-
Notifications
You must be signed in to change notification settings - Fork 20
Expand file tree
/
Copy pathindex.html
More file actions
512 lines (484 loc) · 26.5 KB
/
Copy pathindex.html
File metadata and controls
512 lines (484 loc) · 26.5 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
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
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>TinyRouter — Evolved LLM Coordinator</title>
<meta name="description" content="Routing accuracy competition on Gittensor. Train a coordinator head, beat the current best, earn TAO.">
<link rel="preconnect" href="https://fonts.googleapis.com">
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
<link href="https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600;700;800;900&display=swap" rel="stylesheet">
<style>
/* ── Reset & Base ── */
*,*::before,*::after{box-sizing:border-box;margin:0;padding:0}
html{scroll-behavior:smooth;-webkit-font-smoothing:antialiased}
body{
font-family:'Inter',-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,sans-serif;
background:#08080c;color:#b4b8c5;line-height:1.65;font-size:16px;overflow-x:hidden
}
::selection{background:rgba(99,102,241,0.4);color:#fff}
h1,h2,h3,h4{color:#f1f1f6;line-height:1.2;letter-spacing:-0.02em}
a{color:inherit;text-decoration:none}
img{display:block;max-width:100%}
.container{max-width:1160px;margin:0 auto;padding:0 2rem}
/* ── Buttons ── */
.btn{
display:inline-flex;align-items:center;gap:0.5rem;padding:0.8rem 1.6rem;
font-size:0.875rem;font-weight:600;border-radius:8px;transition:all 0.25s;
cursor:pointer;border:none;white-space:nowrap;letter-spacing:-0.01em
}
.btn-primary{
background:linear-gradient(135deg,#6366f1,#8b5cf6);color:#fff;
box-shadow:0 4px 20px rgba(99,102,241,0.3)
}
.btn-primary:hover{transform:translateY(-1px);box-shadow:0 8px 30px rgba(99,102,241,0.45)}
.btn-ghost{border:1px solid rgba(255,255,255,0.12);color:#c5c8d4;background:transparent}
.btn-ghost:hover{border-color:rgba(255,255,255,0.3);background:rgba(255,255,255,0.03)}
/* ── Nav ── */
nav{
position:fixed;top:0;left:0;right:0;z-index:1000;
background:rgba(8,8,12,0.82);backdrop-filter:blur(16px);-webkit-backdrop-filter:blur(16px);
border-bottom:1px solid rgba(255,255,255,0.06);transition:all 0.3s
}
nav .container{display:flex;justify-content:space-between;align-items:center;height:60px}
nav .logo{display:flex;align-items:center;gap:0.55rem;font-weight:700;font-size:0.95rem;color:#f1f1f6;letter-spacing:-0.01em}
nav .logo .icon{width:28px;height:28px;border-radius:7px;background:linear-gradient(135deg,#6366f1,#8b5cf6);display:flex;align-items:center;justify-content:center;font-size:0.8rem}
nav .links{display:flex;gap:2rem;list-style:none;align-items:center}
nav .links a{font-size:0.8rem;font-weight:500;color:#7c7f8c;transition:color 0.2s;letter-spacing:-0.01em}
nav .links a:hover{color:#e1e1ed}
@media(max-width:768px){nav .links{display:none}}
/* ── Hero ── */
.hero{
padding:10rem 0 6rem;position:relative;overflow:hidden;
background:radial-gradient(ellipse 80% 60% at 50% -10%,rgba(99,102,241,0.12),transparent 70%),
radial-gradient(ellipse 40% 50% at 80% 30%,rgba(139,92,246,0.06),transparent 70%),
#08080c
}
.hero .container{position:relative;z-index:1}
.hero .banner-img{border-radius:16px;box-shadow:0 20px 60px rgba(0,0,0,0.5),0 0 0 1px rgba(255,255,255,0.06);margin-bottom:3rem}
.hero .badge{
display:inline-flex;align-items:center;gap:0.4rem;
background:rgba(99,102,241,0.1);border:1px solid rgba(99,102,241,0.25);
border-radius:100px;padding:0.35rem 0.9rem;font-size:0.75rem;font-weight:500;
color:#a5b4fc;margin-bottom:1.5rem;letter-spacing:-0.01em
}
.hero .badge .dot{width:6px;height:6px;border-radius:50%;background:#818cf8;animation:pulse 2s infinite}
@keyframes pulse{0%,100%{opacity:1}50%{opacity:0.4}}
.hero h1{
font-size:clamp(2.5rem,5.5vw,4rem);font-weight:900;line-height:1.08;
letter-spacing:-0.03em;margin-bottom:1.25rem;color:#f8f8fd
}
.hero h1 .grad{background:linear-gradient(135deg,#818cf8,#c084fc);-webkit-background-clip:text;background-clip:text;-webkit-text-fill-color:transparent}
.hero p{
font-size:1.1rem;color:#888a9a;max-width:560px;margin-bottom:2rem;
line-height:1.7;letter-spacing:-0.01em
}
.hero .actions{display:flex;gap:0.75rem;flex-wrap:wrap}
.hero .stats{margin-top:3rem;display:flex;gap:3rem;flex-wrap:wrap}
.hero .stat .val{font-size:1.6rem;font-weight:800;color:#e8e8f2;letter-spacing:-0.02em}
.hero .stat .lbl{font-size:0.78rem;color:#6b6e7d;margin-top:0.2rem}
/* ── Sections ── */
section{padding:5rem 0}
.section-label{
font-size:0.72rem;font-weight:700;text-transform:uppercase;letter-spacing:0.08em;
color:#6366f1;margin-bottom:0.75rem
}
.section-title{font-size:2rem;font-weight:800;margin-bottom:0.75rem;letter-spacing:-0.025em}
.section-sub{color:#6b6e7d;font-size:1rem;max-width:520px;margin-bottom:3rem;line-height:1.7}
/* ── Alt section bg ── */
.section-alt{background:#0c0c14;border-top:1px solid rgba(255,255,255,0.04);border-bottom:1px solid rgba(255,255,255,0.04)}
/* ── Steps Grid ── */
.steps{display:grid;grid-template-columns:repeat(4,1fr);gap:1.5rem}
@media(max-width:900px){.steps{grid-template-columns:repeat(2,1fr)}}
@media(max-width:500px){.steps{grid-template-columns:1fr}}
.step{
background:rgba(255,255,255,0.02);border:1px solid rgba(255,255,255,0.06);
border-radius:14px;padding:1.75rem;transition:all 0.3s;position:relative;overflow:hidden
}
.step:hover{border-color:rgba(99,102,241,0.3);background:rgba(99,102,241,0.03);transform:translateY(-2px)}
.step .num{
font-size:0.7rem;font-weight:800;color:#6366f1;margin-bottom:0.75rem;
display:inline-block;background:rgba(99,102,241,0.1);padding:0.2rem 0.55rem;border-radius:6px;letter-spacing:0.04em
}
.step h3{font-size:0.95rem;margin-bottom:0.4rem;font-weight:700}
.step p{font-size:0.82rem;color:#6b6e7d;line-height:1.6}
.step code{
display:inline-block;margin-top:0.5rem;font-size:0.7rem;background:rgba(0,0,0,0.4);
color:#8b8fa0;padding:0.3rem 0.55rem;border-radius:5px;font-family:'SF Mono','Fira Code',monospace;
overflow-x:auto;max-width:100%
}
/* ── Leaderboard ── */
.lb-cards{display:grid;grid-template-columns:repeat(2,1fr);gap:1.5rem;margin-bottom:2.5rem}
@media(max-width:700px){.lb-cards{grid-template-columns:1fr}}
.lb-card{
background:rgba(255,255,255,0.015);border:1px solid rgba(255,255,255,0.06);
border-radius:16px;padding:2rem;transition:all 0.3s
}
.lb-card.king{border-color:rgba(249,226,175,0.2);background:linear-gradient(135deg,rgba(249,226,175,0.03),rgba(249,226,175,0.01))}
.lb-card .lb-header{display:flex;justify-content:space-between;align-items:center;margin-bottom:1rem}
.lb-card .lb-bench{font-size:0.7rem;font-weight:700;text-transform:uppercase;letter-spacing:0.06em;color:#6b6e7d}
.lb-card .lb-crown{font-size:1rem}
.lb-card .lb-score{font-size:3rem;font-weight:900;letter-spacing:-0.03em;background:linear-gradient(180deg,#f8f8fd 0%,#b0b3c0 100%);-webkit-background-clip:text;background-clip:text;-webkit-text-fill-color:transparent}
.lb-card.king .lb-score{background:linear-gradient(180deg,#f9e2af 0%,#fab387 100%);-webkit-background-clip:text;background-clip:text;-webkit-text-fill-color:transparent}
.lb-card .lb-miner{font-size:0.82rem;color:#6b6e7d;margin-bottom:1.25rem}
.lb-card .lb-miner strong{color:#c5c8d4}
.lb-bars{margin-top:1.25rem}
.lb-bar{display:flex;align-items:center;gap:0.75rem;margin-bottom:0.45rem}
.lb-bar .lb-bar-lbl{font-size:0.72rem;color:#6b6e7d;width:100px;flex-shrink:0;text-align:right}
.lb-bar .lb-bar-track{flex:1;height:5px;background:rgba(255,255,255,0.05);border-radius:10px;overflow:hidden}
.lb-bar .lb-bar-fill{height:100%;border-radius:10px;background:rgba(255,255,255,0.15);transition:width 0.8s ease}
.lb-bar .lb-bar-val{font-size:0.72rem;color:#8b8fa0;width:38px;flex-shrink:0;text-align:right}
/* ── Table ── */
.tbl-wrap{overflow-x:auto;border-radius:12px;border:1px solid rgba(255,255,255,0.06)}
table{width:100%;border-collapse:collapse;font-size:0.82rem}
th,td{padding:0.7rem 1rem;text-align:left;border-bottom:1px solid rgba(255,255,255,0.04)}
th{font-size:0.68rem;font-weight:700;text-transform:uppercase;letter-spacing:0.05em;color:#5c5f6e;background:rgba(255,255,255,0.015)}
tr:hover{background:rgba(255,255,255,0.015)}
.merged{color:#6ee7b7;font-weight:600}
.rejected{color:#f87171}
/* ── Pool ── */
.pool-grid{display:grid;grid-template-columns:repeat(3,1fr);gap:1rem}
@media(max-width:700px){.pool-grid{grid-template-columns:1fr}}
.pool-card{
background:rgba(255,255,255,0.015);border:1px solid rgba(255,255,255,0.06);
border-radius:14px;padding:1.5rem;display:flex;gap:1rem;align-items:flex-start;
transition:all 0.3s
}
.pool-card:hover{transform:translateY(-1px);border-color:rgba(255,255,255,0.12)}
.pool-card .pool-icon{width:44px;height:44px;border-radius:10px;display:flex;align-items:center;justify-content:center;font-size:1.2rem;flex-shrink:0}
.pool-card:nth-child(1) .pool-icon{background:rgba(99,102,241,0.15);color:#818cf8}
.pool-card:nth-child(2) .pool-icon{background:rgba(16,185,129,0.15);color:#6ee7b7}
.pool-card:nth-child(3) .pool-icon{background:rgba(244,114,182,0.15);color:#f472b6}
.pool-card h4{font-size:0.88rem;font-weight:700;margin-bottom:0.2rem}
.pool-card p{font-size:0.78rem;color:#6b6e7d;line-height:1.55}
/* ── Research Results ── */
.results-grid{display:grid;grid-template-columns:repeat(2,1fr);gap:1rem;margin-bottom:2.5rem}
@media(max-width:600px){.results-grid{grid-template-columns:1fr}}
.result-card{
background:rgba(255,255,255,0.015);border:1px solid rgba(255,255,255,0.06);
border-radius:14px;padding:1.5rem;text-align:center
}
.result-card .big{font-size:2.2rem;font-weight:900;letter-spacing:-0.025em}
.result-card .big.accent{background:linear-gradient(135deg,#6ee7b7,#34d399);-webkit-background-clip:text;background-clip:text;-webkit-text-fill-color:transparent}
.result-card .lbl{font-size:0.78rem;color:#6b6e7d;margin-top:0.3rem}
.compare-table{overflow-x:auto;margin-bottom:1.5rem}
.compare-table table{font-size:0.82rem}
.compare-table td:first-child{font-weight:600;color:#d0d3dc}
.compare-table .winner{color:#6ee7b7;font-weight:700}
.compare-table .highlight{background:rgba(99,102,241,0.06)}
/* ── FAQ ── */
.faq-grid{display:grid;grid-template-columns:repeat(2,1fr);gap:1.5rem}
@media(max-width:700px){.faq-grid{grid-template-columns:1fr}}
.faq-item{
background:rgba(255,255,255,0.015);border:1px solid rgba(255,255,255,0.05);
border-radius:12px;padding:1.5rem;transition:all 0.25s
}
.faq-item:hover{border-color:rgba(255,255,255,0.1)}
.faq-item h3{font-size:0.9rem;font-weight:700;margin-bottom:0.5rem}
.faq-item p{font-size:0.82rem;color:#6b6e7d;line-height:1.65}
/* ── CTA ── */
.cta-section{
padding:5rem 0;text-align:center;
background:radial-gradient(ellipse 60% 50% at 50% 50%,rgba(99,102,241,0.08),transparent 70%),#0c0c14;
border-top:1px solid rgba(255,255,255,0.04)
}
.cta-section h2{font-size:1.8rem;font-weight:800;margin-bottom:0.75rem}
.cta-section p{color:#6b6e7d;margin-bottom:1.5rem;font-size:0.95rem}
/* ── Footer ── */
footer{
padding:4rem 0 3rem;border-top:1px solid rgba(255,255,255,0.04);
background:#08080c;font-size:0.78rem
}
footer .footer-grid{display:grid;grid-template-columns:2fr 1fr 1fr 1fr;gap:2rem;margin-bottom:2rem}
@media(max-width:700px){footer .footer-grid{grid-template-columns:1fr 1fr}}
footer h4{font-size:0.72rem;font-weight:700;text-transform:uppercase;letter-spacing:0.06em;color:#5c5f6e;margin-bottom:1rem}
footer ul{list-style:none}
footer li{margin-bottom:0.45rem}
footer a{color:#7c7f8c;transition:color 0.2s;font-size:0.8rem}
footer a:hover{color:#c5c8d4}
footer .footer-bottom{display:flex;justify-content:space-between;align-items:center;padding-top:2rem;border-top:1px solid rgba(255,255,255,0.04);color:#5c5f6e;flex-wrap:wrap;gap:1rem}
footer .badges{display:flex;gap:0.5rem;flex-wrap:wrap}
footer .badge{
font-size:0.68rem;padding:0.25rem 0.65rem;border-radius:100px;
background:rgba(255,255,255,0.03);border:1px solid rgba(255,255,255,0.06);
color:#7c7f8c
}
/* ── Misc ── */
.text-center{text-align:center}
.muted-note{font-size:0.75rem;color:#5c5f6e;margin-top:0.75rem}
.mt-1{margin-top:1rem}.mt-2{margin-top:2rem}.mt-3{margin-top:3rem}
</style>
</head>
<body>
<!-- ── Nav ── -->
<nav>
<div class="container">
<a href="#" class="logo">
<span class="icon">⭯</span> TinyRouter
</a>
<ul class="links">
<li><a href="#how">How it works</a></li>
<li><a href="#leaderboard">Leaderboard</a></li>
<li><a href="#pool">Models</a></li>
<li><a href="#results">Results</a></li>
<li><a href="#faq">FAQ</a></li>
<li><a href="https://github.com/James-CUDA/Gittensor-TinyRouter" class="btn btn-ghost" style="padding:0.4rem 0.9rem;font-size:0.72rem;">GitHub</a></li>
</ul>
</div>
</nav>
<!-- ── Hero ── -->
<section class="hero">
<div class="container">
<img src="Gittensor-TinyRouter.png" alt="TinyRouter" class="banner-img">
<div class="badge"><span class="dot"></span> Live on Gittensor — Subnet 74</div>
<h1>Train a <span class="grad">routing brain.</span><br>Earn TAO.</h1>
<p>TinyRouter is a routing accuracy competition. You train a small coordinator head that decides which LLM to call for each question. Beat the current best accuracy — your PR gets merged and you earn TAO through Gittensor.</p>
<div class="actions">
<a href="#how" class="btn btn-primary"><span>▶</span> Start Competing</a>
<a href="https://github.com/James-CUDA/Gittensor-TinyRouter" class="btn btn-ghost">View on GitHub</a>
<a href="SUBMITTING.md" class="btn btn-ghost">Submission Guide</a>
</div>
<div class="stats">
<div class="stat"><div class="val">0.858</div><div class="lbl">Best Average Score</div></div>
<div class="stat"><div class="val">13,312</div><div class="lbl">Trainable Parameters</div></div>
<div class="stat"><div class="val">$25–65</div><div class="lbl">Cost per Training Run</div></div>
<div class="stat"><div class="val">3</div><div class="lbl">Models in Pool</div></div>
</div>
</div>
</section>
<!-- ── How It Works ── -->
<section id="how">
<div class="container">
<div class="section-label">The Competition</div>
<h2 class="section-title">Train → Pack → Submit → Earn</h2>
<p class="section-sub">Four steps. One winner per benchmark. Only accuracy-improving heads get merged and earn TAO.</p>
<div class="steps">
<div class="step">
<div class="num">STEP 1</div>
<h3>Train a routing head</h3>
<p>Run CMA-ES on any GPU to evolve a coordinator head. The optimizer breeds thousands of candidates and keeps the best.</p>
<code>$ python -m trinity.train --benchmark math500</code>
</div>
<div class="step">
<div class="num">STEP 2</div>
<h3>Pack your submission</h3>
<p>Extract head weights and SVF scales from your trained model. A training receipt is auto-generated for verification.</p>
<code>$ python scripts/pack_submission.py --run-dir ...</code>
</div>
<div class="step">
<div class="num">STEP 3</div>
<h3>Open a pull request</h3>
<p>Add your files to <code>submissions/your-name/N/</code> and open a PR. One submission per benchmark per week.</p>
</div>
<div class="step">
<div class="num">STEP 4</div>
<h3>Win & earn TAO</h3>
<p>The maintainer runs <code>pr_eval.py</code> against an encrypted hidden benchmark. If you beat the king, your PR is merged.</p>
</div>
</div>
</div>
</section>
<!-- ── Leaderboard ── -->
<section id="leaderboard" class="section-alt">
<div class="container">
<div class="section-label">Live Data</div>
<h2 class="section-title">King of the Hill</h2>
<p class="section-sub">Beat the current best score to claim the crown. Evaluated on an encrypted hidden benchmark — never revealed to miners.</p>
<div class="lb-cards" id="lb-kings"></div>
<div class="tbl-wrap">
<table>
<thead><tr><th>Miner</th><th>Gen</th><th>Benchmark</th><th>Score</th><th>PR</th><th>Status</th></tr></thead>
<tbody id="lb-history"></tbody>
</table>
</div>
<p class="muted-note">Scores from encrypted hidden benchmark · 4 pre-eval gates + 1 post-eval run before every evaluation · <a href="leaderboard.json" style="color:#818cf8">Raw JSON</a></p>
</div>
</section>
<!-- ── Model Pool ── -->
<section id="pool">
<div class="container">
<div class="section-label">The Arena</div>
<h2 class="section-title">One pool. Three models. Fair fight.</h2>
<p class="section-sub">All miners route to the same three OpenRouter models. Routing skill is what matters — not who has the biggest GPU.</p>
<div class="pool-grid">
<div class="pool-card">
<div class="pool-icon">🧠</div>
<div>
<h4>qwen3.5-35b-a3b</h4>
<p>Knowledge specialist. Dominates MMLU at 92.2%. The go-to model for factual and reasoning tasks.</p>
</div>
</div>
<div class="pool-card">
<div class="pool-icon">∑</div>
<div>
<h4>gemini-3.1-flash-lite</h4>
<p>Math specialist. Best math500 at 79.4%. Excels at quantitative problems and symbolic reasoning.</p>
</div>
</div>
<div class="pool-card">
<div class="pool-icon">🌐</div>
<div>
<h4>deepseek-v4-flash</h4>
<p>General purpose. Broad capability across diverse question types and domains.</p>
</div>
</div>
</div>
<p class="muted-note text-center mt-2">All models served via OpenRouter API. Miners only need a T4 GPU ($0.20/hr) to run the 0.6B encoder. The big models run in the cloud.</p>
</div>
</section>
<!-- ── Results ── -->
<section id="results" class="section-alt">
<div class="container">
<div class="section-label">Proven Performance</div>
<h2 class="section-title">Routing beats every single model</h2>
<p class="section-sub">Rigorous eval on 120 held-out questions per benchmark. Single-model baselines averaged over 3 runs. TRINITY method (Xu et al., ICLR 2026).</p>
<div class="results-grid">
<div class="result-card">
<div class="big accent">0.858</div>
<div class="lbl">TinyRouter Average (math + MMLU)</div>
</div>
<div class="result-card">
<div class="big">0.835</div>
<div class="lbl">Best Single Model (qwen3.5-35b-a3b)</div>
</div>
<div class="result-card">
<div class="big">0.833</div>
<div class="lbl">Random Routing Baseline</div>
</div>
<div class="result-card">
<div class="big">~$62</div>
<div class="lbl">Total API Spend (all experiments)</div>
</div>
</div>
<div class="compare-table">
<table>
<thead><tr><th>System</th><th>Math</th><th>MMLU</th><th>Average</th></tr></thead>
<tbody>
<tr class="highlight"><td><strong>TinyRouter</strong></td><td>0.792</td><td class="winner">0.925</td><td class="winner"><strong>0.858</strong></td></tr>
<tr><td>qwen3.5-35b-a3b</td><td>0.747</td><td>0.922</td><td>0.835</td></tr>
<tr><td>random routing</td><td>0.792</td><td>0.875</td><td>0.833</td></tr>
<tr><td>gemini-3.1-flash-lite</td><td>0.794</td><td>0.783</td><td>0.789</td></tr>
<tr><td>deepseek-v4-flash</td><td>0.742</td><td>0.539</td><td>0.640</td></tr>
</tbody>
</table>
</div>
<p style="font-size:0.85rem;color:#6b6e7d;line-height:1.7">The win is <strong style="color:#c5c8d4">across tasks, not within a task</strong>. No single model is good at both: deepseek dominates knowledge, glm dominates math. The router wins the average by sending each question to the right specialist. On math500, the oracle ceiling diagnostic found <strong style="color:#c5c8d4">4.9 points of real headroom</strong> still on the table — that's the gap miners compete to close.</p>
</div>
</section>
<!-- ── FAQ ── -->
<section id="faq">
<div class="container">
<div class="section-label">Questions</div>
<h2 class="section-title">Everything you need to know</h2>
<div class="faq-grid">
<div class="faq-item">
<h3>What does it cost to compete?</h3>
<p>~$25–65 in OpenRouter API credits per training run, plus a GPU (T4 at ~$0.20/hr) or CPU (free, slower). The encoder is only 0.6B parameters — it runs on anything.</p>
</div>
<div class="faq-item">
<h3>How is my submission evaluated?</h3>
<p>Your head is tested against 200 encrypted hidden questions with pre-cached model answers (70% weight), plus 20 live multi-turn questions with real API calls (15%), plus efficiency and novelty scores (15%).</p>
</div>
<div class="faq-item">
<h3>Can I see the hidden benchmark?</h3>
<p>No. The benchmark is AES-256-GCM encrypted and stored outside the repo. It is never revealed. This prevents overfitting and keeps the competition fair for everyone.</p>
</div>
<div class="faq-item">
<h3>What happens if I lose?</h3>
<p>Your PR is closed with your composite score and the current best shown. Component breakdowns are <em>not</em> revealed — this prevents attackers from probing the benchmark through repeated submissions.</p>
</div>
<div class="faq-item">
<h3>How do you prevent cheating?</h3>
<p>Five automated anti-cheat gates: rate limiting, NaN/Inf detection, cosine-similarity duplicate detection against all previous heads, receipt cross-validation (cost ≥ $15, plausible fitness curve), overfit rejection, and minimized score feedback.</p>
</div>
<div class="faq-item">
<h3>How does Gittensor pay me?</h3>
<p>When your routing head PR is merged, Gittensor validators (Bittensor Subnet 74) detect the merged PR and allocate TAO rewards. You earn TAO for genuinely improving routing accuracy.</p>
</div>
</div>
</div>
</section>
<!-- ── CTA ── -->
<section class="cta-section">
<div class="container">
<h2>Ready to compete?</h2>
<p>Train a head, beat the king, earn TAO. The math500 headroom is waiting.</p>
<div class="actions" style="justify-content:center">
<a href="SUBMITTING.md" class="btn btn-primary"><span>▶</span> Read the Submission Guide</a>
<a href="https://github.com/James-CUDA/Gittensor-TinyRouter" class="btn btn-ghost">Fork on GitHub</a>
</div>
</div>
</section>
<!-- ── Footer ── -->
<footer>
<div class="container">
<div class="footer-grid">
<div>
<h4>TinyRouter</h4>
<p style="color:#6b6e7d;font-size:0.8rem;line-height:1.65">An open-source replication of TRINITY: An Evolved LLM Coordinator (Xu et al., ICLR 2026). Routing accuracy competition on Gittensor.</p>
</div>
<div>
<h4>Compete</h4>
<ul>
<li><a href="SUBMITTING.md">Submission Guide</a></li>
<li><a href="CONTRIBUTING.md">Rules</a></li>
<li><a href="leaderboard.json">Leaderboard JSON</a></li>
</ul>
</div>
<div>
<h4>Project</h4>
<ul>
<li><a href="https://github.com/James-CUDA/Gittensor-TinyRouter">GitHub</a></li>
<li><a href="ROADMAP.md">Roadmap</a></li>
<li><a href="docs/JOURNAL.md">Research Journal</a></li>
</ul>
</div>
<div>
<h4>Research</h4>
<ul>
<li><a href="https://arxiv.org/abs/2512.04695">TRINITY Paper</a></li>
<li><a href="docs/SPEC.md">Implementation Spec</a></li>
<li><a href="docs/RESULTS.md">Full Results</a></li>
</ul>
</div>
</div>
<div class="footer-bottom">
<span>© 2026 TinyRouter. TRINITY replication. MIT License.</span>
<div class="badges">
<span class="badge">Gittensor SN74</span>
<span class="badge">ICLR 2026</span>
<span class="badge">13,312 trainable params</span>
</div>
</div>
</div>
</section>
<script>
(async function(){
try{
const r=await fetch('leaderboard.json'),d=await r.json();
const benches=d.benchmarks||{},kings=document.getElementById('lb-kings'),hist=document.getElementById('lb-history');
let all=[];
for(const[n,b]of Object.entries(benches)){
const best=b.best_score||0,isKing=b.best_miner&&b.best_score>(b.best_single_model||0);
const bars=[['Best single',b.best_single_model],['Random',b.baseline_random],['Oracle',b.oracle_ceiling]];
const maxBar=Math.max(...bars.map(x=>x[1]),0.01);
kings.innerHTML+=`
<div class="lb-card${isKing?' king':''}">
<div class="lb-header"><span class="lb-bench">${n}</span>${isKing?'<span class="lb-crown">👑</span>':''}</div>
<div class="lb-score">${(best*100).toFixed(1)}%</div>
<div class="lb-miner">${b.best_miner?`King: <strong>${b.best_miner}</strong> gen ${b.best_generation}`:'<em>No submissions yet</em>'}</div>
<div class="lb-bars">${bars.map(x=>`
<div class="lb-bar"><span class="lb-bar-lbl">${x[0]}</span><div class="lb-bar-track"><div class="lb-bar-fill" style="width:${(x[1]/maxBar*100).toFixed(0)}%"></div></div><span class="lb-bar-val">${(x[1]*100).toFixed(1)}%</span></div>
`).join('')}</div>
</div>`;
for(const h of(b.history||[]))all.push({...h,benchmark:n});
}
all.sort((a,b)=>(b.timestamp||'').localeCompare(a.timestamp||''));
hist.innerHTML=all.length?all.map(h=>`
<tr><td><strong>${h.miner}</strong></td><td>${h.generation}</td><td>${h.benchmark}</td><td>${(h.score*100).toFixed(1)}%</td><td>${h.pr?'#'+h.pr:'—'}</td><td class="${h.merged?'merged':'rejected'}">${h.merged?'MERGED':'rejected'}</td></tr>
`).join(''):'<tr><td colspan="6" style="text-align:center;color:#5c5f6e;padding:2rem;">No submissions yet — train a head and open a PR to be the first</td></tr>';
}catch(e){document.getElementById('lb-kings').innerHTML='<p style="color:#5c5f6e">Leaderboard data loading…</p>'}
})();
</script>
</body>
</html>