cuda.empty_cache()
|
96 |
+
gc.collect()
|
97 |
+
|
98 |
+
# ๋ชจ๋ธ ๋ก๋
|
99 |
+
@spaces.GPU
|
100 |
+
def load_model():
|
101 |
+
try:
|
102 |
+
model = AutoModelForCausalLM.from_pretrained(
|
103 |
+
config.MODEL_ID,
|
104 |
+
torch_dtype=torch.bfloat16,
|
105 |
+
device_map="auto",
|
106 |
+
)
|
107 |
+
return model
|
108 |
+
except Exception as e:
|
109 |
+
logger.error(f"Model loading error: {str(e)}")
|
110 |
+
raise
|
111 |
+
|
112 |
+
# ์ปจํ
์คํธ ๊ฒ์
|
113 |
+
@lru_cache(maxsize=100)
|
114 |
+
def find_relevant_context(query, top_k=3):
|
115 |
+
try:
|
116 |
+
query_vector = vectorizer.transform([query])
|
117 |
+
similarities = (query_vector * question_vectors.T).toarray()[0]
|
118 |
+
top_indices = np.argsort(similarities)[-top_k:][::-1]
|
119 |
+
|
120 |
+
relevant_contexts = []
|
121 |
+
for idx in top_indices:
|
122 |
+
if similarities[idx] > 0:
|
123 |
+
relevant_contexts.append({
|
124 |
+
'question': questions[idx],
|
125 |
+
'answer': wiki_dataset['train']['answer'][idx],
|
126 |
+
'similarity': similarities[idx]
|
127 |
+
})
|
128 |
+
return relevant_contexts
|
129 |
+
except Exception as e:
|
130 |
+
logger.error(f"Context search error: {str(e)}")
|
131 |
+
return []
|
132 |
+
|
133 |
+
# ์คํธ๋ฆฌ๋ฐ ์ฑํ
|
134 |
+
@spaces.GPU
|
135 |
+
def stream_chat(message: str, history: list, uploaded_file, temperature: float,
|
136 |
+
max_new_tokens: int, top_p: float, top_k: int, penalty: float) -> Iterator[Tuple[str, list]]:
|
137 |
+
"""
|
138 |
+
์คํธ๋ฆฌ๋ฐ ์ฑํ
์๋ต์ ์์ฑํฉ๋๋ค.
|
139 |
+
|
140 |
+
Args:
|
141 |
+
message (str): ์ฌ์ฉ์ ์
๋ ฅ ๋ฉ์์ง
|
142 |
+
history (list): ๋ํ ํ์คํ ๋ฆฌ
|
143 |
+
uploaded_file: ์
๋ก๋๋ ํ์ผ
|
144 |
+
temperature (float): ์์ฑ ์จ๋
|
145 |
+
max_new_tokens (int): ์ต๋ ํ ํฐ ์
|
146 |
+
top_p (float): ์์ p ์ํ๋ง
|
147 |
+
top_k (int): ์์ k ์ํ๋ง
|
148 |
+
penalty (float): ๋ฐ๋ณต ํ๋ํฐ
|
149 |
+
|
150 |
+
Returns:
|
151 |
+
Iterator[Tuple[str, list]]: ์์ฑ๋ ์๋ต๊ณผ ์
๋ฐ์ดํธ๋ ํ์คํ ๋ฆฌ
|
152 |
+
"""
|
153 |
+
global model, current_file_context
|
154 |
+
|
155 |
+
try:
|
156 |
+
if model is None:
|
157 |
+
model = load_model()
|
158 |
+
|
159 |
+
logger.info(f'Processing message: {message}')
|
160 |
+
logger.debug(f'History length: {len(history)}')
|
161 |
+
|
162 |
+
# ํ์ผ ์ฒ๋ฆฌ
|
163 |
+
file_context = ""
|
164 |
+
if uploaded_file:
|
165 |
+
try:
|
166 |
+
file_ext = os.path.splitext(uploaded_file.name)[1].lower()
|
167 |
+
if file_ext == '.pdf':
|
168 |
+
content = FileProcessor.process_pdf(uploaded_file.name)
|
169 |
+
elif file_ext == '.csv':
|
170 |
+
content = FileProcessor.process_csv(uploaded_file.name)
|
171 |
+
else:
|
172 |
+
content = safe_file_read(uploaded_file.name)
|
173 |
+
|
174 |
+
file_context = analyze_file_content(content, file_ext)
|
175 |
+
current_file_context = file_context
|
176 |
+
except Exception as e:
|
177 |
+
logger.error(f"File processing error: {str(e)}")
|
178 |
+
file_context = f"\n\nโ File analysis error: {str(e)}"
|
179 |
+
|
180 |
+
# ์ปจํ
์คํธ ๊ฒ์ ๋ฐ ํ๋กฌํํธ ๊ตฌ์ฑ
|
181 |
+
relevant_contexts = find_relevant_context(message)
|
182 |
+
wiki_context = "\n\n๊ด๋ จ ์ํคํผ๋์ ์ ๋ณด:\n" + "\n".join([
|
183 |
+
f"Q: {ctx['question']}\nA: {ctx['answer']}\n์ ์ฌ๋: {ctx['similarity']:.3f}"
|
184 |
+
for ctx in relevant_contexts
|
185 |
+
])
|
186 |
+
|
187 |
+
# ํ ํฐํ ๋ฐ ์์ฑ
|
188 |
+
conversation = [
|
189 |
+
{"role": "user" if i % 2 == 0 else "assistant", "content": msg}
|
190 |
+
for hist in history[-config.MAX_HISTORY:]
|
191 |
+
for i, msg in enumerate(hist)
|
192 |
+
]
|
193 |
+
|
194 |
+
final_message = f"{file_context}{wiki_context}\nํ์ฌ ์ง๋ฌธ: {message}"
|
195 |
+
conversation.append({"role": "user", "content": final_message})
|
196 |
+
|
197 |
+
inputs = tokenizer(
|
198 |
+
tokenizer.apply_chat_template(conversation, tokenize=False, add_generation_prompt=True),
|
199 |
+
return_tensors="pt"
|
200 |
+
).to("cuda")
|
201 |
+
|
202 |
+
streamer = TextIteratorStreamer(tokenizer, timeout=10., skip_prompt=True, skip_special_tokens=True)
|
203 |
+
|
204 |
+
generate_kwargs = dict(
|
205 |
+
inputs,
|
206 |
+
streamer=streamer,
|
207 |
+
top_k=top_k,
|
208 |
+
top_p=top_p,
|
209 |
+
repetition_penalty=penalty,
|
210 |
+
max_new_tokens=min(max_new_tokens, 2048),
|
211 |
+
do_sample=True,
|
212 |
+
temperature=temperature,
|
213 |
+
eos_token_id=[255001],
|
214 |
+
)
|
215 |
+
|
216 |
+
clear_cuda_memory()
|
217 |
+
|
218 |
+
thread = Thread(target=model.generate, kwargs=generate_kwargs)
|
219 |
+
thread.start()
|
220 |
+
|
221 |
+
buffer = ""
|
222 |
+
for new_text in streamer:
|
223 |
+
buffer += new_text
|
224 |
+
yield "", history + [[message, buffer]]
|
225 |
+
|
226 |
+
clear_cuda_memory()
|
227 |
+
|
228 |
+
except Exception as e:
|
229 |
+
logger.error(f"Stream chat error: {str(e)}")
|
230 |
+
yield "", history + [[message, f"Error: {str(e)}"]]
|
231 |
+
clear_cuda_memory()
|
232 |
+
|
233 |
+
# UI ์์ฑ
|
234 |
+
def create_demo():
|
235 |
+
with gr.Blocks(css=UPDATED_CSS) as demo:
|
236 |
+
# UI ์ปดํฌ๋ํธ ๊ตฌ์ฑ
|
237 |
+
with gr.Column(elem_classes="markdown-style"):
|
238 |
+
gr.Markdown("""
|
239 |
+
# ๐ค RAGOndevice
|
240 |
+
#### ๐ RAG: Upload and Analyze Files (TXT, CSV, PDF, Parquet files)
|
241 |
+
Upload your files for data analysis and learning
|
242 |
+
""")
|
243 |
+
|
244 |
+
chatbot = gr.Chatbot(
|
245 |
+
value=[],
|
246 |
+
height=600,
|
247 |
+
label="GiniGEN AI Assistant",
|
248 |
+
elem_classes="chat-container"
|
249 |
+
)
|
250 |
+
|
251 |
+
# ์
๋ ฅ ์ปดํฌ๋ํธ
|
252 |
+
with gr.Row(elem_classes="input-container"):
|
253 |
+
with gr.Column(scale=1, min_width=70):
|
254 |
+
file_upload = gr.File(
|
255 |
+
type="filepath",
|
256 |
+
elem_classes="file-upload-icon",
|
257 |
+
scale=1,
|
258 |
+
container=True,
|
259 |
+
interactive=True,
|
260 |
+
show_label=False
|
261 |
+
)
|
262 |
+
|
263 |
+
with gr.Column(scale=3):
|
264 |
+
msg = gr.Textbox(
|
265 |
+
show_label=False,
|
266 |
+
placeholder="Type your message here... ๐ญ",
|
267 |
+
container=False,
|
268 |
+
elem_classes="input-textbox",
|
269 |
+
scale=1
|
270 |
+
)
|
271 |
+
|
272 |
+
with gr.Column(scale=1, min_width=70):
|
273 |
+
send = gr.Button(
|
274 |
+
"Send",
|
275 |
+
elem_classes="send-button custom-button",
|
276 |
+
scale=1
|
277 |
+
)
|
278 |
+
|
279 |
+
with gr.Column(scale=1, min_width=70):
|
280 |
+
clear = gr.Button(
|
281 |
+
"Clear",
|
282 |
+
elem_classes="clear-button custom-button",
|
283 |
+
scale=1
|
284 |
+
)
|
285 |
+
|
286 |
+
# ๊ณ ๊ธ ์ค์
|
287 |
+
with gr.Accordion("๐ฎ Advanced Settings", open=False):
|
288 |
+
with gr.Row():
|
289 |
+
with gr.Column(scale=1):
|
290 |
+
temperature = gr.Slider(
|
291 |
+
minimum=0, maximum=1, step=0.1, value=config.DEFAULT_TEMPERATURE,
|
292 |
+
label="Creativity Level ๐จ"
|
293 |
+
)
|
294 |
+
max_new_tokens = gr.Slider(
|
295 |
+
minimum=128, maximum=8000, step=1, value=4000,
|
296 |
+
label="Maximum Token Count ๐"
|
297 |
+
)
|
298 |
+
with gr.Column(scale=1):
|
299 |
+
top_p = gr.Slider(
|
300 |
+
minimum=0.0, maximum=1.0, step=0.1, value=0.8,
|
301 |
+
label="Diversity Control ๐ฏ"
|
302 |
+
)
|
303 |
+
top_k = gr.Slider(
|
304 |
+
minimum=1, maximum=20, step=1, value=20,
|
305 |
+
label="Selection Range ๐"
|
306 |
+
)
|
307 |
+
penalty = gr.Slider(
|
308 |
+
minimum=0.0, maximum=2.0, step=0.1, value=1.0,
|
309 |
+
label="Repetition Penalty ๐"
|
310 |
+
)
|
311 |
+
|
312 |
+
# ์ด๋ฒคํธ ๋ฐ์ธ๋ฉ
|
313 |
+
msg.submit(stream_chat, [msg, chatbot, file_upload, temperature, max_new_tokens, top_p, top_k, penalty], [msg, chatbot])
|
314 |
+
send.click(stream_chat, [msg, chatbot, file_upload, temperature, max_new_tokens, top_p, top_k, penalty], [msg, chatbot])
|
315 |
+
clear.click(lambda: ([], None, ""), outputs=[chatbot, file_upload, msg])
|
316 |
+
|
317 |
+
return demo
|
318 |
+
|
319 |
+
# ๋ฉ์ธ ์คํ
|
320 |
+
if __name__ == "__main__":
|
321 |
+
# ์ํคํผ๋์ ๋ฐ์ดํฐ์
๋ก๋
|
322 |
+
wiki_dataset = load_dataset("lcw99/wikipedia-korean-20240501-1million-qna")
|
323 |
+
logger.info("Wikipedia dataset loaded")
|
324 |
+
|
325 |
+
# TF-IDF ๋ฒกํฐ๋ผ์ด์ ์ด๊ธฐํ
|
326 |
+
questions = wiki_dataset['train']['question'][:10000]
|
327 |
+
vectorizer = TfidfVectorizer(max_features=1000)
|
328 |
+
question_vectors = vectorizer.fit_transform(questions)
|
329 |
+
logger.info("TF-IDF vectorization completed")
|
330 |
+
|
331 |
+
# UI ์คํ
|
332 |
+
demo = create_demo()
|
333 |
+
demo.launch()
|
334 |
+
|
335 |
+
# ํ
์คํธ ์ฝ๋
|
336 |
+
class TestChatBot(unittest.TestCase):
|
337 |
+
def test_file_processing(self):
|
338 |
+
# ํ
์คํธ ๊ตฌํ
|
339 |
+
pass
|
340 |
+
|
341 |
+
def test_context_search(self):
|
342 |
+
# ํ
์คํธ ๊ตฌํ
|
343 |
+
pass
|