Upload folder using huggingface_hub

index.html

@@ -552,7 +552,7 @@
         </div>
         <div class="stage-detail"><div class="stage-detail-inner">
           <p>Uses <strong>causal attention</strong>: each audio frame only attends to itself and past frames — never future ones. This is what makes streaming possible.</p>
-          <p><strong>Sliding window attention</strong> (window = 750, ~60s of audio) bounds memory regardless of audio length.</p>
+          <p><strong>Sliding window attention</strong> (window = 750, ~15s of audio) bounds memory regardless of audio length.</p>
           <div class="key-point">Traditional encoders (like Whisper's) use bidirectional attention — every frame sees every other frame. That gives better accuracy but requires full audio upfront. Causal attention trades a small amount of accuracy for the ability to stream.</div>
         </div></div>
       </div>
@@ -689,7 +689,7 @@
 
     <div class="window-viz">
       <div class="window-info">
-        <span class="wi-label">Encoder window: <span class="wi-value">750 frames</span> (~60s)</span>
+        <span class="wi-label">Encoder window: <span class="wi-value">750 frames</span> (~15s)</span>
         <span class="wi-label">Position: <span class="wi-value" id="window-pos">frame 8</span></span>
       </div>