Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
|
@@ -10,18 +10,30 @@ import pandas as pd
|
|
| 10 |
import os
|
| 11 |
import psutil
|
| 12 |
import gc
|
| 13 |
-
import
|
| 14 |
-
|
| 15 |
-
|
|
|
|
|
|
|
| 16 |
|
| 17 |
# Set environment variables to optimize CPU performance
|
| 18 |
os.environ["OMP_NUM_THREADS"] = str(psutil.cpu_count(logical=False))
|
| 19 |
os.environ["MKL_NUM_THREADS"] = str(psutil.cpu_count(logical=False))
|
|
|
|
| 20 |
|
| 21 |
# Set device globally
|
| 22 |
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
|
| 23 |
|
| 24 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 25 |
def load_model(model_name, model_class, is_bc=False, device=None):
|
| 26 |
if device is None:
|
| 27 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
|
@@ -30,18 +42,44 @@ def load_model(model_name, model_class, is_bc=False, device=None):
|
|
| 30 |
model = model_class.from_pretrained(model_name, num_labels=3 if not is_bc else 2)
|
| 31 |
model.eval()
|
| 32 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 33 |
model.to(device)
|
| 34 |
return tokenizer, model
|
| 35 |
|
| 36 |
-
|
|
|
|
| 37 |
def preprocess_text(text):
|
| 38 |
# Add any text cleaning or normalization here
|
| 39 |
return text.strip()
|
| 40 |
|
| 41 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 42 |
def perform_verification(claim, context, model_qatc, tokenizer_qatc, model_tc, tokenizer_tc,
|
| 43 |
-
|
| 44 |
-
# Extract evidence
|
| 45 |
evidence_start_time = time.time()
|
| 46 |
evidence = extract_evidence_tfidf_qatc(
|
| 47 |
claim, context, model_qatc, tokenizer_qatc,
|
|
@@ -51,25 +89,25 @@ def perform_verification(claim, context, model_qatc, tokenizer_qatc, model_tc, t
|
|
| 51 |
)
|
| 52 |
evidence_time = time.time() - evidence_start_time
|
| 53 |
|
| 54 |
-
#
|
|
|
|
|
|
|
| 55 |
gc.collect()
|
| 56 |
|
| 57 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 58 |
verdict_start_time = time.time()
|
| 59 |
with torch.no_grad():
|
| 60 |
verdict = "NEI"
|
| 61 |
-
prob3class, pred_tc = classify_claim(
|
| 62 |
-
claim, evidence, model_tc, tokenizer_tc, DEVICE
|
| 63 |
-
)
|
| 64 |
-
|
| 65 |
-
# Only run binary classifier if needed
|
| 66 |
-
prob2class, pred_bc = 0, 0
|
| 67 |
if pred_tc != 0:
|
| 68 |
-
prob2class, pred_bc = classify_claim(
|
| 69 |
-
claim, evidence, model_bc, tokenizer_bc, DEVICE
|
| 70 |
-
)
|
| 71 |
verdict = "SUPPORTED" if pred_bc == 0 else "REFUTED" if prob2class > prob3class else ["NEI", "SUPPORTED", "REFUTED"][pred_tc]
|
| 72 |
-
|
| 73 |
verdict_time = time.time() - verdict_start_time
|
| 74 |
|
| 75 |
return {
|
|
@@ -83,69 +121,19 @@ def perform_verification(claim, context, model_qatc, tokenizer_qatc, model_tc, t
|
|
| 83 |
"pred_bc": pred_bc
|
| 84 |
}
|
| 85 |
|
| 86 |
-
# Add
|
| 87 |
-
def
|
| 88 |
-
if
|
| 89 |
-
return
|
| 90 |
-
|
| 91 |
-
|
| 92 |
-
|
| 93 |
-
|
| 94 |
-
|
| 95 |
-
|
| 96 |
-
|
| 97 |
-
|
| 98 |
-
|
| 99 |
-
)
|
| 100 |
-
return fig
|
| 101 |
-
|
| 102 |
-
def analyze_processing_time(history):
|
| 103 |
-
if not history:
|
| 104 |
-
return None
|
| 105 |
-
|
| 106 |
-
df = pd.DataFrame(history)
|
| 107 |
-
df['timestamp'] = pd.to_datetime(df['timestamp'])
|
| 108 |
-
|
| 109 |
-
fig = px.line(
|
| 110 |
-
df,
|
| 111 |
-
x='timestamp',
|
| 112 |
-
y=['evidence_time', 'verdict_time', 'total_time'],
|
| 113 |
-
title='Thời gian Xử lý theo Thời gian',
|
| 114 |
-
labels={'value': 'Thời gian (giây)', 'timestamp': 'Thời điểm'}
|
| 115 |
-
)
|
| 116 |
-
return fig
|
| 117 |
-
|
| 118 |
-
def generate_report(result):
|
| 119 |
-
report = f"""
|
| 120 |
-
BÁO CÁO KIỂM CHỨNG THÔNG TIN
|
| 121 |
-
Thời gian: {datetime.now().strftime('%d/%m/%Y %H:%M:%S')}
|
| 122 |
-
|
| 123 |
-
1. THÔNG TIN CƠ BẢN
|
| 124 |
-
-------------------
|
| 125 |
-
Câu khẳng định: {result['claim']}
|
| 126 |
-
Kết luận: {result['verdict']}
|
| 127 |
-
|
| 128 |
-
2. BẰNG CHỨNG
|
| 129 |
-
-------------
|
| 130 |
-
{result['evidence']}
|
| 131 |
-
|
| 132 |
-
3. THỐNG KÊ THỜI GIAN
|
| 133 |
-
---------------------
|
| 134 |
-
- Thời gian trích xuất bằng chứng: {result['evidence_time']:.2f} giây
|
| 135 |
-
- Thời gian phân loại: {result['verdict_time']:.2f} giây
|
| 136 |
-
- Tổng thời gian xử lý: {result['total_time']:.2f} giây
|
| 137 |
-
|
| 138 |
-
4. CHI TIẾT KỸ THUẬT
|
| 139 |
-
-------------------
|
| 140 |
-
{result['details']}
|
| 141 |
-
|
| 142 |
-
5. MÔ HÌNH SỬ DỤNG
|
| 143 |
-
------------------
|
| 144 |
-
- QATC Model: {result['qatc_model']}
|
| 145 |
-
- Binary Classification Model: {result['bc_model']}
|
| 146 |
-
- 3-Class Classification Model: {result['tc_model']}
|
| 147 |
-
"""
|
| 148 |
-
return report
|
| 149 |
|
| 150 |
# Set page configuration
|
| 151 |
st.set_page_config(
|
|
@@ -160,295 +148,195 @@ st.markdown("""
|
|
| 160 |
<style>
|
| 161 |
/* Main theme colors */
|
| 162 |
:root {
|
| 163 |
-
--primary-color: #
|
| 164 |
-
--secondary-color: #
|
| 165 |
--accent-color: #e74c3c;
|
| 166 |
-
--success-color: #2ecc71;
|
| 167 |
-
--warning-color: #f1c40f;
|
| 168 |
--background-color: #f8f9fa;
|
| 169 |
--text-color: #2c3e50;
|
| 170 |
--border-color: #e0e0e0;
|
| 171 |
-
--gradient-start: #2c3e50;
|
| 172 |
-
--gradient-end: #3498db;
|
| 173 |
}
|
| 174 |
|
| 175 |
/* General styling */
|
| 176 |
.stApp {
|
| 177 |
background-color: var(--background-color);
|
| 178 |
color: var(--text-color);
|
| 179 |
-
font-family: 'Segoe UI', Tahoma, Geneva, Verdana, sans-serif;
|
| 180 |
}
|
| 181 |
|
| 182 |
/* Header styling */
|
| 183 |
.main-header {
|
| 184 |
-
background: linear-gradient(135deg, var(--
|
| 185 |
color: white;
|
| 186 |
-
padding:
|
| 187 |
-
border-radius:
|
| 188 |
margin-bottom: 2rem;
|
| 189 |
-
box-shadow: 0
|
| 190 |
-
position: relative;
|
| 191 |
-
overflow: hidden;
|
| 192 |
-
}
|
| 193 |
-
|
| 194 |
-
.main-header::before {
|
| 195 |
-
content: '';
|
| 196 |
-
position: absolute;
|
| 197 |
-
top: 0;
|
| 198 |
-
left: 0;
|
| 199 |
-
right: 0;
|
| 200 |
-
bottom: 0;
|
| 201 |
-
background: url('data:image/svg+xml,<svg width="20" height="20" viewBox="0 0 20 20" xmlns="http://www.w3.org/2000/svg"><rect width="1" height="1" fill="rgba(255,255,255,0.05)"/></svg>');
|
| 202 |
-
opacity: 0.1;
|
| 203 |
}
|
| 204 |
|
| 205 |
.main-title {
|
| 206 |
-
font-size:
|
| 207 |
-
font-weight:
|
| 208 |
text-align: center;
|
| 209 |
margin-bottom: 1rem;
|
| 210 |
-
text-shadow: 2px 2px 4px rgba(0, 0, 0, 0.2);
|
| 211 |
-
letter-spacing: 1px;
|
| 212 |
}
|
| 213 |
|
| 214 |
.sub-title {
|
| 215 |
-
font-size: 1.
|
| 216 |
text-align: center;
|
| 217 |
opacity: 0.9;
|
| 218 |
-
font-weight: 300;
|
| 219 |
}
|
| 220 |
|
| 221 |
/* Input styling */
|
| 222 |
.stTextArea textarea {
|
| 223 |
border: 2px solid var(--border-color);
|
| 224 |
-
border-radius:
|
| 225 |
-
padding:
|
| 226 |
-
font-size:
|
| 227 |
min-height: 150px;
|
| 228 |
background-color: white;
|
| 229 |
-
transition: all 0.3s ease;
|
| 230 |
-
box-shadow: 0 2px 4px rgba(0, 0, 0, 0.05);
|
| 231 |
-
}
|
| 232 |
-
|
| 233 |
-
.stTextArea textarea:focus {
|
| 234 |
-
border-color: var(--secondary-color);
|
| 235 |
-
box-shadow: 0 4px 8px rgba(52, 152, 219, 0.1);
|
| 236 |
}
|
| 237 |
|
| 238 |
/* Button styling */
|
| 239 |
.stButton>button {
|
| 240 |
-
background: linear-gradient(135deg, var(--
|
| 241 |
color: white;
|
| 242 |
border: none;
|
| 243 |
-
border-radius:
|
| 244 |
-
padding:
|
| 245 |
-
font-size: 1.
|
| 246 |
-
font-weight:
|
| 247 |
transition: all 0.3s ease;
|
| 248 |
-
text-transform: uppercase;
|
| 249 |
-
letter-spacing: 1px;
|
| 250 |
}
|
| 251 |
|
| 252 |
.stButton>button:hover {
|
| 253 |
transform: translateY(-2px);
|
| 254 |
-
box-shadow: 0 8px
|
| 255 |
}
|
| 256 |
|
| 257 |
/* Result box styling */
|
| 258 |
.result-box {
|
| 259 |
background-color: white;
|
| 260 |
-
border-radius:
|
| 261 |
-
padding:
|
| 262 |
-
margin:
|
| 263 |
-
box-shadow: 0
|
| 264 |
-
transition: all 0.3s ease;
|
| 265 |
-
border: 1px solid rgba(0, 0, 0, 0.05);
|
| 266 |
}
|
| 267 |
|
| 268 |
-
|
| 269 |
-
|
| 270 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 271 |
}
|
| 272 |
|
| 273 |
.verdict {
|
| 274 |
-
font-size:
|
| 275 |
-
font-weight:
|
| 276 |
-
padding:
|
| 277 |
-
border-radius:
|
| 278 |
-
margin:
|
| 279 |
text-align: center;
|
| 280 |
-
transition: all 0.3s ease;
|
| 281 |
}
|
| 282 |
|
| 283 |
.verdict-supported {
|
| 284 |
-
background:
|
| 285 |
-
color:
|
| 286 |
-
box-shadow: 0 4px 8px rgba(46, 204, 113, 0.2);
|
| 287 |
}
|
| 288 |
|
| 289 |
.verdict-refuted {
|
| 290 |
-
background:
|
| 291 |
-
color:
|
| 292 |
-
box-shadow: 0 4px 8px rgba(231, 76, 60, 0.2);
|
| 293 |
}
|
| 294 |
|
| 295 |
.verdict-nei {
|
| 296 |
-
background:
|
| 297 |
-
color:
|
| 298 |
-
box-shadow: 0 4px 8px rgba(241, 196, 15, 0.2);
|
| 299 |
}
|
| 300 |
|
| 301 |
/* Sidebar styling */
|
| 302 |
.css-1d391kg {
|
| 303 |
background-color: white;
|
| 304 |
-
padding:
|
| 305 |
-
border-radius:
|
| 306 |
-
box-shadow: 0
|
| 307 |
}
|
| 308 |
|
| 309 |
/* Stats box styling */
|
| 310 |
.stats-box {
|
| 311 |
-
background:
|
| 312 |
-
border-radius:
|
| 313 |
-
padding:
|
| 314 |
-
margin:
|
| 315 |
-
box-shadow: 0 4px
|
| 316 |
}
|
| 317 |
|
| 318 |
/* Code block styling */
|
| 319 |
.code-block {
|
| 320 |
-
background-color: #
|
| 321 |
-
border
|
| 322 |
-
|
| 323 |
-
|
| 324 |
-
|
| 325 |
-
|
| 326 |
-
overflow-x: auto;
|
| 327 |
}
|
| 328 |
|
| 329 |
/* Tab styling */
|
| 330 |
.stTabs [data-baseweb="tab-list"] {
|
| 331 |
gap: 2rem;
|
| 332 |
-
background-color: transparent;
|
| 333 |
}
|
| 334 |
|
| 335 |
.stTabs [data-baseweb="tab"] {
|
| 336 |
background-color: white;
|
| 337 |
-
border-radius:
|
| 338 |
-
padding: 0.
|
| 339 |
margin: 0 0.5rem;
|
| 340 |
-
font-weight: 600;
|
| 341 |
-
transition: all 0.3s ease;
|
| 342 |
-
box-shadow: 0 2px 4px rgba(0, 0, 0, 0.05);
|
| 343 |
}
|
| 344 |
|
| 345 |
.stTabs [aria-selected="true"] {
|
| 346 |
-
background: linear-gradient(135deg, var(--secondary-color), var(--primary-color));
|
| 347 |
-
color: white;
|
| 348 |
-
box-shadow: 0 4px 8px rgba(52, 152, 219, 0.2);
|
| 349 |
-
}
|
| 350 |
-
|
| 351 |
-
/* Analysis box styling */
|
| 352 |
-
.analysis-box {
|
| 353 |
-
background-color: white;
|
| 354 |
-
border-radius: 15px;
|
| 355 |
-
padding: 2rem;
|
| 356 |
-
margin: 1.5rem 0;
|
| 357 |
-
box-shadow: 0 8px 16px rgba(0, 0, 0, 0.1);
|
| 358 |
-
border-left: 4px solid var(--secondary-color);
|
| 359 |
-
}
|
| 360 |
-
|
| 361 |
-
/* Search box styling */
|
| 362 |
-
.search-box {
|
| 363 |
-
background-color: white;
|
| 364 |
-
border-radius: 12px;
|
| 365 |
-
padding: 1.5rem;
|
| 366 |
-
margin-bottom: 1.5rem;
|
| 367 |
-
box-shadow: 0 4px 8px rgba(0, 0, 0, 0.05);
|
| 368 |
-
}
|
| 369 |
-
|
| 370 |
-
/* Comparison box styling */
|
| 371 |
-
.comparison-box {
|
| 372 |
-
background-color: white;
|
| 373 |
-
border-radius: 15px;
|
| 374 |
-
padding: 2rem;
|
| 375 |
-
margin: 1.5rem 0;
|
| 376 |
-
box-shadow: 0 8px 16px rgba(0, 0, 0, 0.1);
|
| 377 |
-
border-left: 4px solid var(--secondary-color);
|
| 378 |
-
}
|
| 379 |
-
|
| 380 |
-
/* Selectbox styling */
|
| 381 |
-
.stSelectbox select {
|
| 382 |
-
border-radius: 8px;
|
| 383 |
-
padding: 0.8rem;
|
| 384 |
-
font-size: 1rem;
|
| 385 |
-
border: 2px solid var(--border-color);
|
| 386 |
-
background-color: white;
|
| 387 |
-
transition: all 0.3s ease;
|
| 388 |
-
}
|
| 389 |
-
|
| 390 |
-
.stSelectbox select:focus {
|
| 391 |
-
border-color: var(--secondary-color);
|
| 392 |
-
box-shadow: 0 4px 8px rgba(52, 152, 219, 0.1);
|
| 393 |
-
}
|
| 394 |
-
|
| 395 |
-
/* Slider styling */
|
| 396 |
-
.stSlider > div > div {
|
| 397 |
-
background-color: var(--secondary-color);
|
| 398 |
-
}
|
| 399 |
-
|
| 400 |
-
/* Checkbox styling */
|
| 401 |
-
.stCheckbox > label {
|
| 402 |
-
font-weight: 500;
|
| 403 |
-
}
|
| 404 |
-
|
| 405 |
-
/* Info box styling */
|
| 406 |
-
.info-box {
|
| 407 |
-
background: linear-gradient(135deg, #f8f9fa, #e9ecef);
|
| 408 |
-
border-radius: 15px;
|
| 409 |
-
padding: 2rem;
|
| 410 |
-
margin: 1.5rem 0;
|
| 411 |
-
box-shadow: 0 8px 16px rgba(0, 0, 0, 0.1);
|
| 412 |
-
}
|
| 413 |
-
|
| 414 |
-
/* Chart container styling */
|
| 415 |
-
.js-plotly-plot {
|
| 416 |
-
border-radius: 12px !important;
|
| 417 |
-
box-shadow: 0 4px 8px rgba(0, 0, 0, 0.05) !important;
|
| 418 |
-
}
|
| 419 |
-
|
| 420 |
-
/* Dataframe styling */
|
| 421 |
-
.dataframe {
|
| 422 |
-
border-radius: 12px;
|
| 423 |
-
overflow: hidden;
|
| 424 |
-
box-shadow: 0 4px 8px rgba(0, 0, 0, 0.05);
|
| 425 |
-
}
|
| 426 |
-
|
| 427 |
-
.dataframe thead th {
|
| 428 |
background-color: var(--primary-color);
|
| 429 |
color: white;
|
| 430 |
}
|
| 431 |
-
|
| 432 |
-
/* Download button styling */
|
| 433 |
-
.stDownloadButton > button {
|
| 434 |
-
background: linear-gradient(135deg, var(--success-color), #27ae60);
|
| 435 |
-
color: white;
|
| 436 |
-
border: none;
|
| 437 |
-
border-radius: 12px;
|
| 438 |
-
padding: 0.8rem 2rem;
|
| 439 |
-
font-size: 1.1rem;
|
| 440 |
-
font-weight: 600;
|
| 441 |
-
transition: all 0.3s ease;
|
| 442 |
-
}
|
| 443 |
-
|
| 444 |
-
.stDownloadButton > button:hover {
|
| 445 |
-
transform: translateY(-2px);
|
| 446 |
-
box-shadow: 0 4px 8px rgba(46, 204, 113, 0.2);
|
| 447 |
-
}
|
| 448 |
</style>
|
| 449 |
""", unsafe_allow_html=True)
|
| 450 |
|
| 451 |
-
# Main header
|
| 452 |
st.markdown("""
|
| 453 |
<div class="main-header">
|
| 454 |
<div class="main-title">SemViQA</div>
|
|
@@ -456,15 +344,11 @@ st.markdown("""
|
|
| 456 |
</div>
|
| 457 |
""", unsafe_allow_html=True)
|
| 458 |
|
| 459 |
-
# Sidebar
|
| 460 |
with st.sidebar:
|
| 461 |
-
st.markdown(""
|
| 462 |
-
<div class="info-box">
|
| 463 |
-
<h3>⚙️ Cài đặt Hệ thống</h3>
|
| 464 |
-
</div>
|
| 465 |
-
""", unsafe_allow_html=True)
|
| 466 |
|
| 467 |
-
# Model selection
|
| 468 |
st.markdown("#### 🧠 Chọn Mô hình")
|
| 469 |
qatc_model_name = st.selectbox(
|
| 470 |
"Mô hình QATC",
|
|
@@ -500,7 +384,7 @@ with st.sidebar:
|
|
| 500 |
]
|
| 501 |
)
|
| 502 |
|
| 503 |
-
# Threshold settings
|
| 504 |
st.markdown("#### ⚖️ Ngưỡng Phân tích")
|
| 505 |
tfidf_threshold = st.slider(
|
| 506 |
"Ngưỡng TF-IDF",
|
|
@@ -514,7 +398,7 @@ with st.sidebar:
|
|
| 514 |
help="Điều chỉnh độ dài tối đa của bằng chứng"
|
| 515 |
)
|
| 516 |
|
| 517 |
-
# Display settings
|
| 518 |
st.markdown("#### 👁️ Hiển thị")
|
| 519 |
show_details = st.checkbox(
|
| 520 |
"Hiển thị Chi tiết Xác suất",
|
|
@@ -522,7 +406,7 @@ with st.sidebar:
|
|
| 522 |
help="Hiển thị thông tin chi tiết về xác suất dự đoán"
|
| 523 |
)
|
| 524 |
|
| 525 |
-
# Performance settings
|
| 526 |
st.markdown("#### ⚡ Hiệu suất")
|
| 527 |
num_threads = st.slider(
|
| 528 |
"Số luồng CPU",
|
|
@@ -534,7 +418,7 @@ with st.sidebar:
|
|
| 534 |
os.environ["MKL_NUM_THREADS"] = str(num_threads)
|
| 535 |
|
| 536 |
# Main content
|
| 537 |
-
tabs = st.tabs(["🔍 Kiểm chứng", "📊 Lịch sử", "
|
| 538 |
|
| 539 |
tokenizer_qatc, model_qatc = load_model(qatc_model_name, QATCForQuestionAnswering, device=DEVICE)
|
| 540 |
tokenizer_bc, model_bc = load_model(bc_model_name, ClaimModelForClassification, is_bc=True, device=DEVICE)
|
|
@@ -575,8 +459,14 @@ with tabs[0]:
|
|
| 575 |
|
| 576 |
# Clear memory and perform verification
|
| 577 |
gc.collect()
|
|
|
|
|
|
|
|
|
|
| 578 |
start_time = time.time()
|
| 579 |
|
|
|
|
|
|
|
|
|
|
| 580 |
result = perform_verification(
|
| 581 |
preprocessed_claim, preprocessed_context,
|
| 582 |
model_qatc, tokenizer_qatc,
|
|
@@ -587,6 +477,9 @@ with tabs[0]:
|
|
| 587 |
|
| 588 |
total_time = time.time() - start_time
|
| 589 |
|
|
|
|
|
|
|
|
|
|
| 590 |
# Format details
|
| 591 |
details = ""
|
| 592 |
if show_details:
|
|
@@ -595,9 +488,15 @@ with tabs[0]:
|
|
| 595 |
3-Class Predicted Label: {['NEI', 'SUPPORTED', 'REFUTED'][result['pred_tc']]}
|
| 596 |
2-Class Probability: {result['prob2class']:.2f}
|
| 597 |
2-Class Predicted Label: {['SUPPORTED', 'REFUTED'][result['pred_bc']] if isinstance(result['pred_bc'], int) and result['pred_tc'] != 0 else 'Not used'}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 598 |
"""
|
| 599 |
|
| 600 |
-
# Store result
|
| 601 |
st.session_state.latest_result = {
|
| 602 |
"claim": claim,
|
| 603 |
"evidence": result['evidence'],
|
|
@@ -608,7 +507,8 @@ with tabs[0]:
|
|
| 608 |
"details": details,
|
| 609 |
"qatc_model": qatc_model_name,
|
| 610 |
"bc_model": bc_model_name,
|
| 611 |
-
"tc_model": tc_model_name
|
|
|
|
| 612 |
}
|
| 613 |
|
| 614 |
# Add to history
|
|
@@ -616,7 +516,7 @@ with tabs[0]:
|
|
| 616 |
st.session_state.history = []
|
| 617 |
st.session_state.history.append(st.session_state.latest_result)
|
| 618 |
|
| 619 |
-
# Display result
|
| 620 |
res = st.session_state.latest_result
|
| 621 |
verdict_class = {
|
| 622 |
"SUPPORTED": "verdict-supported",
|
|
@@ -636,13 +536,29 @@ with tabs[0]:
|
|
| 636 |
<p><strong>Thời gian trích xuất bằng chứng:</strong> {res['evidence_time']:.2f} giây</p>
|
| 637 |
<p><strong>Thời gian phân loại:</strong> {res['verdict_time']:.2f} giây</p>
|
| 638 |
<p><strong>Tổng thời gian:</strong> {res['total_time']:.2f} giây</p>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 639 |
</div>
|
| 640 |
{f"<div class='code-block'><pre>{res['details']}</pre></div>" if show_details else ""}
|
| 641 |
</div>
|
| 642 |
""", unsafe_allow_html=True)
|
| 643 |
|
| 644 |
-
# Download button
|
| 645 |
-
result_text = f"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 646 |
st.download_button(
|
| 647 |
"📥 Tải kết quả",
|
| 648 |
data=result_text,
|
|
@@ -655,22 +571,9 @@ with tabs[0]:
|
|
| 655 |
# --- Tab History ---
|
| 656 |
with tabs[1]:
|
| 657 |
st.markdown("### 📊 Lịch sử Kiểm chứng")
|
| 658 |
-
|
| 659 |
-
# Add search functionality
|
| 660 |
-
search_query = st.text_input("🔍 Tìm kiếm trong lịch sử", "")
|
| 661 |
-
|
| 662 |
if 'history' in st.session_state and st.session_state.history:
|
| 663 |
-
# Filter history based on search query
|
| 664 |
-
filtered_history = st.session_state.history
|
| 665 |
-
if search_query:
|
| 666 |
-
filtered_history = [
|
| 667 |
-
record for record in st.session_state.history
|
| 668 |
-
if search_query.lower() in record['claim'].lower() or
|
| 669 |
-
search_query.lower() in record['evidence'].lower()
|
| 670 |
-
]
|
| 671 |
-
|
| 672 |
# Download full history
|
| 673 |
-
history_df = pd.DataFrame(
|
| 674 |
st.download_button(
|
| 675 |
"📥 Tải toàn bộ lịch sử",
|
| 676 |
data=history_df.to_csv(index=False).encode('utf-8'),
|
|
@@ -678,63 +581,23 @@ with tabs[1]:
|
|
| 678 |
mime="text/csv"
|
| 679 |
)
|
| 680 |
|
| 681 |
-
# Display history
|
| 682 |
-
for idx, record in enumerate(reversed(
|
| 683 |
-
|
| 684 |
-
|
| 685 |
-
|
| 686 |
-
|
| 687 |
-
<
|
| 688 |
-
|
| 689 |
-
|
| 690 |
-
|
| 691 |
-
<p><strong>Thời gian:</strong> {record['total_time']:.2f} giây</p>
|
| 692 |
-
</div>
|
| 693 |
-
""", unsafe_allow_html=True)
|
| 694 |
-
|
| 695 |
-
with col2:
|
| 696 |
-
if st.button("🔄 So sánh", key=f"compare_{idx}"):
|
| 697 |
-
st.session_state.selected_for_comparison = record
|
| 698 |
-
|
| 699 |
-
# --- Tab Analysis ---
|
| 700 |
-
with tabs[2]:
|
| 701 |
-
st.markdown("### 📈 Phân tích Chi tiết")
|
| 702 |
-
|
| 703 |
-
if 'history' in st.session_state and st.session_state.history:
|
| 704 |
-
# Add timestamp to history records
|
| 705 |
-
for record in st.session_state.history:
|
| 706 |
-
if 'timestamp' not in record:
|
| 707 |
-
record['timestamp'] = datetime.now()
|
| 708 |
-
|
| 709 |
-
# Distribution analysis
|
| 710 |
-
st.markdown("#### 📊 Phân bố Kết quả")
|
| 711 |
-
verdict_fig = analyze_verdict_distribution(st.session_state.history)
|
| 712 |
-
if verdict_fig:
|
| 713 |
-
st.plotly_chart(verdict_fig, use_container_width=True)
|
| 714 |
-
|
| 715 |
-
# Processing time analysis
|
| 716 |
-
st.markdown("#### ⏱️ Phân tích Thời gian Xử lý")
|
| 717 |
-
time_fig = analyze_processing_time(st.session_state.history)
|
| 718 |
-
if time_fig:
|
| 719 |
-
st.plotly_chart(time_fig, use_container_width=True)
|
| 720 |
-
|
| 721 |
-
# Model performance analysis
|
| 722 |
-
st.markdown("#### 🧠 Phân tích Hiệu suất Mô hình")
|
| 723 |
-
model_stats = pd.DataFrame(st.session_state.history)
|
| 724 |
-
if not model_stats.empty:
|
| 725 |
-
st.markdown("##### Thống kê theo Mô hình")
|
| 726 |
-
model_performance = model_stats.groupby(['qatc_model', 'bc_model', 'tc_model']).agg({
|
| 727 |
-
'total_time': ['mean', 'count'],
|
| 728 |
-
'verdict': lambda x: (x == 'SUPPORTED').mean()
|
| 729 |
-
}).round(2)
|
| 730 |
-
st.dataframe(model_performance)
|
| 731 |
else:
|
| 732 |
-
st.info("Chưa có
|
| 733 |
|
| 734 |
# --- Tab Info ---
|
| 735 |
-
with tabs[
|
| 736 |
st.markdown("""
|
| 737 |
-
<div class="
|
| 738 |
<h3>ℹ️ Thông tin về SemViQA</h3>
|
| 739 |
<p>SemViQA là hệ thống kiểm chứng thông tin tự động cho tiếng Việt, được phát triển bởi nhóm nghiên cứu tại Đại học Công nghệ Thông tin - Đại học Quốc gia TP.HCM.</p>
|
| 740 |
|
|
@@ -759,13 +622,5 @@ with tabs[3]:
|
|
| 759 |
<li><strong>REFUTED:</strong> Câu khẳng định bị bác bỏ bởi bằng chứng</li>
|
| 760 |
<li><strong>NEI:</strong> Không đủ bằng chứng để kết luận</li>
|
| 761 |
</ul>
|
| 762 |
-
|
| 763 |
-
<h4>🆕 Tính năng Mới</h4>
|
| 764 |
-
<ul>
|
| 765 |
-
<li><strong>Phân tích Chi tiết:</strong> Xem thống kê và biểu đồ về kết quả kiểm chứng</li>
|
| 766 |
-
<li><strong>Tìm kiếm Lịch sử:</strong> Dễ dàng tìm kiếm trong lịch sử kiểm chứng</li>
|
| 767 |
-
<li><strong>So sánh Kết quả:</strong> So sánh các kết quả kiểm chứng với nhau</li>
|
| 768 |
-
<li><strong>Báo cáo Chi tiết:</strong> Xuất báo cáo chi tiết về kết quả kiểm chứng</li>
|
| 769 |
-
</ul>
|
| 770 |
</div>
|
| 771 |
""", unsafe_allow_html=True)
|
|
|
|
| 10 |
import os
|
| 11 |
import psutil
|
| 12 |
import gc
|
| 13 |
+
import numpy as np
|
| 14 |
+
from functools import lru_cache
|
| 15 |
+
import threading
|
| 16 |
+
from concurrent.futures import ThreadPoolExecutor
|
| 17 |
+
import torch.nn.functional as F
|
| 18 |
|
| 19 |
# Set environment variables to optimize CPU performance
|
| 20 |
os.environ["OMP_NUM_THREADS"] = str(psutil.cpu_count(logical=False))
|
| 21 |
os.environ["MKL_NUM_THREADS"] = str(psutil.cpu_count(logical=False))
|
| 22 |
+
torch.set_num_threads(psutil.cpu_count(logical=False))
|
| 23 |
|
| 24 |
# Set device globally
|
| 25 |
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
|
| 26 |
|
| 27 |
+
# Cache for model outputs
|
| 28 |
+
@lru_cache(maxsize=1000)
|
| 29 |
+
def cached_classify_claim(claim, evidence, model_name, is_bc=False):
|
| 30 |
+
tokenizer, model = load_model(model_name, ClaimModelForClassification, is_bc=is_bc, device=DEVICE)
|
| 31 |
+
with torch.no_grad():
|
| 32 |
+
prob, pred = classify_claim(claim, evidence, model, tokenizer, DEVICE)
|
| 33 |
+
return prob, pred
|
| 34 |
+
|
| 35 |
+
# Optimized model loading with caching
|
| 36 |
+
@st.cache_resource(ttl=3600) # Cache for 1 hour
|
| 37 |
def load_model(model_name, model_class, is_bc=False, device=None):
|
| 38 |
if device is None:
|
| 39 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
|
|
|
| 42 |
model = model_class.from_pretrained(model_name, num_labels=3 if not is_bc else 2)
|
| 43 |
model.eval()
|
| 44 |
|
| 45 |
+
# Optimize model for inference
|
| 46 |
+
if device == "cuda":
|
| 47 |
+
model = model.half() # Use FP16 for faster inference
|
| 48 |
+
torch.cuda.empty_cache()
|
| 49 |
+
|
| 50 |
model.to(device)
|
| 51 |
return tokenizer, model
|
| 52 |
|
| 53 |
+
# Optimized text preprocessing
|
| 54 |
+
@st.cache_data(ttl=3600)
|
| 55 |
def preprocess_text(text):
|
| 56 |
# Add any text cleaning or normalization here
|
| 57 |
return text.strip()
|
| 58 |
|
| 59 |
+
# Batch processing for evidence extraction
|
| 60 |
+
def batch_extract_evidence(claims, contexts, model_qatc, tokenizer_qatc, batch_size=4):
|
| 61 |
+
results = []
|
| 62 |
+
for i in range(0, len(claims), batch_size):
|
| 63 |
+
batch_claims = claims[i:i + batch_size]
|
| 64 |
+
batch_contexts = contexts[i:i + batch_size]
|
| 65 |
+
|
| 66 |
+
with torch.no_grad():
|
| 67 |
+
batch_results = [
|
| 68 |
+
extract_evidence_tfidf_qatc(
|
| 69 |
+
claim, context, model_qatc, tokenizer_qatc,
|
| 70 |
+
DEVICE,
|
| 71 |
+
confidence_threshold=0.5,
|
| 72 |
+
length_ratio_threshold=0.5
|
| 73 |
+
)
|
| 74 |
+
for claim, context in zip(batch_claims, batch_contexts)
|
| 75 |
+
]
|
| 76 |
+
results.extend(batch_results)
|
| 77 |
+
return results
|
| 78 |
+
|
| 79 |
+
# Optimized verification function with parallel processing
|
| 80 |
def perform_verification(claim, context, model_qatc, tokenizer_qatc, model_tc, tokenizer_tc,
|
| 81 |
+
model_bc, tokenizer_bc, tfidf_threshold, length_ratio_threshold):
|
| 82 |
+
# Extract evidence with optimized settings
|
| 83 |
evidence_start_time = time.time()
|
| 84 |
evidence = extract_evidence_tfidf_qatc(
|
| 85 |
claim, context, model_qatc, tokenizer_qatc,
|
|
|
|
| 89 |
)
|
| 90 |
evidence_time = time.time() - evidence_start_time
|
| 91 |
|
| 92 |
+
# Clear memory after evidence extraction
|
| 93 |
+
if DEVICE == "cuda":
|
| 94 |
+
torch.cuda.empty_cache()
|
| 95 |
gc.collect()
|
| 96 |
|
| 97 |
+
# Parallel classification using ThreadPoolExecutor
|
| 98 |
+
with ThreadPoolExecutor(max_workers=2) as executor:
|
| 99 |
+
future_tc = executor.submit(cached_classify_claim, claim, evidence, tc_model_name, False)
|
| 100 |
+
future_bc = executor.submit(cached_classify_claim, claim, evidence, bc_model_name, True)
|
| 101 |
+
|
| 102 |
+
prob3class, pred_tc = future_tc.result()
|
| 103 |
+
prob2class, pred_bc = future_bc.result()
|
| 104 |
+
|
| 105 |
verdict_start_time = time.time()
|
| 106 |
with torch.no_grad():
|
| 107 |
verdict = "NEI"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 108 |
if pred_tc != 0:
|
|
|
|
|
|
|
|
|
|
| 109 |
verdict = "SUPPORTED" if pred_bc == 0 else "REFUTED" if prob2class > prob3class else ["NEI", "SUPPORTED", "REFUTED"][pred_tc]
|
| 110 |
+
|
| 111 |
verdict_time = time.time() - verdict_start_time
|
| 112 |
|
| 113 |
return {
|
|
|
|
| 121 |
"pred_bc": pred_bc
|
| 122 |
}
|
| 123 |
|
| 124 |
+
# Add performance monitoring
|
| 125 |
+
def monitor_performance():
|
| 126 |
+
if DEVICE == "cuda":
|
| 127 |
+
return {
|
| 128 |
+
"gpu_memory_used": torch.cuda.memory_allocated() / 1024**2,
|
| 129 |
+
"gpu_memory_cached": torch.cuda.memory_reserved() / 1024**2,
|
| 130 |
+
"cpu_percent": psutil.cpu_percent(),
|
| 131 |
+
"memory_percent": psutil.virtual_memory().percent
|
| 132 |
+
}
|
| 133 |
+
return {
|
| 134 |
+
"cpu_percent": psutil.cpu_percent(),
|
| 135 |
+
"memory_percent": psutil.virtual_memory().percent
|
| 136 |
+
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 137 |
|
| 138 |
# Set page configuration
|
| 139 |
st.set_page_config(
|
|
|
|
| 148 |
<style>
|
| 149 |
/* Main theme colors */
|
| 150 |
:root {
|
| 151 |
+
--primary-color: #1f77b4;
|
| 152 |
+
--secondary-color: #2c3e50;
|
| 153 |
--accent-color: #e74c3c;
|
|
|
|
|
|
|
| 154 |
--background-color: #f8f9fa;
|
| 155 |
--text-color: #2c3e50;
|
| 156 |
--border-color: #e0e0e0;
|
|
|
|
|
|
|
| 157 |
}
|
| 158 |
|
| 159 |
/* General styling */
|
| 160 |
.stApp {
|
| 161 |
background-color: var(--background-color);
|
| 162 |
color: var(--text-color);
|
|
|
|
| 163 |
}
|
| 164 |
|
| 165 |
/* Header styling */
|
| 166 |
.main-header {
|
| 167 |
+
background: linear-gradient(135deg, var(--primary-color), var(--secondary-color));
|
| 168 |
color: white;
|
| 169 |
+
padding: 2rem;
|
| 170 |
+
border-radius: 10px;
|
| 171 |
margin-bottom: 2rem;
|
| 172 |
+
box-shadow: 0 4px 6px rgba(0, 0, 0, 0.1);
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 173 |
}
|
| 174 |
|
| 175 |
.main-title {
|
| 176 |
+
font-size: 2.5rem;
|
| 177 |
+
font-weight: bold;
|
| 178 |
text-align: center;
|
| 179 |
margin-bottom: 1rem;
|
|
|
|
|
|
|
| 180 |
}
|
| 181 |
|
| 182 |
.sub-title {
|
| 183 |
+
font-size: 1.2rem;
|
| 184 |
text-align: center;
|
| 185 |
opacity: 0.9;
|
|
|
|
| 186 |
}
|
| 187 |
|
| 188 |
/* Input styling */
|
| 189 |
.stTextArea textarea {
|
| 190 |
border: 2px solid var(--border-color);
|
| 191 |
+
border-radius: 8px;
|
| 192 |
+
padding: 1rem;
|
| 193 |
+
font-size: 1rem;
|
| 194 |
min-height: 150px;
|
| 195 |
background-color: white;
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 196 |
}
|
| 197 |
|
| 198 |
/* Button styling */
|
| 199 |
.stButton>button {
|
| 200 |
+
background: linear-gradient(135deg, var(--primary-color), var(--secondary-color));
|
| 201 |
color: white;
|
| 202 |
border: none;
|
| 203 |
+
border-radius: 8px;
|
| 204 |
+
padding: 0.8rem 2rem;
|
| 205 |
+
font-size: 1.1rem;
|
| 206 |
+
font-weight: bold;
|
| 207 |
transition: all 0.3s ease;
|
|
|
|
|
|
|
| 208 |
}
|
| 209 |
|
| 210 |
.stButton>button:hover {
|
| 211 |
transform: translateY(-2px);
|
| 212 |
+
box-shadow: 0 4px 8px rgba(0, 0, 0, 0.2);
|
| 213 |
}
|
| 214 |
|
| 215 |
/* Result box styling */
|
| 216 |
.result-box {
|
| 217 |
background-color: white;
|
| 218 |
+
border-radius: 12px;
|
| 219 |
+
padding: 2rem;
|
| 220 |
+
margin: 1rem 0;
|
| 221 |
+
box-shadow: 0 4px 6px rgba(0, 0, 0, 0.1);
|
|
|
|
|
|
|
| 222 |
}
|
| 223 |
|
| 224 |
+
/* Info section styling */
|
| 225 |
+
.info-section {
|
| 226 |
+
background-color: white;
|
| 227 |
+
border-radius: 12px;
|
| 228 |
+
padding: 2rem;
|
| 229 |
+
margin: 1rem 0;
|
| 230 |
+
box-shadow: 0 4px 6px rgba(0, 0, 0, 0.1);
|
| 231 |
+
}
|
| 232 |
+
|
| 233 |
+
.info-section h3 {
|
| 234 |
+
color: var(--primary-color);
|
| 235 |
+
font-size: 1.8rem;
|
| 236 |
+
margin-bottom: 1.5rem;
|
| 237 |
+
border-bottom: 2px solid var(--border-color);
|
| 238 |
+
padding-bottom: 0.5rem;
|
| 239 |
+
}
|
| 240 |
+
|
| 241 |
+
.info-section h4 {
|
| 242 |
+
color: var(--secondary-color);
|
| 243 |
+
font-size: 1.4rem;
|
| 244 |
+
margin: 1.5rem 0 1rem 0;
|
| 245 |
+
}
|
| 246 |
+
|
| 247 |
+
.info-section p {
|
| 248 |
+
font-size: 1.1rem;
|
| 249 |
+
line-height: 1.6;
|
| 250 |
+
color: var(--text-color);
|
| 251 |
+
margin-bottom: 1.5rem;
|
| 252 |
+
}
|
| 253 |
+
|
| 254 |
+
.info-section ol, .info-section ul {
|
| 255 |
+
margin-left: 1.5rem;
|
| 256 |
+
margin-bottom: 1.5rem;
|
| 257 |
+
}
|
| 258 |
+
|
| 259 |
+
.info-section li {
|
| 260 |
+
font-size: 1.1rem;
|
| 261 |
+
line-height: 1.6;
|
| 262 |
+
margin-bottom: 0.5rem;
|
| 263 |
+
}
|
| 264 |
+
|
| 265 |
+
.info-section strong {
|
| 266 |
+
color: var(--primary-color);
|
| 267 |
}
|
| 268 |
|
| 269 |
.verdict {
|
| 270 |
+
font-size: 1.8rem;
|
| 271 |
+
font-weight: bold;
|
| 272 |
+
padding: 1rem;
|
| 273 |
+
border-radius: 8px;
|
| 274 |
+
margin: 1rem 0;
|
| 275 |
text-align: center;
|
|
|
|
| 276 |
}
|
| 277 |
|
| 278 |
.verdict-supported {
|
| 279 |
+
background-color: #d4edda;
|
| 280 |
+
color: #155724;
|
|
|
|
| 281 |
}
|
| 282 |
|
| 283 |
.verdict-refuted {
|
| 284 |
+
background-color: #f8d7da;
|
| 285 |
+
color: #721c24;
|
|
|
|
| 286 |
}
|
| 287 |
|
| 288 |
.verdict-nei {
|
| 289 |
+
background-color: #fff3cd;
|
| 290 |
+
color: #856404;
|
|
|
|
| 291 |
}
|
| 292 |
|
| 293 |
/* Sidebar styling */
|
| 294 |
.css-1d391kg {
|
| 295 |
background-color: white;
|
| 296 |
+
padding: 2rem;
|
| 297 |
+
border-radius: 12px;
|
| 298 |
+
box-shadow: 0 4px 6px rgba(0, 0, 0, 0.1);
|
| 299 |
}
|
| 300 |
|
| 301 |
/* Stats box styling */
|
| 302 |
.stats-box {
|
| 303 |
+
background-color: white;
|
| 304 |
+
border-radius: 8px;
|
| 305 |
+
padding: 1rem;
|
| 306 |
+
margin: 0.5rem 0;
|
| 307 |
+
box-shadow: 0 2px 4px rgba(0, 0, 0, 0.05);
|
| 308 |
}
|
| 309 |
|
| 310 |
/* Code block styling */
|
| 311 |
.code-block {
|
| 312 |
+
background-color: #f8f9fa;
|
| 313 |
+
border: 1px solid var(--border-color);
|
| 314 |
+
border-radius: 8px;
|
| 315 |
+
padding: 1rem;
|
| 316 |
+
font-family: monospace;
|
| 317 |
+
margin: 1rem 0;
|
|
|
|
| 318 |
}
|
| 319 |
|
| 320 |
/* Tab styling */
|
| 321 |
.stTabs [data-baseweb="tab-list"] {
|
| 322 |
gap: 2rem;
|
|
|
|
| 323 |
}
|
| 324 |
|
| 325 |
.stTabs [data-baseweb="tab"] {
|
| 326 |
background-color: white;
|
| 327 |
+
border-radius: 8px;
|
| 328 |
+
padding: 0.5rem 1rem;
|
| 329 |
margin: 0 0.5rem;
|
|
|
|
|
|
|
|
|
|
| 330 |
}
|
| 331 |
|
| 332 |
.stTabs [aria-selected="true"] {
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 333 |
background-color: var(--primary-color);
|
| 334 |
color: white;
|
| 335 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 336 |
</style>
|
| 337 |
""", unsafe_allow_html=True)
|
| 338 |
|
| 339 |
+
# Main header
|
| 340 |
st.markdown("""
|
| 341 |
<div class="main-header">
|
| 342 |
<div class="main-title">SemViQA</div>
|
|
|
|
| 344 |
</div>
|
| 345 |
""", unsafe_allow_html=True)
|
| 346 |
|
| 347 |
+
# Sidebar
|
| 348 |
with st.sidebar:
|
| 349 |
+
st.markdown("### ⚙️ Cài đặt Hệ thống")
|
|
|
|
|
|
|
|
|
|
|
|
|
| 350 |
|
| 351 |
+
# Model selection
|
| 352 |
st.markdown("#### 🧠 Chọn Mô hình")
|
| 353 |
qatc_model_name = st.selectbox(
|
| 354 |
"Mô hình QATC",
|
|
|
|
| 384 |
]
|
| 385 |
)
|
| 386 |
|
| 387 |
+
# Threshold settings
|
| 388 |
st.markdown("#### ⚖️ Ngưỡng Phân tích")
|
| 389 |
tfidf_threshold = st.slider(
|
| 390 |
"Ngưỡng TF-IDF",
|
|
|
|
| 398 |
help="Điều chỉnh độ dài tối đa của bằng chứng"
|
| 399 |
)
|
| 400 |
|
| 401 |
+
# Display settings
|
| 402 |
st.markdown("#### 👁️ Hiển thị")
|
| 403 |
show_details = st.checkbox(
|
| 404 |
"Hiển thị Chi tiết Xác suất",
|
|
|
|
| 406 |
help="Hiển thị thông tin chi tiết về xác suất dự đoán"
|
| 407 |
)
|
| 408 |
|
| 409 |
+
# Performance settings
|
| 410 |
st.markdown("#### ⚡ Hiệu suất")
|
| 411 |
num_threads = st.slider(
|
| 412 |
"Số luồng CPU",
|
|
|
|
| 418 |
os.environ["MKL_NUM_THREADS"] = str(num_threads)
|
| 419 |
|
| 420 |
# Main content
|
| 421 |
+
tabs = st.tabs(["🔍 Kiểm chứng", "📊 Lịch sử", "ℹ️ Thông tin"])
|
| 422 |
|
| 423 |
tokenizer_qatc, model_qatc = load_model(qatc_model_name, QATCForQuestionAnswering, device=DEVICE)
|
| 424 |
tokenizer_bc, model_bc = load_model(bc_model_name, ClaimModelForClassification, is_bc=True, device=DEVICE)
|
|
|
|
| 459 |
|
| 460 |
# Clear memory and perform verification
|
| 461 |
gc.collect()
|
| 462 |
+
if DEVICE == "cuda":
|
| 463 |
+
torch.cuda.empty_cache()
|
| 464 |
+
|
| 465 |
start_time = time.time()
|
| 466 |
|
| 467 |
+
# Monitor initial performance
|
| 468 |
+
initial_perf = monitor_performance()
|
| 469 |
+
|
| 470 |
result = perform_verification(
|
| 471 |
preprocessed_claim, preprocessed_context,
|
| 472 |
model_qatc, tokenizer_qatc,
|
|
|
|
| 477 |
|
| 478 |
total_time = time.time() - start_time
|
| 479 |
|
| 480 |
+
# Monitor final performance
|
| 481 |
+
final_perf = monitor_performance()
|
| 482 |
+
|
| 483 |
# Format details
|
| 484 |
details = ""
|
| 485 |
if show_details:
|
|
|
|
| 488 |
3-Class Predicted Label: {['NEI', 'SUPPORTED', 'REFUTED'][result['pred_tc']]}
|
| 489 |
2-Class Probability: {result['prob2class']:.2f}
|
| 490 |
2-Class Predicted Label: {['SUPPORTED', 'REFUTED'][result['pred_bc']] if isinstance(result['pred_bc'], int) and result['pred_tc'] != 0 else 'Not used'}
|
| 491 |
+
|
| 492 |
+
Performance Metrics:
|
| 493 |
+
- GPU Memory Used: {final_perf.get('gpu_memory_used', 'N/A'):.2f} MB
|
| 494 |
+
- GPU Memory Cached: {final_perf.get('gpu_memory_cached', 'N/A'):.2f} MB
|
| 495 |
+
- CPU Usage: {final_perf['cpu_percent']}%
|
| 496 |
+
- Memory Usage: {final_perf['memory_percent']}%
|
| 497 |
"""
|
| 498 |
|
| 499 |
+
# Store result with performance metrics
|
| 500 |
st.session_state.latest_result = {
|
| 501 |
"claim": claim,
|
| 502 |
"evidence": result['evidence'],
|
|
|
|
| 507 |
"details": details,
|
| 508 |
"qatc_model": qatc_model_name,
|
| 509 |
"bc_model": bc_model_name,
|
| 510 |
+
"tc_model": tc_model_name,
|
| 511 |
+
"performance_metrics": final_perf
|
| 512 |
}
|
| 513 |
|
| 514 |
# Add to history
|
|
|
|
| 516 |
st.session_state.history = []
|
| 517 |
st.session_state.history.append(st.session_state.latest_result)
|
| 518 |
|
| 519 |
+
# Display result with performance metrics
|
| 520 |
res = st.session_state.latest_result
|
| 521 |
verdict_class = {
|
| 522 |
"SUPPORTED": "verdict-supported",
|
|
|
|
| 536 |
<p><strong>Thời gian trích xuất bằng chứng:</strong> {res['evidence_time']:.2f} giây</p>
|
| 537 |
<p><strong>Thời gian phân loại:</strong> {res['verdict_time']:.2f} giây</p>
|
| 538 |
<p><strong>Tổng thời gian:</strong> {res['total_time']:.2f} giây</p>
|
| 539 |
+
<p><strong>Hiệu suất:</strong></p>
|
| 540 |
+
<ul>
|
| 541 |
+
<li>CPU: {res['performance_metrics']['cpu_percent']}%</li>
|
| 542 |
+
<li>RAM: {res['performance_metrics']['memory_percent']}%</li>
|
| 543 |
+
{f"<li>GPU Memory: {res['performance_metrics'].get('gpu_memory_used', 'N/A'):.2f} MB</li>" if DEVICE == "cuda" else ""}
|
| 544 |
+
</ul>
|
| 545 |
</div>
|
| 546 |
{f"<div class='code-block'><pre>{res['details']}</pre></div>" if show_details else ""}
|
| 547 |
</div>
|
| 548 |
""", unsafe_allow_html=True)
|
| 549 |
|
| 550 |
+
# Download button with performance metrics
|
| 551 |
+
result_text = f"""
|
| 552 |
+
Câu khẳng định: {res['claim']}
|
| 553 |
+
Bằng chứng: {res['evidence']}
|
| 554 |
+
Kết luận: {res['verdict']}
|
| 555 |
+
Chi tiết: {res['details']}
|
| 556 |
+
|
| 557 |
+
Hiệu suất:
|
| 558 |
+
- CPU: {res['performance_metrics']['cpu_percent']}%
|
| 559 |
+
- RAM: {res['performance_metrics']['memory_percent']}%
|
| 560 |
+
{f"- GPU Memory: {res['performance_metrics'].get('gpu_memory_used', 'N/A'):.2f} MB" if DEVICE == "cuda" else ""}
|
| 561 |
+
"""
|
| 562 |
st.download_button(
|
| 563 |
"📥 Tải kết quả",
|
| 564 |
data=result_text,
|
|
|
|
| 571 |
# --- Tab History ---
|
| 572 |
with tabs[1]:
|
| 573 |
st.markdown("### 📊 Lịch sử Kiểm chứng")
|
|
|
|
|
|
|
|
|
|
|
|
|
| 574 |
if 'history' in st.session_state and st.session_state.history:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 575 |
# Download full history
|
| 576 |
+
history_df = pd.DataFrame(st.session_state.history)
|
| 577 |
st.download_button(
|
| 578 |
"📥 Tải toàn bộ lịch sử",
|
| 579 |
data=history_df.to_csv(index=False).encode('utf-8'),
|
|
|
|
| 581 |
mime="text/csv"
|
| 582 |
)
|
| 583 |
|
| 584 |
+
# Display history
|
| 585 |
+
for idx, record in enumerate(reversed(st.session_state.history), 1):
|
| 586 |
+
st.markdown(f"""
|
| 587 |
+
<div class="result-box">
|
| 588 |
+
<h4>Kiểm chứng #{idx}</h4>
|
| 589 |
+
<p><strong>Câu khẳng định:</strong> {record['claim']}</p>
|
| 590 |
+
<p><strong>Kết luận:</strong> {verdict_icons.get(record['verdict'], '')} {record['verdict']}</p>
|
| 591 |
+
<p><strong>Thời gian:</strong> {record['total_time']:.2f} giây</p>
|
| 592 |
+
</div>
|
| 593 |
+
""", unsafe_allow_html=True)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 594 |
else:
|
| 595 |
+
st.info("Chưa có lịch sử kiểm chứng.")
|
| 596 |
|
| 597 |
# --- Tab Info ---
|
| 598 |
+
with tabs[2]:
|
| 599 |
st.markdown("""
|
| 600 |
+
<div class="info-section">
|
| 601 |
<h3>ℹ️ Thông tin về SemViQA</h3>
|
| 602 |
<p>SemViQA là hệ thống kiểm chứng thông tin tự động cho tiếng Việt, được phát triển bởi nhóm nghiên cứu tại Đại học Công nghệ Thông tin - Đại học Quốc gia TP.HCM.</p>
|
| 603 |
|
|
|
|
| 622 |
<li><strong>REFUTED:</strong> Câu khẳng định bị bác bỏ bởi bằng chứng</li>
|
| 623 |
<li><strong>NEI:</strong> Không đủ bằng chứng để kết luận</li>
|
| 624 |
</ul>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 625 |
</div>
|
| 626 |
""", unsafe_allow_html=True)
|