Hugging Face's logo

Spaces:
VIDraft
/
AI
Running

App Files Community
4
AI
File size: 17,410 Bytes
e65f85b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>POCKET — a big model, no GPU required</title>
<meta name="description" content="POCKET runs a 35B-class open model on a phone, a laptop, or a GPU-less server — decoding 2.69× faster than the most-downloaded on-device model, at matched quality.">
<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=Space+Grotesk:wght@400;500;600;700&family=Inter:wght@400;500;600&family=JetBrains+Mono:wght@400;500;600&display=swap" rel="stylesheet">
<style>
/* Hallmark · genre: modern-minimal · macrostructure: stat-led · theme: cobalt · enrichment: none · nav: N5 floating-pill · footer: Ft2 inline · pre-emit critique: P5 H4 E5 S5 R5 V5 · contrast: pass (40–41) · slop: pass (42–45) */
:root{
  --paper:#F6F8FC;            /* cool engineered near-white (not #fff) */
  --paper-2:#EEF2F9;
  --panel:#FFFFFF;
  --ink:#141922;             /* cool near-black (not #000) */
  --ink-2:#3A4453;
  --muted:#5C6675;
  --line:#DCE2EC;            /* ruler hairline */
  --line-2:#C9D1DE;
  --accent:#1E4FE0;          /* electric cobalt — the single signal */
  --accent-ink:#FFFFFF;      /* text on accent (WCAG >7:1) */
  --accent-wash:#EAF0FE;
  --ok:#0E7C66;              /* muted teal for "win", not decorative */
  --space-3xs:4px; --space-2xs:8px; --space-xs:12px; --space-s:16px;
  --space-m:24px; --space-l:40px; --space-xl:64px; --space-2xl:96px; --space-3xl:140px;
  --r:6px;                   /* tight radii */
  --r-pill:100px;
  --font-display:'Space Grotesk',ui-sans-serif,system-ui,sans-serif;
  --font-body:'Inter',ui-sans-serif,system-ui,sans-serif;
  --font-mono:'JetBrains Mono',ui-monospace,SFMono-Regular,Menlo,Consolas,monospace;
}
*{box-sizing:border-box}
html,body{overflow-x:clip;margin:0}          /* gate 34 */
html{-webkit-text-size-adjust:100%}
body{
  background:var(--paper);color:var(--ink);
  font-family:var(--font-body);font-weight:400;line-height:1.55;
  font-size:16px;letter-spacing:-0.006em;
  -webkit-font-smoothing:antialiased;
}
.wrap{max-width:1120px;margin-inline:auto;padding-inline:var(--space-m)}
h1,h2,h3{font-family:var(--font-display);font-weight:600;font-style:normal;letter-spacing:-0.03em;line-height:1.04;margin:0}
p{margin:0}
a{color:inherit}
.tnum{font-variant-numeric:tabular-nums;font-feature-settings:"tnum" 1}
.mono{font-family:var(--font-mono)}
.eyebrow{font-family:var(--font-mono);font-size:.72rem;font-weight:500;letter-spacing:.14em;text-transform:uppercase;color:var(--muted)}

/* ---- N5 floating-pill nav (content-sized, detached, bordered) ---- */
.nav-shell{position:sticky;top:0;z-index:50;padding-top:var(--space-s)}
.nav{
  max-width:1120px;margin-inline:auto;
  display:flex;align-items:center;gap:var(--space-m);
  width:max-content;max-width:calc(100% - 2*var(--space-m));
  margin-inline:auto;
  padding:9px 10px 9px 18px;
  background:rgba(255,255,255,.82);backdrop-filter:blur(10px);-webkit-backdrop-filter:blur(10px);
  border:1px solid var(--line-2);border-radius:var(--r-pill);
  box-shadow:0 1px 2px rgba(20,25,34,.04),0 8px 24px rgba(20,25,34,.06);
}
.nav .brand{font-family:var(--font-display);font-weight:700;letter-spacing:-0.02em;font-size:1.02rem}
.nav .brand .dot{color:var(--accent)}
.nav .nlinks{display:flex;gap:var(--space-s);margin-left:var(--space-xs)}
.nav .nlinks a{font-size:.85rem;color:var(--ink-2);text-decoration:none;padding:4px 2px}
.nav .nlinks a:hover{color:var(--ink)}
.nav a:focus-visible,.btn:focus-visible,.chip:focus-visible{outline:2px solid var(--accent);outline-offset:2px;border-radius:4px}
.nav .btn{margin-left:var(--space-2xs)}
@media(max-width:640px){ .nav .nlinks{display:none} }

.btn{
  display:inline-flex;align-items:center;gap:8px;line-height:1;
  font-family:var(--font-body);font-size:.86rem;font-weight:500;
  padding:10px 16px;border-radius:var(--r-pill);text-decoration:none;cursor:pointer;
  border:1px solid transparent;transition:background-color .16s ease,color .16s ease,border-color .16s ease;
}
.btn-primary{background:var(--accent);color:var(--accent-ink)}
.btn-primary:hover{background:#173FBF}
.btn-primary:active{background:#12329B}
.btn-ghost{background:transparent;color:var(--ink);border-color:var(--line-2)}
.btn-ghost:hover{border-color:var(--ink-2);background:var(--panel)}
.btn-ghost:active{background:var(--paper-2)}

/* ---- Hero · stat-led (figure + worded headline, left-biased, off-axis eyebrow) ---- */
.hero{padding-top:var(--space-2xl);padding-bottom:var(--space-3xl)}   /* bottom>top: gate 44a */
.hero-grid{display:grid;grid-template-columns:minmax(0,1.15fr) minmax(0,.85fr);gap:var(--space-xl);align-items:end}
.hero-figure{font-family:var(--font-display);font-weight:700;letter-spacing:-0.05em;line-height:.86;
  font-size:clamp(5.5rem,15vw,12.5rem);color:var(--ink)}
.hero-figure .x{color:var(--accent);font-weight:500}
.hero-eyebrow{display:flex;align-items:center;gap:10px;margin-bottom:var(--space-m)}
.hero-eyebrow .rule{height:1px;width:44px;background:var(--line-2)}
.hero h1{font-size:clamp(1.55rem,2.4vw + .6rem,2.5rem);font-weight:600;margin-top:var(--space-s)}
.hero .lede{color:var(--ink-2);font-size:1.06rem;line-height:1.55;max-width:56ch;margin-top:var(--space-s)}
.hero .cta{display:flex;flex-wrap:wrap;align-items:center;gap:var(--space-xs);margin-top:var(--space-l)}
.hero-side{padding-bottom:var(--space-2xs)}
.hero-side .qual{font-size:.98rem;color:var(--ink-2);line-height:1.5;border-left:2px solid var(--accent);padding-left:var(--space-s)}
.hero-side .foot{margin-top:var(--space-m);font-family:var(--font-mono);font-size:.72rem;color:var(--muted);letter-spacing:.01em;line-height:1.6}
@media(max-width:820px){ .hero-grid{grid-template-columns:1fr;gap:var(--space-l);align-items:start} .hero{padding-top:var(--space-xl);padding-bottom:var(--space-2xl)} }

/* ---- Supporting stat strip (hairline-separated, tabular) ---- */
.rule-top{border-top:1px solid var(--line)}
.strip{display:grid;grid-template-columns:repeat(4,1fr);gap:0}
.strip .cell{padding:var(--space-l) var(--space-m) var(--space-l) 0;border-right:1px solid var(--line)}
.strip .cell:last-child{border-right:none}
.strip .n{font-family:var(--font-display);font-weight:600;font-size:2.3rem;letter-spacing:-0.03em;line-height:1}
.strip .n.acc{color:var(--accent)}
.strip .k{font-family:var(--font-mono);font-size:.72rem;letter-spacing:.06em;text-transform:uppercase;color:var(--muted);margin-top:var(--space-2xs)}
.strip .d{font-size:.9rem;color:var(--ink-2);margin-top:6px;line-height:1.45}
@media(max-width:820px){ .strip{grid-template-columns:1fr 1fr} .strip .cell:nth-child(2n){border-right:none} }
@media(max-width:460px){ .strip{grid-template-columns:1fr} .strip .cell{border-right:none;border-bottom:1px solid var(--line);padding-right:0} .strip .cell:last-child{border-bottom:none} }

.section{padding-block:var(--space-2xl)}
.sec-head{max-width:60ch}
.sec-head .eyebrow{margin-bottom:var(--space-s)}
.sec-head h2{font-size:clamp(1.5rem,2vw + .5rem,2.15rem)}
.sec-head p{color:var(--ink-2);margin-top:var(--space-s);font-size:1.02rem}

/* ---- Scorecard (honest 4-axis, incl. the loss) ---- */
.card{background:var(--panel);border:1px solid var(--line);border-radius:var(--r);margin-top:var(--space-l)}
.score{display:grid;grid-template-columns:1fr}
.score .row{display:grid;grid-template-columns:minmax(0,1fr) 128px 120px;align-items:center;gap:var(--space-m);
  padding:var(--space-m) var(--space-l);border-top:1px solid var(--line)}
.score .row:first-child{border-top:none}
.score .metric{font-weight:500;color:var(--ink)}
.score .metric small{display:block;font-weight:400;color:var(--muted);font-size:.82rem;margin-top:2px}
.score .bar{height:8px;border-radius:100px;background:var(--paper-2);position:relative;overflow:hidden}
.score .bar > i{position:absolute;inset:0 auto 0 0;border-radius:100px;background:var(--accent)}
.score .bar.loss > i{background:var(--line-2)}
.score .val{font-family:var(--font-mono);font-weight:600;font-size:1.02rem;text-align:right}
.score .val.win{color:var(--accent)}
.score .val.eq{color:var(--ink-2)}
.score .val.loss{color:var(--muted)}
@media(max-width:640px){ .score .row{grid-template-columns:1fr 84px;gap:var(--space-s)} .score .bar{display:none} }

.note{margin-top:var(--space-m);font-family:var(--font-mono);font-size:.78rem;color:var(--muted);line-height:1.7;max-width:74ch}
.note b{color:var(--ink-2);font-weight:600}

/* ---- Where it runs (typographic rows, no emoji/no icon-tiles) ---- */
.runs{border-top:1px solid var(--line);margin-top:var(--space-l)}
.runs .r{display:grid;grid-template-columns:180px 1fr;gap:var(--space-l);align-items:baseline;
  padding:var(--space-l) 0;border-bottom:1px solid var(--line)}
.runs .r .t{font-family:var(--font-display);font-weight:600;font-size:1.15rem;letter-spacing:-0.02em}
.runs .r .idx{font-family:var(--font-mono);font-size:.72rem;color:var(--accent);display:block;margin-bottom:6px}
.runs .r p{color:var(--ink-2);max-width:60ch}
@media(max-width:640px){ .runs .r{grid-template-columns:1fr;gap:var(--space-2xs)} }

/* ---- Close / CTA band ---- */
.close{background:var(--ink);color:var(--paper);border-radius:12px;padding:var(--space-2xl) var(--space-xl);margin-block:var(--space-2xl)}
.close h2{color:var(--paper);font-size:clamp(1.6rem,2.4vw + .5rem,2.6rem);max-width:20ch}
.close p{color:#AEB7C6;margin-top:var(--space-s);max-width:52ch}
.close .cta{display:flex;flex-wrap:wrap;gap:var(--space-xs);margin-top:var(--space-l)}
.close .btn-ghost{color:var(--paper);border-color:#3A4453}
.close .btn-ghost:hover{border-color:#5C6675;background:rgba(255,255,255,.04)}

/* ---- Ft2 inline footer (single line, hairline above) ---- */
footer{border-top:1px solid var(--line);padding-block:var(--space-l);margin-top:var(--space-xl)}
.foot-row{display:flex;flex-wrap:wrap;align-items:center;gap:var(--space-xs) var(--space-m);justify-content:space-between}
.foot-row .fb{font-family:var(--font-display);font-weight:700}
.foot-row .fb .dot{color:var(--accent)}
.foot-row .ft{color:var(--muted);font-size:.86rem}
.foot-row .fc{font-family:var(--font-mono);font-size:.72rem;color:var(--muted)}
.foot-row a{color:var(--muted);text-decoration:none;font-size:.86rem}
.foot-row a:hover{color:var(--ink)}

@media(prefers-reduced-motion:reduce){ *{transition:none!important;animation:none!important} }
@media(prefers-color-scheme:dark){
  :root{--paper:#0E1218;--paper-2:#161C25;--panel:#151B24;--ink:#EEF2F8;--ink-2:#B7C0CE;--muted:#8A94A4;--line:#232B36;--line-2:#313B49;--accent:#5B85FF;--accent-wash:#141E36}
  .nav{background:rgba(21,27,36,.78);}
  .close{background:#080B10}
}
</style>
</head>
<body>

<div class="nav-shell">
  <nav class="nav" aria-label="Primary">
    <span class="brand">POCKET<span class="dot">.</span></span>
    <span class="nlinks">
      <a href="#numbers">Benchmarks</a>
      <a href="#run">Where it runs</a>
      <a href="#sovereign">Sovereign</a>
    </span>
    <a class="btn btn-primary" href="#get">Get POCKET</a>
  </nav>
</div>

<main>
  <!-- HERO · stat-led -->
  <section class="hero">
    <div class="wrap hero-grid">
      <div>
        <div class="hero-eyebrow"><span class="rule"></span><span class="eyebrow">On-device inference · GGUF + MLX</span></div>
        <div class="hero-figure tnum">2.69<span class="x">×</span></div>
        <h1>Big-model answers, faster on the CPU you already own.</h1>
        <p class="lede">POCKET runs a 35B-class open model on a phone, a laptop, or a GPU-less server — and decodes <strong>2.69× faster</strong> than the most-downloaded on-device model, at matched quality.</p>
        <div class="cta">
          <a class="btn btn-primary" href="#numbers">See the benchmarks</a>
          <a class="btn btn-ghost" href="#run">How it runs on-device</a>
        </div>
      </div>
      <div class="hero-side">
        <p class="qual">Same laptop. Same prompt. No accelerator. POCKET returns tokens while the baseline is still warming up.</p>
        <p class="foot">measured · CPU decode throughput<br>vs. the most-downloaded on-device model<br>Apple M3, single request</p>
      </div>
    </div>
  </section>

  <!-- SUPPORTING STAT STRIP -->
  <section id="numbers" class="wrap rule-top">
    <div class="strip">
      <div class="cell">
        <div class="n acc tnum">2.22×</div>
        <div class="k">GPU decode</div>
        <div class="d">Throughput on a single consumer GPU, same quality target.</div>
      </div>
      <div class="cell">
        <div class="n tnum">≈ 1.0×</div>
        <div class="k">Quality</div>
        <div class="d">Matched on standard evals — parity, not a downgrade.</div>
      </div>
      <div class="cell">
        <div class="n tnum">0</div>
        <div class="k">GPUs required</div>
        <div class="d">Runs on CPU. A phone or an air-gapped box is enough.</div>
      </div>
      <div class="cell">
        <div class="n tnum">2</div>
        <div class="k">Formats</div>
        <div class="d">GGUF for llama.cpp, MLX for Apple silicon. No custom fork.</div>
      </div>
    </div>
  </section>

  <!-- HONEST SCORECARD -->
  <section class="section wrap">
    <div class="sec-head">
      <div class="eyebrow">Measured, not marketed</div>
      <h2>The full scorecard — including where we trail.</h2>
      <p>Four axes against the most-downloaded on-device model, same quality target. We publish the axis we lose, because a benchmark you can't check is just a claim.</p>
    </div>
    <div class="card">
      <div class="score">
        <div class="row">
          <div class="metric">CPU decode throughput<small>tokens per second, single request</small></div>
          <div class="bar"><i style="right:0"></i></div>
          <div class="val win tnum">2.69×</div>
        </div>
        <div class="row">
          <div class="metric">GPU decode throughput<small>single consumer GPU</small></div>
          <div class="bar"><i style="right:17.5%"></i></div>
          <div class="val win tnum">2.22×</div>
        </div>
        <div class="row">
          <div class="metric">Answer quality<small>standard reasoning + knowledge evals</small></div>
          <div class="bar"><i style="right:62.8%"></i></div>
          <div class="val eq tnum">matched</div>
        </div>
        <div class="row">
          <div class="metric">Prefill on long prompts<small>time-to-first-token, 8k context</small></div>
          <div class="bar loss"><i style="right:84.8%"></i></div>
          <div class="val loss tnum">0.41×</div>
        </div>
      </div>
    </div>
    <p class="note"><b>Reading it straight:</b> POCKET wins decode by a wide margin and holds quality, but the baseline ingests very long prompts faster — POCKET's prefill trails at 0.41×. For chat, agents and short-to-medium prompts, decode is what you feel; for 8k-token document dumps, weigh the trade. Full harness and configs are published.</p>
  </section>

  <!-- WHERE IT RUNS -->
  <section id="run" class="section wrap">
    <div class="sec-head">
      <div class="eyebrow">On-device by design</div>
      <h2>One model. Three places it has no business running — and does.</h2>
    </div>
    <div class="runs">
      <div class="r"><div><span class="idx mono">01</span><span class="t">A phone in your hand</span></div><p>The model lives on the device. Prompts and answers never leave it — no round-trip, no cloud bill, no signal required.</p></div>
      <div class="r"><div><span class="idx mono">02</span><span class="t">A laptop with no GPU</span></div><p>Apple silicon via MLX, everything else via GGUF. A MacBook Air is a workstation; POCKET decodes 2.69× faster than the on-device model most people already downloaded.</p></div>
      <div class="r"><div><span class="idx mono">03</span><span class="t">An air-gapped server</span></div><p>Closed networks, no accelerators, no outbound. The model runs inside the perimeter, which is exactly where regulated work has to stay.</p></div>
    </div>
  </section>

  <!-- SOVEREIGN -->
  <section id="sovereign" class="section wrap">
    <div class="sec-head">
      <div class="eyebrow">For the public sector</div>
      <h2>Sovereign by default.</h2>
      <p>No cloud dependency, no GPU procurement, no data leaving the building. POCKET ships as software you license or as a sealed appliance — built for closed networks, defense, healthcare, and administrative systems that cloud AI cannot enter. Built on Google’s open Gemma, tuned for Korean, distributed as GGUF and MLX.</p>
    </div>
  </section>

  <!-- CLOSE -->
  <section id="get" class="wrap">
    <div class="close">
      <h2>Big-model answers. Pocket-sized footprint.</h2>
      <p>Put a frontier-class model where the work is — on the device, inside the network, off the cloud.</p>
      <div class="cta">
        <a class="btn btn-primary" href="#">Download POCKET</a>
        <a class="btn btn-ghost" href="#">Read the benchmark report</a>
      </div>
    </div>
  </section>
</main>

<footer>
  <div class="wrap foot-row">
    <span class="fb">POCKET<span class="dot">.</span></span>
    <span class="ft">Frontier AI that runs where you are.</span>
    <span class="fc">A VIDRAFT model · GGUF + MLX</span>
  </div>
</footer>

</body>
</html>