front matter: publication dateline replaces internal draft number
Browse files- index.html +2 -3
index.html
CHANGED
|
@@ -79,10 +79,9 @@ hr{border:none; border-top:1px solid var(--rule); margin:44px 0}
|
|
| 79 |
<p class="eyebrow">Quantization · Apple Silicon · 2026</p>
|
| 80 |
<h1>Data-Free Vector Quantization Beats Affine Quantization at Matched Bytes Below 6 Bits</h1>
|
| 81 |
<p><strong>Noah Zelezny</strong></p>
|
| 82 |
-
<p><em>
|
| 83 |
record; every margin is stated against a measured fit-to-fit noise floor
|
| 84 |
-
for its geometry.
|
| 85 |
-
the corrections they produced are recorded in the ledger alongside it.</em></p>
|
| 86 |
<h2 id="abstract">Abstract</h2>
|
| 87 |
<p>Memory, not compute, is the binding constraint on local inference of
|
| 88 |
large language models: a model is runnable on a given machine only if
|
|
|
|
| 79 |
<p class="eyebrow">Quantization · Apple Silicon · 2026</p>
|
| 80 |
<h1>Data-Free Vector Quantization Beats Affine Quantization at Matched Bytes Below 6 Bits</h1>
|
| 81 |
<p><strong>Noah Zelezny</strong></p>
|
| 82 |
+
<p><em>August 2026. Every number traces to a committed entry in the laboratory
|
| 83 |
record; every margin is stated against a measured fit-to-fit noise floor
|
| 84 |
+
for its geometry.</em></p>
|
|
|
|
| 85 |
<h2 id="abstract">Abstract</h2>
|
| 86 |
<p>Memory, not compute, is the binding constraint on local inference of
|
| 87 |
large language models: a model is runnable on a given machine only if
|