Update app.py
Browse files
app.py
CHANGED
|
@@ -138,12 +138,12 @@ def build_trace(problem, displayed, first_fail, done=False):
|
|
| 138 |
lbl, conf = probe_step(problem, displayed[:i+1])
|
| 139 |
if lbl == "FAIL": faults += 1
|
| 140 |
dot = f"<span class='dot dot-{'ok' if lbl == 'OK' else 'fail'}'></span>"
|
| 141 |
-
rows += f"<div class='trace-row'>{dot} Step {i+1} — {int(conf*100)}%</div>"
|
| 142 |
else:
|
| 143 |
rows += "<div class='trace-row'><span class='dot dot-final'></span> Final Answer</div>"
|
| 144 |
|
| 145 |
health = max(0, 100 - faults * 20)
|
| 146 |
-
bar_color = "#
|
| 147 |
summary = f"""
|
| 148 |
<div class='trace-summary'>
|
| 149 |
<div class='trace-label'>Reasoning Health</div>
|
|
@@ -153,50 +153,245 @@ def build_trace(problem, displayed, first_fail, done=False):
|
|
| 153 |
return f"{rows}{summary}"
|
| 154 |
|
| 155 |
|
| 156 |
-
# ββ CSS ββββββββββββββββββββββββββββββββββ
|
| 157 |
CSS = """
|
| 158 |
-
@import url('https://fonts.googleapis.com/css2?family=Inter:wght@300;400;500;600;700&family=JetBrains+Mono:wght@400;500&display=swap');
|
|
|
|
|
|
|
| 159 |
*, *::before, *::after { box-sizing: border-box; }
|
| 160 |
-
body, .gradio-container, .gradio-container * {
|
| 161 |
-
|
| 162 |
-
|
| 163 |
-
.
|
| 164 |
-
|
| 165 |
-
|
| 166 |
-
|
| 167 |
-
|
| 168 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 169 |
textarea::placeholder { color: #94a3b8 !important; }
|
| 170 |
-
|
| 171 |
-
|
| 172 |
-
.
|
| 173 |
-
|
| 174 |
-
|
| 175 |
-
|
| 176 |
-
|
| 177 |
-
|
| 178 |
-
|
| 179 |
-
|
| 180 |
-
|
| 181 |
-
|
| 182 |
-
|
| 183 |
-
|
| 184 |
-
|
| 185 |
-
|
| 186 |
-
.
|
| 187 |
-
|
| 188 |
-
|
| 189 |
-
|
| 190 |
-
.
|
| 191 |
-
|
| 192 |
-
|
| 193 |
-
|
| 194 |
-
|
| 195 |
-
.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 196 |
.gradio-container .prose { color: #1e293b !important; }
|
| 197 |
-
|
| 198 |
"""
|
| 199 |
|
|
|
|
| 200 |
HEADER = """
|
| 201 |
<div class='twt-header'>
|
| 202 |
<div class='twt-title'>Think<span>While</span>Thinking</div>
|
|
@@ -216,17 +411,25 @@ with gr.Blocks(css=CSS, title="ThinkWhileThinking") as demo:
|
|
| 216 |
with gr.Row():
|
| 217 |
with gr.Column(scale=2):
|
| 218 |
gr.HTML("<div class='input-label'>Problem</div>")
|
| 219 |
-
problem_input = gr.Textbox(
|
|
|
|
|
|
|
|
|
|
|
|
|
| 220 |
run_btn = gr.Button("βΆ Analyze Reasoning", variant="primary")
|
| 221 |
gr.Examples(examples=EXAMPLES, inputs=problem_input, label="Try an example")
|
| 222 |
with gr.Column(scale=3):
|
| 223 |
with gr.Row():
|
| 224 |
with gr.Column(scale=3):
|
| 225 |
gr.HTML("<div class='panel-label'>Chain of Thought Β· Step Scores</div>")
|
| 226 |
-
cot_output = gr.HTML(
|
|
|
|
|
|
|
| 227 |
with gr.Column(scale=1):
|
| 228 |
gr.HTML("<div class='panel-label'>Step Trace</div>")
|
| 229 |
-
trace_output = gr.HTML(
|
|
|
|
|
|
|
| 230 |
raw_output = gr.Textbox(visible=False)
|
| 231 |
run_btn.click(fn=run_twt, inputs=[problem_input], outputs=[raw_output, cot_output, trace_output])
|
| 232 |
problem_input.submit(fn=run_twt, inputs=[problem_input], outputs=[raw_output, cot_output, trace_output])
|
|
|
|
| 138 |
lbl, conf = probe_step(problem, displayed[:i+1])
|
| 139 |
if lbl == "FAIL": faults += 1
|
| 140 |
dot = f"<span class='dot dot-{'ok' if lbl == 'OK' else 'fail'}'></span>"
|
| 141 |
+
rows += f"<div class='trace-row'>{dot} Step {i+1} <span class='trace-conf'>— {int(conf*100)}%</span></div>"
|
| 142 |
else:
|
| 143 |
rows += "<div class='trace-row'><span class='dot dot-final'></span> Final Answer</div>"
|
| 144 |
|
| 145 |
health = max(0, 100 - faults * 20)
|
| 146 |
+
bar_color = "#10b981" if health > 60 else "#ef4444"
|
| 147 |
summary = f"""
|
| 148 |
<div class='trace-summary'>
|
| 149 |
<div class='trace-label'>Reasoning Health</div>
|
|
|
|
| 153 |
return f"{rows}{summary}"
|
| 154 |
|
| 155 |
|
| 156 |
+
# ββ CSS (SaaS / Modern Professional Upgrade) ββββββββββββββββββββββββββββββββββ
|
| 157 |
CSS = """
|
| 158 |
+
@import url('https://fonts.googleapis.com/css2?family=Inter:wght@300;400;500;600;700&family=JetBrains+Mono:wght@400;500;600&display=swap');
|
| 159 |
+
|
| 160 |
+
/* Globals */
|
| 161 |
*, *::before, *::after { box-sizing: border-box; }
|
| 162 |
+
body, .gradio-container, .gradio-container * {
|
| 163 |
+
font-family: 'Inter', -apple-system, sans-serif !important;
|
| 164 |
+
}
|
| 165 |
+
.gradio-container {
|
| 166 |
+
background-color: #f8fafc !important;
|
| 167 |
+
max-width: 1280px !important;
|
| 168 |
+
}
|
| 169 |
+
|
| 170 |
+
/* Header */
|
| 171 |
+
.twt-header {
|
| 172 |
+
text-align: center;
|
| 173 |
+
padding: 3rem 1rem 2rem;
|
| 174 |
+
margin-bottom: 2.5rem;
|
| 175 |
+
background: transparent;
|
| 176 |
+
border-bottom: 1px solid #e2e8f0;
|
| 177 |
+
}
|
| 178 |
+
.twt-title {
|
| 179 |
+
font-size: 2.75rem;
|
| 180 |
+
font-weight: 800;
|
| 181 |
+
letter-spacing: -0.04em;
|
| 182 |
+
color: #0f172a;
|
| 183 |
+
font-family: 'JetBrains Mono', monospace !important;
|
| 184 |
+
margin-bottom: 0.5rem;
|
| 185 |
+
}
|
| 186 |
+
.twt-title span {
|
| 187 |
+
background: linear-gradient(135deg, #7c3aed 0%, #4f46e5 100%);
|
| 188 |
+
-webkit-background-clip: text;
|
| 189 |
+
-webkit-text-fill-color: transparent;
|
| 190 |
+
}
|
| 191 |
+
.twt-sub {
|
| 192 |
+
font-size: 0.85rem;
|
| 193 |
+
color: #64748b;
|
| 194 |
+
letter-spacing: 0.05em;
|
| 195 |
+
font-weight: 500;
|
| 196 |
+
}
|
| 197 |
+
|
| 198 |
+
/* Typography & Labels */
|
| 199 |
+
.input-label, .panel-label {
|
| 200 |
+
font-size: 0.75rem;
|
| 201 |
+
font-weight: 700;
|
| 202 |
+
color: #475569;
|
| 203 |
+
letter-spacing: 0.1em;
|
| 204 |
+
text-transform: uppercase;
|
| 205 |
+
margin-bottom: 0.75rem;
|
| 206 |
+
display: block;
|
| 207 |
+
}
|
| 208 |
+
.panel-label {
|
| 209 |
+
padding-bottom: 0.5rem;
|
| 210 |
+
border-bottom: 2px solid #e2e8f0;
|
| 211 |
+
}
|
| 212 |
+
|
| 213 |
+
/* Inputs & Forms */
|
| 214 |
+
textarea {
|
| 215 |
+
background: #ffffff !important;
|
| 216 |
+
border: 1px solid #cbd5e1 !important;
|
| 217 |
+
border-radius: 12px !important;
|
| 218 |
+
color: #0f172a !important;
|
| 219 |
+
font-size: 0.95rem !important;
|
| 220 |
+
line-height: 1.6 !important;
|
| 221 |
+
padding: 1rem !important;
|
| 222 |
+
box-shadow: 0 1px 2px 0 rgba(0, 0, 0, 0.05) !important;
|
| 223 |
+
transition: all 0.2s ease-in-out !important;
|
| 224 |
+
}
|
| 225 |
+
textarea:focus {
|
| 226 |
+
border-color: #7c3aed !important;
|
| 227 |
+
box-shadow: 0 0 0 4px rgba(124, 58, 237, 0.15) !important;
|
| 228 |
+
outline: none !important;
|
| 229 |
+
}
|
| 230 |
textarea::placeholder { color: #94a3b8 !important; }
|
| 231 |
+
|
| 232 |
+
/* Primary Button */
|
| 233 |
+
button.primary {
|
| 234 |
+
background: linear-gradient(135deg, #7c3aed 0%, #6d28d9 100%) !important;
|
| 235 |
+
color: #ffffff !important;
|
| 236 |
+
font-weight: 600 !important;
|
| 237 |
+
border: none !important;
|
| 238 |
+
border-radius: 10px !important;
|
| 239 |
+
padding: 0.75rem 1.5rem !important;
|
| 240 |
+
font-size: 0.95rem !important;
|
| 241 |
+
letter-spacing: 0.02em !important;
|
| 242 |
+
box-shadow: 0 4px 6px -1px rgba(124, 58, 237, 0.3), 0 2px 4px -1px rgba(124, 58, 237, 0.2) !important;
|
| 243 |
+
transition: all 0.2s ease !important;
|
| 244 |
+
cursor: pointer !important;
|
| 245 |
+
width: 100% !important;
|
| 246 |
+
}
|
| 247 |
+
button.primary:hover {
|
| 248 |
+
transform: translateY(-2px) !important;
|
| 249 |
+
box-shadow: 0 10px 15px -3px rgba(124, 58, 237, 0.3), 0 4px 6px -2px rgba(124, 58, 237, 0.2) !important;
|
| 250 |
+
}
|
| 251 |
+
button.primary:active {
|
| 252 |
+
transform: translateY(0) !important;
|
| 253 |
+
}
|
| 254 |
+
|
| 255 |
+
/* Step Cards */
|
| 256 |
+
.step-card {
|
| 257 |
+
display: flex;
|
| 258 |
+
align-items: flex-start;
|
| 259 |
+
gap: 1rem;
|
| 260 |
+
padding: 1.25rem;
|
| 261 |
+
margin-bottom: 0.75rem;
|
| 262 |
+
background: #ffffff;
|
| 263 |
+
border: 1px solid #e2e8f0;
|
| 264 |
+
border-radius: 12px;
|
| 265 |
+
font-size: 0.9rem;
|
| 266 |
+
line-height: 1.6;
|
| 267 |
+
animation: fadeUp 0.3s cubic-bezier(0.16, 1, 0.3, 1) both;
|
| 268 |
+
box-shadow: 0 1px 3px 0 rgba(0, 0, 0, 0.05), 0 1px 2px -1px rgba(0, 0, 0, 0.05);
|
| 269 |
+
transition: box-shadow 0.2s ease, border-color 0.2s ease;
|
| 270 |
+
}
|
| 271 |
+
.step-card:hover {
|
| 272 |
+
box-shadow: 0 4px 6px -1px rgba(0, 0, 0, 0.05), 0 2px 4px -2px rgba(0, 0, 0, 0.05);
|
| 273 |
+
border-color: #cbd5e1;
|
| 274 |
+
}
|
| 275 |
+
.step-card.step-fault {
|
| 276 |
+
border: 1px solid #fecaca;
|
| 277 |
+
background: #fff1f2;
|
| 278 |
+
}
|
| 279 |
+
.step-card.step-fault:hover {
|
| 280 |
+
border-color: #fca5a5;
|
| 281 |
+
}
|
| 282 |
+
|
| 283 |
+
@keyframes fadeUp {
|
| 284 |
+
from { opacity: 0; transform: translateY(12px); }
|
| 285 |
+
to { opacity: 1; transform: translateY(0); }
|
| 286 |
+
}
|
| 287 |
+
|
| 288 |
+
.step-icon {
|
| 289 |
+
font-size: 1.1rem;
|
| 290 |
+
flex-shrink: 0;
|
| 291 |
+
margin-top: 2px;
|
| 292 |
+
}
|
| 293 |
+
.step-text {
|
| 294 |
+
flex: 1;
|
| 295 |
+
color: #1e293b;
|
| 296 |
+
}
|
| 297 |
+
|
| 298 |
+
/* Badges */
|
| 299 |
+
.badge {
|
| 300 |
+
flex-shrink: 0;
|
| 301 |
+
font-size: 0.65rem;
|
| 302 |
+
font-weight: 700;
|
| 303 |
+
font-family: 'JetBrains Mono', monospace !important;
|
| 304 |
+
letter-spacing: 0.05em;
|
| 305 |
+
padding: 0.25rem 0.75rem;
|
| 306 |
+
border-radius: 999px; /* Pill shape */
|
| 307 |
+
align-self: flex-start;
|
| 308 |
+
margin-top: 2px;
|
| 309 |
+
}
|
| 310 |
+
.badge-ok { background: #ecfdf5; color: #059669; border: 1px solid #a7f3d0; }
|
| 311 |
+
.badge-fail { background: #fef2f2; color: #dc2626; border: 1px solid #fecaca; }
|
| 312 |
+
.badge-final { background: #f3e8ff; color: #7c3aed; border: 1px solid #d8b4fe; }
|
| 313 |
+
|
| 314 |
+
/* Trace Panel */
|
| 315 |
+
.trace-row {
|
| 316 |
+
display: flex;
|
| 317 |
+
align-items: center;
|
| 318 |
+
gap: 0.75rem;
|
| 319 |
+
font-size: 0.8rem;
|
| 320 |
+
color: #334155;
|
| 321 |
+
font-family: 'JetBrains Mono', monospace !important;
|
| 322 |
+
padding: 0.5rem 0;
|
| 323 |
+
border-bottom: 1px solid #f1f5f9;
|
| 324 |
+
}
|
| 325 |
+
.trace-conf {
|
| 326 |
+
color: #94a3b8;
|
| 327 |
+
font-weight: 500;
|
| 328 |
+
}
|
| 329 |
+
.dot {
|
| 330 |
+
width: 10px; height: 10px;
|
| 331 |
+
border-radius: 50%;
|
| 332 |
+
flex-shrink: 0;
|
| 333 |
+
}
|
| 334 |
+
.dot-ok { background: #10b981; box-shadow: 0 0 0 3px #d1fae5; }
|
| 335 |
+
.dot-fail { background: #ef4444; box-shadow: 0 0 0 3px #fee2e2; }
|
| 336 |
+
.dot-final { background: #8b5cf6; box-shadow: 0 0 0 3px #ede9fe; }
|
| 337 |
+
|
| 338 |
+
.trace-summary {
|
| 339 |
+
margin-top: 1.5rem;
|
| 340 |
+
padding-top: 1rem;
|
| 341 |
+
border-top: 1px solid #e2e8f0;
|
| 342 |
+
}
|
| 343 |
+
.trace-label {
|
| 344 |
+
font-size: 0.7rem;
|
| 345 |
+
font-weight: 700;
|
| 346 |
+
color: #475569;
|
| 347 |
+
text-transform: uppercase;
|
| 348 |
+
letter-spacing: 0.1em;
|
| 349 |
+
margin-bottom: 0.5rem;
|
| 350 |
+
}
|
| 351 |
+
.trace-bar-bg {
|
| 352 |
+
background: #e2e8f0;
|
| 353 |
+
border-radius: 999px;
|
| 354 |
+
height: 8px;
|
| 355 |
+
overflow: hidden;
|
| 356 |
+
}
|
| 357 |
+
.trace-bar {
|
| 358 |
+
height: 100%;
|
| 359 |
+
border-radius: 999px;
|
| 360 |
+
transition: width 0.8s cubic-bezier(0.16, 1, 0.3, 1);
|
| 361 |
+
}
|
| 362 |
+
.trace-pct {
|
| 363 |
+
font-size: 0.75rem;
|
| 364 |
+
color: #334155;
|
| 365 |
+
margin-top: 0.4rem;
|
| 366 |
+
font-family: 'JetBrains Mono', monospace !important;
|
| 367 |
+
font-weight: 600;
|
| 368 |
+
}
|
| 369 |
+
|
| 370 |
+
/* System Messages */
|
| 371 |
+
.msg {
|
| 372 |
+
padding: 1rem 1.25rem;
|
| 373 |
+
border-radius: 10px;
|
| 374 |
+
font-size: 0.85rem;
|
| 375 |
+
font-weight: 500;
|
| 376 |
+
}
|
| 377 |
+
.msg.warn { background: #fffbeb; color: #b45309; border: 1px solid #fde68a; }
|
| 378 |
+
.msg.error { background: #fef2f2; color: #dc2626; border: 1px solid #fecaca; }
|
| 379 |
+
.msg.empty {
|
| 380 |
+
color: #94a3b8;
|
| 381 |
+
font-size: 0.85rem;
|
| 382 |
+
font-family: 'JetBrains Mono', monospace !important;
|
| 383 |
+
padding: 1.5rem 1rem;
|
| 384 |
+
text-align: center;
|
| 385 |
+
border: 1px dashed #cbd5e1;
|
| 386 |
+
border-radius: 12px;
|
| 387 |
+
}
|
| 388 |
+
|
| 389 |
+
/* Gradio Overrides */
|
| 390 |
.gradio-container .prose { color: #1e293b !important; }
|
| 391 |
+
.gr-box, .gr-form { background: transparent !important; border: none !important; }
|
| 392 |
"""
|
| 393 |
|
| 394 |
+
# ββ Layout ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 395 |
HEADER = """
|
| 396 |
<div class='twt-header'>
|
| 397 |
<div class='twt-title'>Think<span>While</span>Thinking</div>
|
|
|
|
| 411 |
with gr.Row():
|
| 412 |
with gr.Column(scale=2):
|
| 413 |
gr.HTML("<div class='input-label'>Problem</div>")
|
| 414 |
+
problem_input = gr.Textbox(
|
| 415 |
+
placeholder="Enter a math, logic, or reasoning problem...",
|
| 416 |
+
lines=5,
|
| 417 |
+
show_label=False,
|
| 418 |
+
)
|
| 419 |
run_btn = gr.Button("βΆ Analyze Reasoning", variant="primary")
|
| 420 |
gr.Examples(examples=EXAMPLES, inputs=problem_input, label="Try an example")
|
| 421 |
with gr.Column(scale=3):
|
| 422 |
with gr.Row():
|
| 423 |
with gr.Column(scale=3):
|
| 424 |
gr.HTML("<div class='panel-label'>Chain of Thought Β· Step Scores</div>")
|
| 425 |
+
cot_output = gr.HTML(
|
| 426 |
+
value="<div class='msg empty'>// awaiting problem input</div>"
|
| 427 |
+
)
|
| 428 |
with gr.Column(scale=1):
|
| 429 |
gr.HTML("<div class='panel-label'>Step Trace</div>")
|
| 430 |
+
trace_output = gr.HTML(
|
| 431 |
+
value="<div class='msg empty'>// trace</div>"
|
| 432 |
+
)
|
| 433 |
raw_output = gr.Textbox(visible=False)
|
| 434 |
run_btn.click(fn=run_twt, inputs=[problem_input], outputs=[raw_output, cot_output, trace_output])
|
| 435 |
problem_input.submit(fn=run_twt, inputs=[problem_input], outputs=[raw_output, cot_output, trace_output])
|