File size: 10,009 Bytes
363963d
 
 
 
 
9210bde
 
 
 
 
 
 
 
 
 
 
 
363963d
9210bde
363963d
 
9210bde
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
363963d
9210bde
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
363963d
9210bde
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
363963d
9210bde
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
363963d
 
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
<!DOCTYPE html>
<html lang="en">
  <head>
    <meta charset="UTF-8" />
    <meta name="viewport" content="width=device-width, initial-scale=1.0" />
    <meta
      name="description"
      content="Share your GPU with a browser-based cluster. Run object detection, vision models and more on WebGPU, callable from any app via HTTP."
    />
    <title>WebGPU Cluster — Distributed WebGPU inference</title>
    <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&display=swap"
      rel="stylesheet"
    />
    <script type="module" crossorigin src="/assets/main-DWTtW_kl.js"></script>
    <link rel="modulepreload" crossorigin href="/assets/modulepreload-polyfill-B5Qt9EMX.js">
    <link rel="stylesheet" crossorigin href="/assets/main-Dqiya_1f.css">
  </head>
  <body>
    <header class="site-header">
      <div class="container">
        <a href="/" class="logo" aria-label="WebGPU Cluster home">
          <span class="logo-mark" aria-hidden="true">⚡</span>
          WebGPU Cluster
        </a>
        <nav class="nav-links" aria-label="Main">
          <a href="#how-it-works" class="nav-hide-mobile">How it works</a>
          <a href="#features" class="nav-hide-mobile">Features</a>
          <a
            href="https://github.com/apssouza22/webgpu-video-cluster"
            rel="noopener noreferrer"
            class="nav-hide-mobile"
            >GitHub</a
          >
          <a href="/monitor.html">View the grid</a>
          <a href="/host.html" class="btn btn-primary btn-sm">Join the grid</a>
        </nav>
      </div>
    </header>

    <main>
      <section class="hero">
        <div class="container">
          <div class="badge">
            <span class="badge-dot" aria-hidden="true"></span>
            Distributed GPU network
          </div>
          <h1>
            AI on the grid,<br />
            <span class="gradient-text">powered by your WebGPU.</span>
          </h1>
          <p class="hero-lead">
            Turn any browser with WebGPU into a cluster node. Share inference for LLM models — Host a model on your powerful workstation and access it securely from your phone, laptop, or let others connect to it.
          </p>
          <div class="hero-actions">
            <a href="/host.html" class="btn btn-primary">Join the grid</a>
            <a href="/monitor.html" class="btn btn-secondary">View available grid</a>
          </div>
          <p class="hero-trust">
            Powered by <strong>WebGPU</strong> &amp; <strong>Transformers.js</strong> · No GPU
            drivers to install · Open HTTP API
          </p>

          <div class="live-stats" aria-live="polite" aria-label="Cluster status">
            <div class="live-stat">
              <div id="stat-online" class="live-stat-value online">…</div>
              <div class="live-stat-label">Hosts online</div>
            </div>
            <div class="live-stat">
              <div id="stat-total" class="live-stat-value">…</div>
              <div class="live-stat-label">Hosts registered</div>
            </div>
          </div>
        </div>
      </section>

      <section id="features">
        <div class="container">
          <span class="section-label">Why join</span>
          <h2 class="section-title">Browser-native GPU sharing</h2>
          <p class="section-desc">
            Contribute spare GPU cycles from your workstation. Clients send images over HTTP;
            your browser runs the model and returns results — privately, on your hardware.
          </p>
          <div class="features-grid">
            <article class="feature-card">
              <div class="feature-icon" aria-hidden="true">🖥️</div>
              <h3>Instant local hosting</h3>
              <p>
                Open a tab, pick a model, and start hosting. RF-DETR and SmolVLM load in a Web
                Worker on WebGPU — no Python environment or driver setup.
              </p>
            </article>
            <article class="feature-card">
              <div class="feature-icon" aria-hidden="true">🔒</div>
              <h3>Your data stays local</h3>
              <p>
                Inference runs on your GPU in the browser. Images are processed on your machine;
                nothing is sent to third-party AI APIs.
              </p>
            </article>
            <article class="feature-card">
              <div class="feature-icon" aria-hidden="true">🔌</div>
              <h3>Universal HTTP API</h3>
              <p>
                Connect from curl, Python, Node, or any HTTP client. Simple JSON endpoints for
                detection and image description — queue and broker included.
              </p>
            </article>
          </div>
        </div>
      </section>

      <section id="how-it-works">
        <div class="container">
          <span class="section-label">Architecture</span>
          <h2 class="section-title">How the grid works</h2>
          <p class="section-desc">
            A lightweight Node broker coordinates tasks. Browser hosts stay connected via SSE
            and pull jobs when idle.
          </p>
          <div class="arch-diagram" role="img" aria-label="Request flow diagram">
            <div class="arch-flow">
              <div class="arch-node">
                <strong>Client</strong>
                curl / Python / app
              </div>
              <span class="arch-arrow" aria-hidden="true">→</span>
              <div class="arch-node">
                <strong>Broker</strong>
                task queue · /v1/detect
              </div>
              <span class="arch-arrow" aria-hidden="true">→</span>
              <div class="arch-node highlight">
                <strong>Browser host</strong>
                WebGPU inference
              </div>
              <span class="arch-arrow" aria-hidden="true">→</span>
              <div class="arch-node">
                <strong>Response</strong>
                boxes · labels · text
              </div>
            </div>
          </div>
          <div class="steps" style="margin-top: 48px">
            <div class="step">
              <h3>Register your node</h3>
              <p>
                Open the host page, choose a host id and model, then click Start hosting. Keep
                the tab open while you share GPU time.
              </p>
            </div>
            <div class="step">
              <h3>Jobs arrive via SSE</h3>
              <p>
                The broker forwards detection and description tasks to your browser. One job
                runs at a time per host.
              </p>
            </div>
            <div class="step">
              <h3>Anyone can call the API</h3>
              <p>
                Point clients at <code>POST /v1/detect</code> or <code>/v1/describe</code> with
                your host id. Results return as JSON.
              </p>
            </div>
          </div>
        </div>
      </section>

      <section id="api">
        <div class="container">
          <span class="section-label">API</span>
          <h2 class="section-title">Call the grid from anywhere</h2>
          <p class="section-desc">
            Use the cluster monitor to see online hosts and copy ready-made curl examples.
          </p>
          <div class="code-panel">
            <div class="code-header">
              <span>detect.sh</span>
              <span>POST /v1/detect</span>
            </div>
            <pre><code>curl -X POST '<span class="str">http://localhost:5180/v1/detect</span>' \
  -H <span class="str">'Content-Type: application/json'</span> \
  -d <span class="str">'{
    "host": "my-gpu-node",
    "image_url": "https://example.com/photo.jpg",
    "threshold": 0.5
  }'</span></code></pre>
          </div>
        </div>
      </section>

      <section id="models">
        <div class="container">
          <span class="section-label">Models</span>
          <h2 class="section-title">What you can host today</h2>
          <p class="section-desc">
            Models download from Hugging Face on first load. Pick one per host session.
          </p>
          <div class="models-grid">
            <article class="model-card">
              <span class="model-tag">Detection</span>
              <h3>RF-DETR Medium</h3>
              <p>
                Real-time object detection (COCO) via ONNX on WebGPU. Endpoint:
                <code>POST /v1/detect</code>
              </p>
            </article>
            <article class="model-card">
              <span class="model-tag">Vision-language</span>
              <h3>SmolVLM-500M</h3>
              <p>
                Describe images with a compact VLM on WebGPU. Endpoint:
                <code>POST /v1/describe</code>
              </p>
            </article>
          </div>
        </div>
      </section>

      <section class="cta-section">
        <div class="container">
          <h2>Ready to power the grid?</h2>
          <p>Share your GPU or explore nodes already online.</p>
          <div class="cta-actions">
            <a href="/host.html" class="btn btn-primary">Join the grid</a>
            <a href="/monitor.html" class="btn btn-secondary">View available grid</a>
          </div>
        </div>
      </section>
    </main>

    <footer class="site-footer">
      <div class="container">
        <p>
          WebGPU Cluster · MIT ·
          <a
            href="https://github.com/apssouza22/webgpu-video-cluster"
            rel="noopener noreferrer"
            >Source on GitHub</a
          >
          ·
          <a
            href="https://huggingface.co/spaces/apssouza22/webgpu-cluster"
            rel="noopener noreferrer"
            >Live on Hugging Face</a
          >
        </p>
      </div>
    </footer>

  </body>
</html>