update website

Author: chromecast56

docs/index.html

@@ -272,6 +272,14 @@ <h2 class="title is-4">Insight: Information Disparity in Pre-training vs. Fine-t
           This addresses the <b>serving challenge</b>.
         </p>
 
+        <p style="margin-bottom: 20px;">
+          Past work (GPT-Zip, DeltaZip) has also explored quantization of the weight delta, achieving 
+          quantization levels as low as 2-bits by applying methods introduced by GPTQ. We find that 
+          the weight delta is extremely compressible, and are able to achieve <b>1-bit quantization</b> 
+          with minimal performance degradation using a simpler methodology.
+        </p>
+
+
         <h2 class="title is-4">BitDelta Overview</h2>
         <h2 class="title is-5">1-bit quantization</h2>