diff --git a/app/dist/index.html b/app/dist/index.html
index cd5c2d358b4a06af3952fda1f345d0cda1d9410a..662dd192d71ccbffecc1a836533e7ef97cc2aa34 100644
--- a/app/dist/index.html
+++ b/app/dist/index.html
@@ -12,8 +12,8 @@
           document.documentElement.setAttribute("data-theme", theme);
         } catch {}
       })();
-    </script><script type="module" src="/scripts/color-palettes.js"></script><!-- TO MANAGE PROPERLY --><script src="https://cdn.plot.ly/plotly-3.0.0.min.js" charset="utf-8"></script><link rel="stylesheet" href="/_astro/index.B7tpapLd.css"><script type="module" src="/_astro/hoisted.DK-CdsVg.js"></script>
-<script type="module" src="/_astro/page.CH0W_C1Z.js"></script></head> <body> <button id="theme-toggle" aria-label="Toggle color theme" data-astro-cid-x3pjskd3> <svg class="icon light" width="20" height="20" viewBox="0 0 24 24" aria-hidden="true" focusable="false" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" data-astro-cid-x3pjskd3> <circle cx="12" cy="12" r="5" data-astro-cid-x3pjskd3></circle> <line x1="12" y1="1" x2="12" y2="4" data-astro-cid-x3pjskd3></line> <line x1="12" y1="20" x2="12" y2="23" data-astro-cid-x3pjskd3></line> <line x1="1" y1="12" x2="4" y2="12" data-astro-cid-x3pjskd3></line> <line x1="20" y1="12" x2="23" y2="12" data-astro-cid-x3pjskd3></line> <line x1="4.22" y1="4.22" x2="6.34" y2="6.34" data-astro-cid-x3pjskd3></line> <line x1="17.66" y1="17.66" x2="19.78" y2="19.78" data-astro-cid-x3pjskd3></line> <line x1="4.22" y1="19.78" x2="6.34" y2="17.66" data-astro-cid-x3pjskd3></line> <line x1="17.66" y1="6.34" x2="19.78" y2="4.22" data-astro-cid-x3pjskd3></line> </svg> <svg class="icon dark" width="20" height="20" viewBox="0 0 24 24" aria-hidden="true" focusable="false" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" data-astro-cid-x3pjskd3> <path d="M21 12.79A9 9 0 1 1 11.21 3 7 7 0 0 0 21 12.79z" data-astro-cid-x3pjskd3></path> </svg>  </button>  <section class="hero" data-astro-cid-bbe6dxrz> <h1 class="hero-title" data-astro-cid-bbe6dxrz>Maintain the unmaintainable:<br/>1M python loc, 400+ models</h1> <div class="hero-banner" data-astro-cid-bbe6dxrz> <figure class="html-embed"><div class="html-embed__card is-frameless"><div id="frag-2ywxxg8i8lo"><style>
+    </script><script type="module" src="/scripts/color-palettes.js"></script><!-- TO MANAGE PROPERLY --><script src="https://cdn.plot.ly/plotly-3.0.0.min.js" charset="utf-8"></script><link rel="stylesheet" href="/_astro/index.C8LanvBP.css"><script type="module" src="/_astro/hoisted.DK-CdsVg.js"></script>
+<script type="module" src="/_astro/page.CH0W_C1Z.js"></script></head> <body> <button id="theme-toggle" aria-label="Toggle color theme" data-astro-cid-x3pjskd3> <svg class="icon light" width="20" height="20" viewBox="0 0 24 24" aria-hidden="true" focusable="false" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" data-astro-cid-x3pjskd3> <circle cx="12" cy="12" r="5" data-astro-cid-x3pjskd3></circle> <line x1="12" y1="1" x2="12" y2="4" data-astro-cid-x3pjskd3></line> <line x1="12" y1="20" x2="12" y2="23" data-astro-cid-x3pjskd3></line> <line x1="1" y1="12" x2="4" y2="12" data-astro-cid-x3pjskd3></line> <line x1="20" y1="12" x2="23" y2="12" data-astro-cid-x3pjskd3></line> <line x1="4.22" y1="4.22" x2="6.34" y2="6.34" data-astro-cid-x3pjskd3></line> <line x1="17.66" y1="17.66" x2="19.78" y2="19.78" data-astro-cid-x3pjskd3></line> <line x1="4.22" y1="19.78" x2="6.34" y2="17.66" data-astro-cid-x3pjskd3></line> <line x1="17.66" y1="6.34" x2="19.78" y2="4.22" data-astro-cid-x3pjskd3></line> </svg> <svg class="icon dark" width="20" height="20" viewBox="0 0 24 24" aria-hidden="true" focusable="false" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" data-astro-cid-x3pjskd3> <path d="M21 12.79A9 9 0 1 1 11.21 3 7 7 0 0 0 21 12.79z" data-astro-cid-x3pjskd3></path> </svg>  </button>  <section class="hero" data-astro-cid-bbe6dxrz> <h1 class="hero-title" data-astro-cid-bbe6dxrz>Maintain the unmaintainable:<br/>1M python loc, 400+ models</h1> <div class="hero-banner" data-astro-cid-bbe6dxrz> <figure class="html-embed"><div class="html-embed__card is-frameless"><div id="frag-1rxf8nus1ov"><style>
 @import url('https://fonts.googleapis.com/css2?family=Inter:wght@500;600&display=swap');
 
 .banner-container {
@@ -462,23 +462,30 @@ We continue to support all new models and expect to do so for the foreseeable fu
 <p>We formalize and articulate the “tenets” that have been guiding our development, demonstrate how they are implemented in code, and show the measurable impact they have on the library’s sustainability and growth.</p>
 <p>For any OSS maintainer, power user, or contributor, this is the map to understanding, using, and building upon <code>transformers</code>, but not only: any project of comparable size will require you to make deep choices, not only on design and choice of abstraction, but on the very mindset of the software you are building. These tenets may or may not be applicable to your project, but they provide a glimpse on how we work that could be helpful or inspirational.</p>
 <p>Conventions used throughout this post:</p>
-<p><span class="glossary-term " style="" data-tooltip-id="tenet-vfqfhttx57q" onmouseenter="window.showTenetTooltip(event, 'tenet-vfqfhttx57q')" onmousemove="window.updateTenetTooltip(event, 'tenet-vfqfhttx57q')" onmouseleave="window.hideTenetTooltip('tenet-vfqfhttx57q')">Tenets exemplified</span> <div id="tenet-vfqfhttx57q" class="glossary-tooltip" data-position="top"> <div class="glossary-tooltip__content"> <div class="glossary-tooltip__term">source-of-truth</div> <div class="glossary-tooltip__definition">Model implementations should be reliable, reproducible, and faithful to original performances.</div> </div> <div class="glossary-tooltip__arrow"></div> </div> <script>
+<p><span class="glossary-term " style="" data-tooltip-id="tenet-yfjcgvbyhlr" onmouseenter="window.showTenetTooltip(event, 'tenet-yfjcgvbyhlr')" onmousemove="window.updateTenetTooltip(event, 'tenet-yfjcgvbyhlr')" onmouseleave="window.hideTenetTooltip('tenet-yfjcgvbyhlr')">Tenets exemplified</span><span id="tenet-yfjcgvbyhlr" class="glossary-tooltip" data-position="top"> <span class="glossary-tooltip__content"> <span class="glossary-tooltip__term">source-of-truth</span> <span class="glossary-tooltip__definition">Model implementations should be reliable, reproducible, and faithful to original performances.</span> </span> <span class="glossary-tooltip__arrow"></span> </span> <script>
 if (!window.tenetTooltipInitialized) {
   window.tenetTooltipInitialized = true;
 
   window.showTenetTooltip = function(event, id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'block';
+    tooltip.style.visibility = 'visible';
     tooltip.style.opacity = '1';
     tooltip.style.top = (event.clientY + 10) + 'px';
     tooltip.style.left = (event.clientX + 10) + 'px';
   };
 
+  window.updateTenetTooltip = function(event, id) {
+    const tooltip = document.getElementById(id);
+    if (!tooltip) return;
+    tooltip.style.top = (event.clientY + 10) + 'px';
+    tooltip.style.left = (event.clientX + 10) + 'px';
+  };
+
   window.hideTenetTooltip = function(id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'none';
+    tooltip.style.visibility = 'hidden';
     tooltip.style.opacity = '0';
   };
 }
@@ -492,65 +499,86 @@ if (!window.tenetTooltipInitialized) {
 <p>These principles were not decided in a vacuum. The library <em>evolved</em> towards them, and once they <em>emerged</em>, they were recognized as critical.</p>
 <div class="tenet-list"><ol><li class="tenet"><a id="source-of-truth"></a><strong>Source of Truth</strong><p>We aim to be a <a href="https://huggingface.co/blog/transformers-model-definition">source of truth for all model definitions</a>. This is more of a goal than a tenet, but it strongly guides our decisions. Model implementations should be reliable, reproducible, and faithful to the original implementations. If we are successful, they should become reference baselines for the ecosystem, so they’ll be easily adopted by downstream libraries and projects. It’s much easier for a project to always refer to the transformers implementation, than to learn a different research codebase every time a new architecture is released.</p><em>This overarching guideline ensures quality and reproducibility across all models in the library, and aspires to make the community work easier.</em></li><li class="tenet"><a id="one-model-one-file"></a><strong>One Model, One File</strong><p>All inference and training core logic has to be visible, top‑to‑bottom, to maximize each model’s hackability.</p><em>Every model should be completely understandable and hackable by reading a single file from top to bottom.</em></li><li class="tenet"><a id="code-is-product"></a><strong>Code is Product</strong><p>Optimize for reading, diffing, and tweaking, our users are power users. Variables should be explicit, full words, even several words, readability is primordial.</p><em>Code quality matters as much as functionality - optimize for human readers, not just computers.</em></li><li class="tenet"><a id="standardize-dont-abstract"></a><strong>Standardize, Don’t Abstract</strong><p>If it’s model behavior, keep it in the file; use abstractions only for generic infra.</p><em>Model-specific logic belongs in the model file, not hidden behind abstractions.</em></li><li class="tenet"><a id="do-repeat-yourself"></a><strong>DRY* (DO Repeat Yourself)</strong><p>Copy when it helps users; keep successors in sync without centralizing behavior.</p><p><b>Evolution:</b></p><p> With the introduction and global adoption of <a href="#modular">modular</a> transformers, we do not repeat any logic in the modular files, but end user files remain faithful to the original tenet.</p><em>Strategic duplication can improve readability and maintainability when done thoughtfully.</em></li><li class="tenet"><a id="minimal-user-api"></a><strong>Minimal User API</strong><p>Config, model, preprocessing; from_pretrained, save_pretrained, push_to_hub. We want the least amount of codepaths. Reading should be obvious, configurations should be obvious.</p><em>Keep the public interface simple and predictable, users should know what to expect.</em></li><li class="tenet"><a id="backwards-compatibility"></a><strong>Backwards Compatibility</strong><p>Evolve by additive standardization, never break public APIs.</p><p>Any artifact that was once on the hub and loadable with transformers should be usable indefinitely with the same interface. Further, public methods should not change to avoid breaking dependencies. If we do deprecate something, it’s with very long cycles beforehand.</p><em>Once something is public, it stays public, evolution through addition, not breaking changes.</em></li><li class="tenet"><a id="consistent-public-surface"></a><strong>Consistent Public Surface</strong><p>Same argument names, same outputs, hidden states and attentions exposed, enforced by tests. This is a goal as well as a tenet.</p><em>All models should feel familiar - consistent interfaces reduce cognitive load.</em></li></ol></div>
 <p>When a PR is merged, it is because the contribution is worthwhile, and because the <code>transformers</code> team finds the design of the contribution to be aligned with the tenets.</p>
-<p>Does all the code in the library strictly follow these tenets? No. The library is a gigantic house with connected nooks, corridors, crannies everywhere, built by thousands of different workers. We <em>try</em> to make it so all the code added is compliant, because if we fail and merge it, we cannot change it lest we break <span class="glossary-term " style="" data-tooltip-id="tenet-yxer8c3t8d" onmouseenter="window.showTenetTooltip(event, 'tenet-yxer8c3t8d')" onmousemove="window.updateTenetTooltip(event, 'tenet-yxer8c3t8d')" onmouseleave="window.hideTenetTooltip('tenet-yxer8c3t8d')">backwards compatibility</span> <div id="tenet-yxer8c3t8d" class="glossary-tooltip" data-position="top"> <div class="glossary-tooltip__content"> <div class="glossary-tooltip__term">backwards-compatibility</div> <div class="glossary-tooltip__definition">Any artifact once on the hub must remain loadable. Breaking changes are unacceptable.</div> </div> <div class="glossary-tooltip__arrow"></div> </div> <script>
+<p>Does all the code in the library strictly follow these tenets? No. The library is a gigantic house with connected nooks, corridors, crannies everywhere, built by thousands of different workers. We <em>try</em> to make it so all the code added is compliant, because if we fail and merge it, we cannot change it lest we break <span class="glossary-term " style="" data-tooltip-id="tenet-2861dkirkm1" onmouseenter="window.showTenetTooltip(event, 'tenet-2861dkirkm1')" onmousemove="window.updateTenetTooltip(event, 'tenet-2861dkirkm1')" onmouseleave="window.hideTenetTooltip('tenet-2861dkirkm1')">backwards compatibility</span><span id="tenet-2861dkirkm1" class="glossary-tooltip" data-position="top"> <span class="glossary-tooltip__content"> <span class="glossary-tooltip__term">backwards-compatibility</span> <span class="glossary-tooltip__definition">Any artifact once on the hub must remain loadable. Breaking changes are unacceptable.</span> </span> <span class="glossary-tooltip__arrow"></span> </span> <script>
 if (!window.tenetTooltipInitialized) {
   window.tenetTooltipInitialized = true;
 
   window.showTenetTooltip = function(event, id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'block';
+    tooltip.style.visibility = 'visible';
     tooltip.style.opacity = '1';
     tooltip.style.top = (event.clientY + 10) + 'px';
     tooltip.style.left = (event.clientX + 10) + 'px';
   };
 
+  window.updateTenetTooltip = function(event, id) {
+    const tooltip = document.getElementById(id);
+    if (!tooltip) return;
+    tooltip.style.top = (event.clientY + 10) + 'px';
+    tooltip.style.left = (event.clientX + 10) + 'px';
+  };
+
   window.hideTenetTooltip = function(id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'none';
+    tooltip.style.visibility = 'hidden';
     tooltip.style.opacity = '0';
   };
 }
 </script>.</p>
 <p>To see what constitutes adherence to the tenets, let’s take the example of code repetition.</p>
-<p>The following function, essential to the implementation of <a href="https://huggingface.co/papers/2104.09864">Rotary Positional Embeddings</a> can be found in more than 70 <code>modeling_&lt;file&gt;.py</code> across <code>src/transformers/models/.</code>  Why keep it? Because we want all the model logic to be <span class="glossary-term " style="" data-tooltip-id="tenet-mhdomygdtq" onmouseenter="window.showTenetTooltip(event, 'tenet-mhdomygdtq')" onmousemove="window.updateTenetTooltip(event, 'tenet-mhdomygdtq')" onmouseleave="window.hideTenetTooltip('tenet-mhdomygdtq')">contained in the modeling file</span> <div id="tenet-mhdomygdtq" class="glossary-tooltip" data-position="top"> <div class="glossary-tooltip__content"> <div class="glossary-tooltip__term">one-model-one-file</div> <div class="glossary-tooltip__definition">All inference and training core logic visible, top‑to‑bottom, in a single file.</div> </div> <div class="glossary-tooltip__arrow"></div> </div> <script>
+<p>The following function, essential to the implementation of <a href="https://huggingface.co/papers/2104.09864">Rotary Positional Embeddings</a> can be found in more than 70 <code>modeling_&lt;file&gt;.py</code> across <code>src/transformers/models/.</code>  Why keep it? Because we want all the model logic to be <span class="glossary-term " style="" data-tooltip-id="tenet-ofinvupiood" onmouseenter="window.showTenetTooltip(event, 'tenet-ofinvupiood')" onmousemove="window.updateTenetTooltip(event, 'tenet-ofinvupiood')" onmouseleave="window.hideTenetTooltip('tenet-ofinvupiood')">contained in the modeling file</span><span id="tenet-ofinvupiood" class="glossary-tooltip" data-position="top"> <span class="glossary-tooltip__content"> <span class="glossary-tooltip__term">one-model-one-file</span> <span class="glossary-tooltip__definition">All inference and training core logic visible, top‑to‑bottom, in a single file.</span> </span> <span class="glossary-tooltip__arrow"></span> </span> <script>
 if (!window.tenetTooltipInitialized) {
   window.tenetTooltipInitialized = true;
 
   window.showTenetTooltip = function(event, id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'block';
+    tooltip.style.visibility = 'visible';
     tooltip.style.opacity = '1';
     tooltip.style.top = (event.clientY + 10) + 'px';
     tooltip.style.left = (event.clientX + 10) + 'px';
   };
 
+  window.updateTenetTooltip = function(event, id) {
+    const tooltip = document.getElementById(id);
+    if (!tooltip) return;
+    tooltip.style.top = (event.clientY + 10) + 'px';
+    tooltip.style.left = (event.clientX + 10) + 'px';
+  };
+
   window.hideTenetTooltip = function(id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'none';
+    tooltip.style.visibility = 'hidden';
     tooltip.style.opacity = '0';
   };
 }
-</script>. In order to do that, we <span class="glossary-term " style="" data-tooltip-id="tenet-yszy20kbe7" onmouseenter="window.showTenetTooltip(event, 'tenet-yszy20kbe7')" onmousemove="window.updateTenetTooltip(event, 'tenet-yszy20kbe7')" onmouseleave="window.hideTenetTooltip('tenet-yszy20kbe7')">do repeat ourselves</span> <div id="tenet-yszy20kbe7" class="glossary-tooltip" data-position="top"> <div class="glossary-tooltip__content"> <div class="glossary-tooltip__term">do-repeat-yourself</div> <div class="glossary-tooltip__definition">Strategic duplication can improve readability and maintainability when done thoughtfully.</div> </div> <div class="glossary-tooltip__arrow"></div> </div> <script>
+</script>. In order to do that, we <span class="glossary-term " style="" data-tooltip-id="tenet-aspwl783c7s" onmouseenter="window.showTenetTooltip(event, 'tenet-aspwl783c7s')" onmousemove="window.updateTenetTooltip(event, 'tenet-aspwl783c7s')" onmouseleave="window.hideTenetTooltip('tenet-aspwl783c7s')">do repeat ourselves</span><span id="tenet-aspwl783c7s" class="glossary-tooltip" data-position="top"> <span class="glossary-tooltip__content"> <span class="glossary-tooltip__term">do-repeat-yourself</span> <span class="glossary-tooltip__definition">Strategic duplication can improve readability and maintainability when done thoughtfully.</span> </span> <span class="glossary-tooltip__arrow"></span> </span> <script>
 if (!window.tenetTooltipInitialized) {
   window.tenetTooltipInitialized = true;
 
   window.showTenetTooltip = function(event, id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'block';
+    tooltip.style.visibility = 'visible';
     tooltip.style.opacity = '1';
     tooltip.style.top = (event.clientY + 10) + 'px';
     tooltip.style.left = (event.clientX + 10) + 'px';
   };
 
+  window.updateTenetTooltip = function(event, id) {
+    const tooltip = document.getElementById(id);
+    if (!tooltip) return;
+    tooltip.style.top = (event.clientY + 10) + 'px';
+    tooltip.style.left = (event.clientX + 10) + 'px';
+  };
+
   window.hideTenetTooltip = function(id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'none';
+    tooltip.style.visibility = 'hidden';
     tooltip.style.opacity = '0';
   };
 }
@@ -564,108 +592,143 @@ if (!window.tenetTooltipInitialized) {
 <p>We want all models to have self-contained modeling code. Every core functionality <em>must</em> be in the modeling code, every non-core functionality <em>can</em> be outside of it.</p>
 <p>This comes at a great cost. For years, we have used what we call the <code>#Copied from...</code> mechanism: we added comments of a specific format documenting that some code was copied from another model, saving time both for the reviewers and for the CI: we had tooling to ensure that the copied blocks remained in sync.</p>
 <p>But the LOC count kept creeping up. Each new model copied over hundreds of lines that we considered largely boilerplate, yet, we could not remove them.</p>
-<p>We needed to separate two principles that were so far intertwined, <span class="glossary-term " style="" data-tooltip-id="tenet-gb9v99mbsah" onmouseenter="window.showTenetTooltip(event, 'tenet-gb9v99mbsah')" onmousemove="window.updateTenetTooltip(event, 'tenet-gb9v99mbsah')" onmouseleave="window.hideTenetTooltip('tenet-gb9v99mbsah')">repetition</span> <div id="tenet-gb9v99mbsah" class="glossary-tooltip" data-position="top"> <div class="glossary-tooltip__content"> <div class="glossary-tooltip__term">do-repeat-yourself</div> <div class="glossary-tooltip__definition">Strategic duplication can improve readability and maintainability when done thoughtfully.</div> </div> <div class="glossary-tooltip__arrow"></div> </div> <script>
+<p>We needed to separate two principles that were so far intertwined, <span class="glossary-term " style="" data-tooltip-id="tenet-ab4jtfea6wq" onmouseenter="window.showTenetTooltip(event, 'tenet-ab4jtfea6wq')" onmousemove="window.updateTenetTooltip(event, 'tenet-ab4jtfea6wq')" onmouseleave="window.hideTenetTooltip('tenet-ab4jtfea6wq')">repetition</span><span id="tenet-ab4jtfea6wq" class="glossary-tooltip" data-position="top"> <span class="glossary-tooltip__content"> <span class="glossary-tooltip__term">do-repeat-yourself</span> <span class="glossary-tooltip__definition">Strategic duplication can improve readability and maintainability when done thoughtfully.</span> </span> <span class="glossary-tooltip__arrow"></span> </span> <script>
 if (!window.tenetTooltipInitialized) {
   window.tenetTooltipInitialized = true;
 
   window.showTenetTooltip = function(event, id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'block';
+    tooltip.style.visibility = 'visible';
     tooltip.style.opacity = '1';
     tooltip.style.top = (event.clientY + 10) + 'px';
     tooltip.style.left = (event.clientX + 10) + 'px';
   };
 
+  window.updateTenetTooltip = function(event, id) {
+    const tooltip = document.getElementById(id);
+    if (!tooltip) return;
+    tooltip.style.top = (event.clientY + 10) + 'px';
+    tooltip.style.left = (event.clientX + 10) + 'px';
+  };
+
   window.hideTenetTooltip = function(id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'none';
+    tooltip.style.visibility = 'hidden';
     tooltip.style.opacity = '0';
   };
 }
-</script> and <span class="glossary-term " style="" data-tooltip-id="tenet-23g571itqla" onmouseenter="window.showTenetTooltip(event, 'tenet-23g571itqla')" onmousemove="window.updateTenetTooltip(event, 'tenet-23g571itqla')" onmouseleave="window.hideTenetTooltip('tenet-23g571itqla')">hackability</span> <div id="tenet-23g571itqla" class="glossary-tooltip" data-position="top"> <div class="glossary-tooltip__content"> <div class="glossary-tooltip__term">one-model-one-file</div> <div class="glossary-tooltip__definition">All inference and training core logic visible, top‑to‑bottom, in a single file.</div> </div> <div class="glossary-tooltip__arrow"></div> </div> <script>
+</script> and <span class="glossary-term " style="" data-tooltip-id="tenet-f7qfirhzdrj" onmouseenter="window.showTenetTooltip(event, 'tenet-f7qfirhzdrj')" onmousemove="window.updateTenetTooltip(event, 'tenet-f7qfirhzdrj')" onmouseleave="window.hideTenetTooltip('tenet-f7qfirhzdrj')">hackability</span><span id="tenet-f7qfirhzdrj" class="glossary-tooltip" data-position="top"> <span class="glossary-tooltip__content"> <span class="glossary-tooltip__term">one-model-one-file</span> <span class="glossary-tooltip__definition">All inference and training core logic visible, top‑to‑bottom, in a single file.</span> </span> <span class="glossary-tooltip__arrow"></span> </span> <script>
 if (!window.tenetTooltipInitialized) {
   window.tenetTooltipInitialized = true;
 
   window.showTenetTooltip = function(event, id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'block';
+    tooltip.style.visibility = 'visible';
     tooltip.style.opacity = '1';
     tooltip.style.top = (event.clientY + 10) + 'px';
     tooltip.style.left = (event.clientX + 10) + 'px';
   };
 
+  window.updateTenetTooltip = function(event, id) {
+    const tooltip = document.getElementById(id);
+    if (!tooltip) return;
+    tooltip.style.top = (event.clientY + 10) + 'px';
+    tooltip.style.left = (event.clientX + 10) + 'px';
+  };
+
   window.hideTenetTooltip = function(id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'none';
+    tooltip.style.visibility = 'hidden';
     tooltip.style.opacity = '0';
   };
 }
 </script>.</p>
 <p>What was the solution to this? Let’s talk about modular transformers.</p>
-<div class="crumbs"><p><strong>TL;DR:</strong> Read the code in one place, <span class="glossary-term " style="" data-tooltip-id="tenet-jxioesjdedp" onmouseenter="window.showTenetTooltip(event, 'tenet-jxioesjdedp')" onmousemove="window.updateTenetTooltip(event, 'tenet-jxioesjdedp')" onmouseleave="window.hideTenetTooltip('tenet-jxioesjdedp')">one model, one file.</span> <div id="tenet-jxioesjdedp" class="glossary-tooltip" data-position="top"> <div class="glossary-tooltip__content"> <div class="glossary-tooltip__term">one-model-one-file</div> <div class="glossary-tooltip__definition">All inference and training core logic visible, top‑to‑bottom, in a single file.</div> </div> <div class="glossary-tooltip__arrow"></div> </div> <script>
+<div class="note note--info" data-astro-cid-qg6lmfty> <div class="note__layout" data-astro-cid-qg6lmfty>  <div class="note__body" data-astro-cid-qg6lmfty>  <div class="note__content" data-astro-cid-qg6lmfty> <p><strong>TL;DR:</strong> Read the code in one place, <span class="glossary-term " style="" data-tooltip-id="tenet-btv1xkls4h" onmouseenter="window.showTenetTooltip(event, 'tenet-btv1xkls4h')" onmousemove="window.updateTenetTooltip(event, 'tenet-btv1xkls4h')" onmouseleave="window.hideTenetTooltip('tenet-btv1xkls4h')">one model, one file.</span><span id="tenet-btv1xkls4h" class="glossary-tooltip" data-position="top"> <span class="glossary-tooltip__content"> <span class="glossary-tooltip__term">one-model-one-file</span> <span class="glossary-tooltip__definition">All inference and training core logic visible, top‑to‑bottom, in a single file.</span> </span> <span class="glossary-tooltip__arrow"></span> </span> <script>
 if (!window.tenetTooltipInitialized) {
   window.tenetTooltipInitialized = true;
 
   window.showTenetTooltip = function(event, id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'block';
+    tooltip.style.visibility = 'visible';
     tooltip.style.opacity = '1';
     tooltip.style.top = (event.clientY + 10) + 'px';
     tooltip.style.left = (event.clientX + 10) + 'px';
   };
 
+  window.updateTenetTooltip = function(event, id) {
+    const tooltip = document.getElementById(id);
+    if (!tooltip) return;
+    tooltip.style.top = (event.clientY + 10) + 'px';
+    tooltip.style.left = (event.clientX + 10) + 'px';
+  };
+
   window.hideTenetTooltip = function(id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'none';
+    tooltip.style.visibility = 'hidden';
     tooltip.style.opacity = '0';
   };
 }
-</script>. Keep semantics local (<a href="#standardize-dont-abstract">Standardize, Don’t Abstract</a>). Allow strategic duplication for end users (<a href="#do-repeat-yourself">DRY*</a>). Keep the public surface minimal and stable (<a href="#minimal-user-api">Minimal API</a>, <a href="#backwards-compatibility">Backwards Compatibility</a>, <a href="#consistent-public-surface">Consistent Surface</a>).</p><p><strong>Next:</strong> how modular transformers honor these while removing boilerplate.</p></div>
+</script>. Keep semantics local (<a href="#standardize-dont-abstract">Standardize, Don’t Abstract</a>). Allow strategic duplication for end users (<a href="#do-repeat-yourself">DRY*</a>). Keep the public surface minimal and stable (<a href="#minimal-user-api">Minimal API</a>, <a href="#backwards-compatibility">Backwards Compatibility</a>, <a href="#consistent-public-surface">Consistent Surface</a>).</p><p><strong>Next:</strong> how modular transformers honor these while removing boilerplate.</p> </div> </div> </div> </div> 
 <h2 id="-modular-transformers"><a href="#-modular-transformers"><a id="modular"></a> Modular transformers</a></h2>
-<p>Transformers is an opinionated library. The previous <a href="https://huggingface.co/docs/transformers/en/philosophy">philosophy</a> page, and the <a href="https://huggingface.co/blog/transformers-design-philosophy">blog post</a> were already pointing at the drawbacks mentioned just above, which have been iteratively addressed. <a href="https://huggingface.co/docs/transformers/en/modular_transformers"><code>modular</code> transformers was introduced</a> to allow a form of inheritance without breaking <span class="glossary-term " style="" data-tooltip-id="tenet-z83ygjkl7b" onmouseenter="window.showTenetTooltip(event, 'tenet-z83ygjkl7b')" onmousemove="window.updateTenetTooltip(event, 'tenet-z83ygjkl7b')" onmouseleave="window.hideTenetTooltip('tenet-z83ygjkl7b')">the one model, one file rule.</span> <div id="tenet-z83ygjkl7b" class="glossary-tooltip" data-position="top"> <div class="glossary-tooltip__content"> <div class="glossary-tooltip__term">one-model-one-file</div> <div class="glossary-tooltip__definition">All inference and training core logic visible, top‑to‑bottom, in a single file.</div> </div> <div class="glossary-tooltip__arrow"></div> </div> <script>
+<p>Transformers is an opinionated library. The previous <a href="https://huggingface.co/docs/transformers/en/philosophy">philosophy</a> page, and the <a href="https://huggingface.co/blog/transformers-design-philosophy">blog post</a> were already pointing at the drawbacks mentioned just above, which have been iteratively addressed. <a href="https://huggingface.co/docs/transformers/en/modular_transformers"><code>modular</code> transformers was introduced</a> to allow a form of inheritance without breaking <span class="glossary-term " style="" data-tooltip-id="tenet-34w21zjvfby" onmouseenter="window.showTenetTooltip(event, 'tenet-34w21zjvfby')" onmousemove="window.updateTenetTooltip(event, 'tenet-34w21zjvfby')" onmouseleave="window.hideTenetTooltip('tenet-34w21zjvfby')">the one model, one file rule.</span><span id="tenet-34w21zjvfby" class="glossary-tooltip" data-position="top"> <span class="glossary-tooltip__content"> <span class="glossary-tooltip__term">one-model-one-file</span> <span class="glossary-tooltip__definition">All inference and training core logic visible, top‑to‑bottom, in a single file.</span> </span> <span class="glossary-tooltip__arrow"></span> </span> <script>
 if (!window.tenetTooltipInitialized) {
   window.tenetTooltipInitialized = true;
 
   window.showTenetTooltip = function(event, id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'block';
+    tooltip.style.visibility = 'visible';
     tooltip.style.opacity = '1';
     tooltip.style.top = (event.clientY + 10) + 'px';
     tooltip.style.left = (event.clientX + 10) + 'px';
   };
 
+  window.updateTenetTooltip = function(event, id) {
+    const tooltip = document.getElementById(id);
+    if (!tooltip) return;
+    tooltip.style.top = (event.clientY + 10) + 'px';
+    tooltip.style.left = (event.clientX + 10) + 'px';
+  };
+
   window.hideTenetTooltip = function(id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'none';
+    tooltip.style.visibility = 'hidden';
     tooltip.style.opacity = '0';
   };
 }
 </script></p>
-<p>We amended the principle of <span class="glossary-term " style="" data-tooltip-id="tenet-7aijc3uf215" onmouseenter="window.showTenetTooltip(event, 'tenet-7aijc3uf215')" onmousemove="window.updateTenetTooltip(event, 'tenet-7aijc3uf215')" onmouseleave="window.hideTenetTooltip('tenet-7aijc3uf215')">DRY*</span> <div id="tenet-7aijc3uf215" class="glossary-tooltip" data-position="top"> <div class="glossary-tooltip__content"> <div class="glossary-tooltip__term">do-repeat-yourself</div> <div class="glossary-tooltip__definition">Strategic duplication can improve readability and maintainability when done thoughtfully.</div> </div> <div class="glossary-tooltip__arrow"></div> </div> <script>
+<p>We amended the principle of <span class="glossary-term " style="" data-tooltip-id="tenet-oqey1c1arlp" onmouseenter="window.showTenetTooltip(event, 'tenet-oqey1c1arlp')" onmousemove="window.updateTenetTooltip(event, 'tenet-oqey1c1arlp')" onmouseleave="window.hideTenetTooltip('tenet-oqey1c1arlp')">DRY*</span><span id="tenet-oqey1c1arlp" class="glossary-tooltip" data-position="top"> <span class="glossary-tooltip__content"> <span class="glossary-tooltip__term">do-repeat-yourself</span> <span class="glossary-tooltip__definition">Strategic duplication can improve readability and maintainability when done thoughtfully.</span> </span> <span class="glossary-tooltip__arrow"></span> </span> <script>
 if (!window.tenetTooltipInitialized) {
   window.tenetTooltipInitialized = true;
 
   window.showTenetTooltip = function(event, id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'block';
+    tooltip.style.visibility = 'visible';
     tooltip.style.opacity = '1';
     tooltip.style.top = (event.clientY + 10) + 'px';
     tooltip.style.left = (event.clientX + 10) + 'px';
   };
 
+  window.updateTenetTooltip = function(event, id) {
+    const tooltip = document.getElementById(id);
+    if (!tooltip) return;
+    tooltip.style.top = (event.clientY + 10) + 'px';
+    tooltip.style.left = (event.clientX + 10) + 'px';
+  };
+
   window.hideTenetTooltip = function(id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'none';
+    tooltip.style.visibility = 'hidden';
     tooltip.style.opacity = '0';
   };
 }
@@ -686,7 +749,7 @@ The modular file can use inheritance across models: and then, it will be unravel
 <span class="line"></span>
 <span class="line"><span style="--shiki-light:#D73A49;--shiki-dark:#F97583">class</span><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0"> GlmForCausalLM</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8">(</span><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0">LlamaForCausalLM</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8">):</span></span>
 <span class="line"><span style="--shiki-light:#D73A49;--shiki-dark:#F97583">    pass</span></span>
-<span class="line"></span></code></pre></div></div><div style="overflow-y: auto;"><h4 id="modeling_glmpy-auto-expanded"><a href="#modeling_glmpy-auto-expanded">modeling_glm.py (auto-expanded)</a></h4><div class="code-card"><button class="code-copy button--ghost" type="button" aria-label="Copy code"><svg viewBox="0 0 24 24" aria-hidden="true" focusable="false"><path d="M16 1H4c-1.1 0-2 .9-2 2v12h2V3h12V1zm3 4H8c-1.1 0-2 .9-2 2v14c0 1.1.9 2 2 2h11c1.1 0 2-.9 2-2V7c0-1.1-.9-2-2-2zm0 16H8V7h11v14z"></path></svg></button><pre class="astro-code astro-code-themes github-light github-dark" style="--shiki-light:#24292e;--shiki-dark:#e1e4e8;--shiki-light-bg:#fff;--shiki-dark-bg:#24292e;overflow-x:auto" tabindex="0" data-language="python"><code><span class="line"><span style="--shiki-light:#D73A49;--shiki-dark:#F97583">class</span><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0"> GlmMLP</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8">(</span><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0">nn</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8">.</span><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0">Module</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8">):</span></span>
+<span class="line"></span></code></pre></div></div><div style="max-height: 400px; overflow-y: auto !important; overflow-x: hidden;"><h4 id="modeling_glmpy-auto-expanded"><a href="#modeling_glmpy-auto-expanded">modeling_glm.py (auto-expanded)</a></h4><div class="code-card"><button class="code-copy button--ghost" type="button" aria-label="Copy code"><svg viewBox="0 0 24 24" aria-hidden="true" focusable="false"><path d="M16 1H4c-1.1 0-2 .9-2 2v12h2V3h12V1zm3 4H8c-1.1 0-2 .9-2 2v14c0 1.1.9 2 2 2h11c1.1 0 2-.9 2-2V7c0-1.1-.9-2-2-2zm0 16H8V7h11v14z"></path></svg></button><pre class="astro-code astro-code-themes github-light github-dark" style="--shiki-light:#24292e;--shiki-dark:#e1e4e8;--shiki-light-bg:#fff;--shiki-dark-bg:#24292e;overflow-x:auto" tabindex="0" data-language="python"><code><span class="line"><span style="--shiki-light:#D73A49;--shiki-dark:#F97583">class</span><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0"> GlmMLP</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8">(</span><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0">nn</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8">.</span><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0">Module</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8">):</span></span>
 <span class="line"><span style="--shiki-light:#D73A49;--shiki-dark:#F97583">    def</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF"> __init__</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8">(self, config):</span></span>
 <span class="line"><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF">        super</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8">().</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF">__init__</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8">()</span></span>
 <span class="line"><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF">        self</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8">.config </span><span style="--shiki-light:#D73A49;--shiki-dark:#F97583">=</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8"> config</span></span>
@@ -805,29 +868,36 @@ The modular file can use inheritance across models: and then, it will be unravel
 <span class="line"><span style="--shiki-light:#6A737D;--shiki-dark:#6A737D"># ... (many more classes and functions would follow)</span></span>
 <span class="line"></span></code></pre></div></div> </div>  </div> </div> <figcaption class="reference__caption" data-astro-cid-e5g6tzce><strong>Left:</strong> Clean modular definition with inheritance. <strong>Right:</strong> Auto-expanded version with all inherited functionality visible.</figcaption> </figure> </div> 
 <p>As you can see, we can define a new model as a <em>modular</em> combination of fragments taken from others.</p>
-<p>You might think “well that’s just how inheritance works”. The crucial difference is that we do <em>visibly</em> what is essentially the <em>compiler</em>’s job: by unrolling the inheritances, we make visible all of the modeling code, keeping it <span class="glossary-term " style="" data-tooltip-id="tenet-249n1d9yo6x" onmouseenter="window.showTenetTooltip(event, 'tenet-249n1d9yo6x')" onmousemove="window.updateTenetTooltip(event, 'tenet-249n1d9yo6x')" onmouseleave="window.hideTenetTooltip('tenet-249n1d9yo6x')">all in one piece.</span> <div id="tenet-249n1d9yo6x" class="glossary-tooltip" data-position="top"> <div class="glossary-tooltip__content"> <div class="glossary-tooltip__term">one-model-one-file</div> <div class="glossary-tooltip__definition">All inference and training core logic visible, top‑to‑bottom, in a single file.</div> </div> <div class="glossary-tooltip__arrow"></div> </div> <script>
+<p>You might think “well that’s just how inheritance works”. The crucial difference is that we do <em>visibly</em> what is essentially the <em>compiler</em>’s job: by unrolling the inheritances, we make visible all of the modeling code, keeping it <span class="glossary-term " style="" data-tooltip-id="tenet-huctzo3katf" onmouseenter="window.showTenetTooltip(event, 'tenet-huctzo3katf')" onmousemove="window.updateTenetTooltip(event, 'tenet-huctzo3katf')" onmouseleave="window.hideTenetTooltip('tenet-huctzo3katf')">all in one piece.</span><span id="tenet-huctzo3katf" class="glossary-tooltip" data-position="top"> <span class="glossary-tooltip__content"> <span class="glossary-tooltip__term">one-model-one-file</span> <span class="glossary-tooltip__definition">All inference and training core logic visible, top‑to‑bottom, in a single file.</span> </span> <span class="glossary-tooltip__arrow"></span> </span> <script>
 if (!window.tenetTooltipInitialized) {
   window.tenetTooltipInitialized = true;
 
   window.showTenetTooltip = function(event, id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'block';
+    tooltip.style.visibility = 'visible';
     tooltip.style.opacity = '1';
     tooltip.style.top = (event.clientY + 10) + 'px';
     tooltip.style.left = (event.clientX + 10) + 'px';
   };
 
+  window.updateTenetTooltip = function(event, id) {
+    const tooltip = document.getElementById(id);
+    if (!tooltip) return;
+    tooltip.style.top = (event.clientY + 10) + 'px';
+    tooltip.style.left = (event.clientX + 10) + 'px';
+  };
+
   window.hideTenetTooltip = function(id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'none';
+    tooltip.style.visibility = 'hidden';
     tooltip.style.opacity = '0';
   };
 }
 </script></p>
 <p>You can see below the difference between <code>GlmAttention</code> and <code>LlamaAttention</code>, with the latter having been copied with minimal changes.</p>
-<div class="wide"> <div class="ri-root" data-ri-root="ri_qcvk2kbsnr" data-has-caption data-astro-cid-6kov3kig> <figure class="" data-astro-cid-6kov3kig> <img src="/_astro/llama_glm_attn.D7pkKjAT_1axKuC.webp" alt="Llama vs GLM" data-zoomable="1" data-astro-cid-6kov3kig width="2169" height="482" loading="lazy" decoding="async"> <figcaption data-astro-cid-6kov3kig> <span data-astro-cid-6kov3kig><strong>Figure 1:</strong> Comparison of attention implementations between Llama and GLM, showing code reuse with minimal modifications.</span> </figcaption> </figure> </div> <script>
+<div class="wide"> <div class="ri-root" data-ri-root="ri_f6pon23f7gu" data-has-caption data-astro-cid-6kov3kig> <figure class="" data-astro-cid-6kov3kig> <img src="/_astro/llama_glm_attn.D7pkKjAT_1axKuC.webp" alt="Llama vs GLM" data-zoomable="1" data-astro-cid-6kov3kig width="2169" height="482" loading="lazy" decoding="async"> <figcaption data-astro-cid-6kov3kig> <span data-astro-cid-6kov3kig><strong>Figure 1:</strong> Comparison of attention implementations between Llama and GLM, showing code reuse with minimal modifications.</span> </figcaption> </figure> </div> <script>
   (() => {
     const scriptEl = document.currentScript;
     const root = scriptEl ? scriptEl.previousElementSibling : null;
@@ -998,34 +1068,41 @@ if (!window.tenetTooltipInitialized) {
 <p>When <code>AutoModel.from_pretrained(...)</code> is called, it is indeed the modeling (right side) that is ran, and all the tests run on the modeling code.</p>
 <p>More importantly, the auto-generated modeling file is what users <em>read</em> to understand the code, what they step through in their debuggers and what they hack for their needs.</p>
 <p>What does that give us?</p>
-<div class="crumbs"><p><strong>TL;DR:</strong> A small <code>modular_*.py</code> declares reuse; the expanded modeling file stays visible and <span class="glossary-term " style="" data-tooltip-id="tenet-4wbjr8m9aeg" onmouseenter="window.showTenetTooltip(event, 'tenet-4wbjr8m9aeg')" onmousemove="window.updateTenetTooltip(event, 'tenet-4wbjr8m9aeg')" onmouseleave="window.hideTenetTooltip('tenet-4wbjr8m9aeg')">unique</span> <div id="tenet-4wbjr8m9aeg" class="glossary-tooltip" data-position="top"> <div class="glossary-tooltip__content"> <div class="glossary-tooltip__term">one-model-one-file</div> <div class="glossary-tooltip__definition">All inference and training core logic visible, top‑to‑bottom, in a single file.</div> </div> <div class="glossary-tooltip__arrow"></div> </div> <script>
+<div class="note note--info" data-astro-cid-qg6lmfty> <div class="note__layout" data-astro-cid-qg6lmfty>  <div class="note__body" data-astro-cid-qg6lmfty>  <div class="note__content" data-astro-cid-qg6lmfty> <p><strong>TL;DR:</strong> A small <code>modular_*.py</code> declares reuse; the expanded modeling file stays visible and <span class="glossary-term " style="" data-tooltip-id="tenet-o4j5vh3jbz" onmouseenter="window.showTenetTooltip(event, 'tenet-o4j5vh3jbz')" onmousemove="window.updateTenetTooltip(event, 'tenet-o4j5vh3jbz')" onmouseleave="window.hideTenetTooltip('tenet-o4j5vh3jbz')">unique</span><span id="tenet-o4j5vh3jbz" class="glossary-tooltip" data-position="top"> <span class="glossary-tooltip__content"> <span class="glossary-tooltip__term">one-model-one-file</span> <span class="glossary-tooltip__definition">All inference and training core logic visible, top‑to‑bottom, in a single file.</span> </span> <span class="glossary-tooltip__arrow"></span> </span> <script>
 if (!window.tenetTooltipInitialized) {
   window.tenetTooltipInitialized = true;
 
   window.showTenetTooltip = function(event, id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'block';
+    tooltip.style.visibility = 'visible';
     tooltip.style.opacity = '1';
     tooltip.style.top = (event.clientY + 10) + 'px';
     tooltip.style.left = (event.clientX + 10) + 'px';
   };
 
+  window.updateTenetTooltip = function(event, id) {
+    const tooltip = document.getElementById(id);
+    if (!tooltip) return;
+    tooltip.style.top = (event.clientY + 10) + 'px';
+    tooltip.style.left = (event.clientX + 10) + 'px';
+  };
+
   window.hideTenetTooltip = function(id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'none';
+    tooltip.style.visibility = 'hidden';
     tooltip.style.opacity = '0';
   };
 }
-</script>. Reviewers and contributors maintain the shard, not the repetition.</p><p><strong>Next:</strong> the measurable effect on effective LOC and maintenance cost.</p></div>
+</script>. Reviewers and contributors maintain the shard, not the repetition.</p><p><strong>Next:</strong> the measurable effect on effective LOC and maintenance cost.</p> </div> </div> </div> </div> 
 <h3 id="a-maintainable-control-surface"><a href="#a-maintainable-control-surface">A maintainable control surface</a></h3>
 <p>The effect of modular can be measured in lines of code (LOC). If a model only has a modeling file, we add its LOC count.
 However, if a model has a modular_<em>.py and a corresponding automatically generated modeling_</em>/.py, we only count the LOC under the modular file. The modeling code has no maintenance cost as it is strictly dependent on the modular file.</p>
 <p>That gives an “effective LOC” curve: the 𝗺𝗮𝗶𝗻𝘁𝗲𝗻𝗮𝗻𝗰𝗲 𝘀𝘂𝗿𝗳𝗮𝗰𝗲.</p>
 <p>Measured on git history, raw <code>modeling_*.py</code> grew at ~362 LOC/day before modular; counting only modular shards yields ~25 LOC/day after — about <strong>15× lower</strong>. The effective curve (blue line below) represents the <strong>maintenance surface</strong> today: what maintainers actually read and review.</p>
 <p>Less code to hand-maintain means fewer places to break. Naturally LOC is not a direct measure of complexity, but they correlate in review effort and change risk.</p>
-<figure class="html-embed"><div class="html-embed__card"><div id="frag-cbos8wxjy6d"><div class="d3-loc-growth"></div>
+<figure class="html-embed"><div class="html-embed__card"><div id="frag-s012olc2z3q"><div class="d3-loc-growth"></div>
 <style>
   .d3-loc-growth { position: relative; }
   
@@ -1446,49 +1523,63 @@ If you zoom in, you’ll notice there’s a sharp drop near the end, it’s esse
 <p>But this was not the only effort that allowed us to reduce maintenance load.</p>
 <p>We recently underwent a deep refactor of the attention implementation. You’ve likely heard about <a href="https://huggingface.co/docs/text-generation-inference/en/conceptual/flash_attention">flash attention</a> and its several variants.</p>
 <p>The <em>attention computation</em> itself happens at a <em>lower</em> level of abstraction than the model itself.</p>
-<p>However, we were adding specific torch operations for each backend (sdpa, the several flash-attention iterations, flex attention) but it wasn’t a <span class="glossary-term " style="" data-tooltip-id="tenet-m14tq94igj9" onmouseenter="window.showTenetTooltip(event, 'tenet-m14tq94igj9')" onmousemove="window.updateTenetTooltip(event, 'tenet-m14tq94igj9')" onmouseleave="window.hideTenetTooltip('tenet-m14tq94igj9')">minimal user api</span> <div id="tenet-m14tq94igj9" class="glossary-tooltip" data-position="top"> <div class="glossary-tooltip__content"> <div class="glossary-tooltip__term">minimal-user-api</div> <div class="glossary-tooltip__definition">Config, model, preprocessing; from_pretrained, save_pretrained, push_to_hub. Least amount of codepaths.</div> </div> <div class="glossary-tooltip__arrow"></div> </div> <script>
+<p>However, we were adding specific torch operations for each backend (sdpa, the several flash-attention iterations, flex attention) but it wasn’t a <span class="glossary-term " style="" data-tooltip-id="tenet-cdn8l7nt1os" onmouseenter="window.showTenetTooltip(event, 'tenet-cdn8l7nt1os')" onmousemove="window.updateTenetTooltip(event, 'tenet-cdn8l7nt1os')" onmouseleave="window.hideTenetTooltip('tenet-cdn8l7nt1os')">minimal user api</span><span id="tenet-cdn8l7nt1os" class="glossary-tooltip" data-position="top"> <span class="glossary-tooltip__content"> <span class="glossary-tooltip__term">minimal-user-api</span> <span class="glossary-tooltip__definition">Config, model, preprocessing; from_pretrained, save_pretrained, push_to_hub. Least amount of codepaths.</span> </span> <span class="glossary-tooltip__arrow"></span> </span> <script>
 if (!window.tenetTooltipInitialized) {
   window.tenetTooltipInitialized = true;
 
   window.showTenetTooltip = function(event, id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'block';
+    tooltip.style.visibility = 'visible';
     tooltip.style.opacity = '1';
     tooltip.style.top = (event.clientY + 10) + 'px';
     tooltip.style.left = (event.clientX + 10) + 'px';
   };
 
+  window.updateTenetTooltip = function(event, id) {
+    const tooltip = document.getElementById(id);
+    if (!tooltip) return;
+    tooltip.style.top = (event.clientY + 10) + 'px';
+    tooltip.style.left = (event.clientX + 10) + 'px';
+  };
+
   window.hideTenetTooltip = function(id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'none';
+    tooltip.style.visibility = 'hidden';
     tooltip.style.opacity = '0';
   };
 }
 </script>. Next section explains what we did.</p>
-<div class="crumbs"><p>Evidence: effective (i.e., maintainable) LOC growth drops ~15× when counting shards instead of expanded modeling files. Less code to read, fewer places to break.</p><p><strong>Next:</strong> how the attention interface stays standard without hiding semantics.</p></div>
+<div class="note note--info" data-astro-cid-qg6lmfty> <div class="note__layout" data-astro-cid-qg6lmfty>  <div class="note__body" data-astro-cid-qg6lmfty>  <div class="note__content" data-astro-cid-qg6lmfty> <p>Evidence: effective (i.e., maintainable) LOC growth drops ~15× when counting shards instead of expanded modeling files. Less code to read, fewer places to break.</p><p><strong>Next:</strong> how the attention interface stays standard without hiding semantics.</p> </div> </div> </div> </div> 
 <h3 id="-external-attention-classes"><a href="#-external-attention-classes"><a id="attention-classes"></a> External Attention classes</a></h3>
 <p>The solution for the “attention abstraction problem” was to move to a standard <a href="https://huggingface.co/docs/transformers/en/attention_interface">attention interface</a> that allows the following:</p>
 <p>The naive implementation of attention, called “eager”, is available by default. We use a <code>Callable</code> called <code>eager_attention_forward</code>, which can run as long as the user has PyTorch installed – which is a requirement any way.</p>
 <p>Instead of using a class interface and a class hierarchy, we just moved to a function interface. When a more complex attention implementation is needed, we use other Callables, including much faster kernel bindings when available. The decision to use a different attention implementation is based on the model configuration file we download from the Hub, and it can also be overridden by the user.</p>
-<p>This is a clear example that that we prefer an interface that is <span class="glossary-term " style="" data-tooltip-id="tenet-ihfmqnz2vi" onmouseenter="window.showTenetTooltip(event, 'tenet-ihfmqnz2vi')" onmousemove="window.updateTenetTooltip(event, 'tenet-ihfmqnz2vi')" onmouseleave="window.hideTenetTooltip('tenet-ihfmqnz2vi')">standard, but not abstract</span> <div id="tenet-ihfmqnz2vi" class="glossary-tooltip" data-position="top"> <div class="glossary-tooltip__content"> <div class="glossary-tooltip__term">standardize-dont-abstract</div> <div class="glossary-tooltip__definition">Model-specific logic belongs in the model file, not hidden behind abstractions.</div> </div> <div class="glossary-tooltip__arrow"></div> </div> <script>
+<p>This is a clear example that that we prefer an interface that is <span class="glossary-term " style="" data-tooltip-id="tenet-746jsxy9p34" onmouseenter="window.showTenetTooltip(event, 'tenet-746jsxy9p34')" onmousemove="window.updateTenetTooltip(event, 'tenet-746jsxy9p34')" onmouseleave="window.hideTenetTooltip('tenet-746jsxy9p34')">standard, but not abstract</span><span id="tenet-746jsxy9p34" class="glossary-tooltip" data-position="top"> <span class="glossary-tooltip__content"> <span class="glossary-tooltip__term">standardize-dont-abstract</span> <span class="glossary-tooltip__definition">Model-specific logic belongs in the model file, not hidden behind abstractions.</span> </span> <span class="glossary-tooltip__arrow"></span> </span> <script>
 if (!window.tenetTooltipInitialized) {
   window.tenetTooltipInitialized = true;
 
   window.showTenetTooltip = function(event, id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'block';
+    tooltip.style.visibility = 'visible';
     tooltip.style.opacity = '1';
     tooltip.style.top = (event.clientY + 10) + 'px';
     tooltip.style.left = (event.clientX + 10) + 'px';
   };
 
+  window.updateTenetTooltip = function(event, id) {
+    const tooltip = document.getElementById(id);
+    if (!tooltip) return;
+    tooltip.style.top = (event.clientY + 10) + 'px';
+    tooltip.style.left = (event.clientX + 10) + 'px';
+  };
+
   window.hideTenetTooltip = function(id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'none';
+    tooltip.style.visibility = 'hidden';
     tooltip.style.opacity = '0';
   };
 }
@@ -1497,46 +1588,60 @@ if (!window.tenetTooltipInitialized) {
 <span class="line"><span style="--shiki-light:#D73A49;--shiki-dark:#F97583">if</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF"> self</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8">.config._attn_implementation </span><span style="--shiki-light:#D73A49;--shiki-dark:#F97583">!=</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF"> &quot;eager&quot;</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8">:</span></span>
 <span class="line"><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8">    attention_interface </span><span style="--shiki-light:#D73A49;--shiki-dark:#F97583">=</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF"> ALL_ATTENTION_FUNCTIONS</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8">[</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF">self</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8">.config._attn_implementation]</span></span>
 <span class="line"></span></code></pre></div>
-<p>Having the attention interfaces functionalized allows to do dynamic switching of attentions as well, increasing their <span class="glossary-term " style="" data-tooltip-id="tenet-lufpuywh5gm" onmouseenter="window.showTenetTooltip(event, 'tenet-lufpuywh5gm')" onmousemove="window.updateTenetTooltip(event, 'tenet-lufpuywh5gm')" onmouseleave="window.hideTenetTooltip('tenet-lufpuywh5gm')">hackability</span> <div id="tenet-lufpuywh5gm" class="glossary-tooltip" data-position="top"> <div class="glossary-tooltip__content"> <div class="glossary-tooltip__term">code-is-product</div> <div class="glossary-tooltip__definition">Optimize for reading, diffing, and tweaking. Code quality matters as much as functionality.</div> </div> <div class="glossary-tooltip__arrow"></div> </div> <script>
+<p>Having the attention interfaces functionalized allows to do dynamic switching of attentions as well, increasing their <span class="glossary-term " style="" data-tooltip-id="tenet-p803817hgh" onmouseenter="window.showTenetTooltip(event, 'tenet-p803817hgh')" onmousemove="window.updateTenetTooltip(event, 'tenet-p803817hgh')" onmouseleave="window.hideTenetTooltip('tenet-p803817hgh')">hackability</span><span id="tenet-p803817hgh" class="glossary-tooltip" data-position="top"> <span class="glossary-tooltip__content"> <span class="glossary-tooltip__term">code-is-product</span> <span class="glossary-tooltip__definition">Optimize for reading, diffing, and tweaking. Code quality matters as much as functionality.</span> </span> <span class="glossary-tooltip__arrow"></span> </span> <script>
 if (!window.tenetTooltipInitialized) {
   window.tenetTooltipInitialized = true;
 
   window.showTenetTooltip = function(event, id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'block';
+    tooltip.style.visibility = 'visible';
     tooltip.style.opacity = '1';
     tooltip.style.top = (event.clientY + 10) + 'px';
     tooltip.style.left = (event.clientX + 10) + 'px';
   };
 
+  window.updateTenetTooltip = function(event, id) {
+    const tooltip = document.getElementById(id);
+    if (!tooltip) return;
+    tooltip.style.top = (event.clientY + 10) + 'px';
+    tooltip.style.left = (event.clientX + 10) + 'px';
+  };
+
   window.hideTenetTooltip = function(id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'none';
+    tooltip.style.visibility = 'hidden';
     tooltip.style.opacity = '0';
   };
 }
 </script>.
 Another strength of the new attention interface is the possibility to enforce specific kwargs, which are needed by kernel providers and other dependencies.</p>
 <p>Backend integrations sometimes require specific kwargs.</p>
-<p>We know that kwargs are often a necessary evil that plagues tools with widespread compatibility; and it is something we have aimed to reduce, and will continue reduce in order to improve readability - with them, the current system is a <span class="glossary-term " style="" data-tooltip-id="tenet-b2xym4pimri" onmouseenter="window.showTenetTooltip(event, 'tenet-b2xym4pimri')" onmousemove="window.updateTenetTooltip(event, 'tenet-b2xym4pimri')" onmouseleave="window.hideTenetTooltip('tenet-b2xym4pimri')">minimal user api</span> <div id="tenet-b2xym4pimri" class="glossary-tooltip" data-position="top"> <div class="glossary-tooltip__content"> <div class="glossary-tooltip__term">minimal-user-api</div> <div class="glossary-tooltip__definition">Config, model, preprocessing; from_pretrained, save_pretrained, push_to_hub. Least amount of codepaths.</div> </div> <div class="glossary-tooltip__arrow"></div> </div> <script>
+<p>We know that kwargs are often a necessary evil that plagues tools with widespread compatibility; and it is something we have aimed to reduce, and will continue reduce in order to improve readability - with them, the current system is a <span class="glossary-term " style="" data-tooltip-id="tenet-thjud1nhqb" onmouseenter="window.showTenetTooltip(event, 'tenet-thjud1nhqb')" onmousemove="window.updateTenetTooltip(event, 'tenet-thjud1nhqb')" onmouseleave="window.hideTenetTooltip('tenet-thjud1nhqb')">minimal user api</span><span id="tenet-thjud1nhqb" class="glossary-tooltip" data-position="top"> <span class="glossary-tooltip__content"> <span class="glossary-tooltip__term">minimal-user-api</span> <span class="glossary-tooltip__definition">Config, model, preprocessing; from_pretrained, save_pretrained, push_to_hub. Least amount of codepaths.</span> </span> <span class="glossary-tooltip__arrow"></span> </span> <script>
 if (!window.tenetTooltipInitialized) {
   window.tenetTooltipInitialized = true;
 
   window.showTenetTooltip = function(event, id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'block';
+    tooltip.style.visibility = 'visible';
     tooltip.style.opacity = '1';
     tooltip.style.top = (event.clientY + 10) + 'px';
     tooltip.style.left = (event.clientX + 10) + 'px';
   };
 
+  window.updateTenetTooltip = function(event, id) {
+    const tooltip = document.getElementById(id);
+    if (!tooltip) return;
+    tooltip.style.top = (event.clientY + 10) + 'px';
+    tooltip.style.left = (event.clientX + 10) + 'px';
+  };
+
   window.hideTenetTooltip = function(id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'none';
+    tooltip.style.visibility = 'hidden';
     tooltip.style.opacity = '0';
   };
 }
@@ -1546,30 +1651,37 @@ if (!window.tenetTooltipInitialized) {
 <span class="line"></span>
 <span class="line"><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8">MyModelOutputAnnotated </span><span style="--shiki-light:#D73A49;--shiki-dark:#F97583">=</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8"> Annotated[MyModelOutput, </span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF">&quot;shape: (B, C, H, W)&quot;</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8">]</span></span>
 <span class="line"></span></code></pre></div>
-<div class="crumbs"><p>Attention semantics remain in <code>eager_attention_forward</code>; faster backends are opt-in via config. We inform via types/annotations rather than enforce rigid kwargs, preserving integrations.</p><p><strong>Next:</strong> parallel partitioning is declared as a plan, not through model surgery.</p></div>
+<div class="note note--info" data-astro-cid-qg6lmfty> <div class="note__layout" data-astro-cid-qg6lmfty>  <div class="note__body" data-astro-cid-qg6lmfty>  <div class="note__content" data-astro-cid-qg6lmfty> <p>Attention semantics remain in <code>eager_attention_forward</code>; faster backends are opt-in via config. We inform via types/annotations rather than enforce rigid kwargs, preserving integrations.</p><p><strong>Next:</strong> parallel partitioning is declared as a plan, not through model surgery.</p> </div> </div> </div> </div> 
 <h3 id="-configurable-tensor-parallelism"><a href="#-configurable-tensor-parallelism"><a id="simpler-tensor-parallelism"></a> Configurable Tensor Parallelism</a></h3>
 <p>If you’re not familiar with the different flavours of parallelism, I recommend checking out <a href="https://huggingface.co/blog/accelerate-nd-parallel">this blog post</a> first, and of course a full <a href="https://huggingface.co/spaces/nanotron/ultrascale-playbook">dive into the ultra-scale playbook</a> is always recommended.</p>
 <p>The essential part is that, as <a href="https://huggingface.co/docs/transformers/v4.56.2/perf_train_gpu_many#tensor-parallelism">the documentation states</a>, when tensors get too large to fit on a single GPU, they are sliced along a particular dimension and every slice is sent to a different GPU.</p>
 <p>Why does it matter?</p>
 <p>Because we want to avoid code modifications that are unrelated to the model.</p>
 <p>We choose to place the level of abstraction higher than the device placement: a matrix multiplication - a <code>nn.Linear</code> layer - should be always expressed in the same way, regardless of how it is placed.</p>
-<p>Hence, we want to touch the modeling code <span class="glossary-term " style="" data-tooltip-id="tenet-7e15fxaemrq" onmouseenter="window.showTenetTooltip(event, 'tenet-7e15fxaemrq')" onmousemove="window.updateTenetTooltip(event, 'tenet-7e15fxaemrq')" onmouseleave="window.hideTenetTooltip('tenet-7e15fxaemrq')">as little as possible</span> <div id="tenet-7e15fxaemrq" class="glossary-tooltip" data-position="top"> <div class="glossary-tooltip__content"> <div class="glossary-tooltip__term">minimal-user-api</div> <div class="glossary-tooltip__definition">Config, model, preprocessing; from_pretrained, save_pretrained, push_to_hub. Least amount of codepaths.</div> </div> <div class="glossary-tooltip__arrow"></div> </div> <script>
+<p>Hence, we want to touch the modeling code <span class="glossary-term " style="" data-tooltip-id="tenet-t218cp1i4h" onmouseenter="window.showTenetTooltip(event, 'tenet-t218cp1i4h')" onmousemove="window.updateTenetTooltip(event, 'tenet-t218cp1i4h')" onmouseleave="window.hideTenetTooltip('tenet-t218cp1i4h')">as little as possible</span><span id="tenet-t218cp1i4h" class="glossary-tooltip" data-position="top"> <span class="glossary-tooltip__content"> <span class="glossary-tooltip__term">minimal-user-api</span> <span class="glossary-tooltip__definition">Config, model, preprocessing; from_pretrained, save_pretrained, push_to_hub. Least amount of codepaths.</span> </span> <span class="glossary-tooltip__arrow"></span> </span> <script>
 if (!window.tenetTooltipInitialized) {
   window.tenetTooltipInitialized = true;
 
   window.showTenetTooltip = function(event, id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'block';
+    tooltip.style.visibility = 'visible';
     tooltip.style.opacity = '1';
     tooltip.style.top = (event.clientY + 10) + 'px';
     tooltip.style.left = (event.clientX + 10) + 'px';
   };
 
+  window.updateTenetTooltip = function(event, id) {
+    const tooltip = document.getElementById(id);
+    if (!tooltip) return;
+    tooltip.style.top = (event.clientY + 10) + 'px';
+    tooltip.style.left = (event.clientX + 10) + 'px';
+  };
+
   window.hideTenetTooltip = function(id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'none';
+    tooltip.style.visibility = 'hidden';
     tooltip.style.opacity = '0';
   };
 }
@@ -1603,7 +1715,7 @@ if (!window.tenetTooltipInitialized) {
 <p>The <code>tp_plan</code> solution allows users to run the same model on a single GPU, or distribute it using multiple processes per node, e.g. 4 GPUs:</p>
 <p><code>torchrun --nproc-per-node 4 demo.py</code></p>
 <p>Semantics stay in the model (a Linear stays a Linear), parallelization is orthogonal and declared via strings: “colwise” splits columns of weights/bias across ranks; “rowwise” splits rows; packed variants shard fused weights; The mapping keys accept glob patterns like <code>layers.*.mlp.down_proj</code> to target repeated submodules.</p>
-<div class="crumbs"><p>Parallelization is specified in the configuration (<code>tp_plan</code>), not through edits to <code>Linear</code>s. Glob patterns target repeated blocks; modeling semantics stay intact.</p><p><strong>Next:</strong> per-layer attention/caching schedules declared in config, not hardcoded.</p></div>
+<div class="note note--info" data-astro-cid-qg6lmfty> <div class="note__layout" data-astro-cid-qg6lmfty>  <div class="note__body" data-astro-cid-qg6lmfty>  <div class="note__content" data-astro-cid-qg6lmfty> <p>Parallelization is specified in the configuration (<code>tp_plan</code>), not through edits to <code>Linear</code>s. Glob patterns target repeated blocks; modeling semantics stay intact.</p><p><strong>Next:</strong> per-layer attention/caching schedules declared in config, not hardcoded.</p> </div> </div> </div> </div> 
 <h3 id="-layers-attentions-and-caches"><a href="#-layers-attentions-and-caches"><a id="layers-attentions-caches"></a> Layers, attentions and caches</a></h3>
 <p>Following the same logic, the <em>nature</em> of attention and per-layer caching should not be hardcoded. We should be able to specify in the configuration how each layer is implemented. Thus, we define a mapping like:</p>
 <div class="code-card"><button class="code-copy button--ghost" type="button" aria-label="Copy code"><svg viewBox="0 0 24 24" aria-hidden="true" focusable="false"><path d="M16 1H4c-1.1 0-2 .9-2 2v12h2V3h12V1zm3 4H8c-1.1 0-2 .9-2 2v14c0 1.1.9 2 2 2h11c1.1 0 2-.9 2-2V7c0-1.1-.9-2-2-2zm0 16H8V7h11v14z"></path></svg></button><pre class="astro-code astro-code-themes github-light github-dark" style="--shiki-light:#24292e;--shiki-dark:#e1e4e8;--shiki-light-bg:#fff;--shiki-dark-bg:#24292e;overflow-x:auto" tabindex="0" data-language="python"><code><span class="line"><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF">ALLOWED_LAYER_TYPES</span><span style="--shiki-light:#D73A49;--shiki-dark:#F97583"> =</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8"> (</span></span>
@@ -1623,46 +1735,60 @@ if (!window.tenetTooltipInitialized) {
 <span class="line"><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF">    &quot;full_attention&quot;</span></span>
 <span class="line"><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8">  ],</span></span>
 <span class="line"></span></code></pre></div>
-<p>This is <span class="glossary-term " style="" data-tooltip-id="tenet-2jk5475q4v4" onmouseenter="window.showTenetTooltip(event, 'tenet-2jk5475q4v4')" onmousemove="window.updateTenetTooltip(event, 'tenet-2jk5475q4v4')" onmouseleave="window.hideTenetTooltip('tenet-2jk5475q4v4')">minimal</span> <div id="tenet-2jk5475q4v4" class="glossary-tooltip" data-position="top"> <div class="glossary-tooltip__content"> <div class="glossary-tooltip__term">minimal-user-api</div> <div class="glossary-tooltip__definition">Config, model, preprocessing; from_pretrained, save_pretrained, push_to_hub. Least amount of codepaths.</div> </div> <div class="glossary-tooltip__arrow"></div> </div> <script>
+<p>This is <span class="glossary-term " style="" data-tooltip-id="tenet-f8s3np3ge36" onmouseenter="window.showTenetTooltip(event, 'tenet-f8s3np3ge36')" onmousemove="window.updateTenetTooltip(event, 'tenet-f8s3np3ge36')" onmouseleave="window.hideTenetTooltip('tenet-f8s3np3ge36')">minimal</span><span id="tenet-f8s3np3ge36" class="glossary-tooltip" data-position="top"> <span class="glossary-tooltip__content"> <span class="glossary-tooltip__term">minimal-user-api</span> <span class="glossary-tooltip__definition">Config, model, preprocessing; from_pretrained, save_pretrained, push_to_hub. Least amount of codepaths.</span> </span> <span class="glossary-tooltip__arrow"></span> </span> <script>
 if (!window.tenetTooltipInitialized) {
   window.tenetTooltipInitialized = true;
 
   window.showTenetTooltip = function(event, id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'block';
+    tooltip.style.visibility = 'visible';
     tooltip.style.opacity = '1';
     tooltip.style.top = (event.clientY + 10) + 'px';
     tooltip.style.left = (event.clientX + 10) + 'px';
   };
 
+  window.updateTenetTooltip = function(event, id) {
+    const tooltip = document.getElementById(id);
+    if (!tooltip) return;
+    tooltip.style.top = (event.clientY + 10) + 'px';
+    tooltip.style.left = (event.clientX + 10) + 'px';
+  };
+
   window.hideTenetTooltip = function(id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'none';
+    tooltip.style.visibility = 'hidden';
     tooltip.style.opacity = '0';
   };
 }
 </script> to implement on the user side, and allows to keep the modeling code untouched. It is also easy to tweak.</p>
-<div class="crumbs"><p>Allowed layer types are explicit; schedules (e.g., sliding/full alternation) live in config. This keeps the file readable and easy to tweak.</p><p><strong>Next:</strong> speedups come from kernels that don’t change semantics.</p></div>
+<div class="note note--info" data-astro-cid-qg6lmfty> <div class="note__layout" data-astro-cid-qg6lmfty>  <div class="note__body" data-astro-cid-qg6lmfty>  <div class="note__content" data-astro-cid-qg6lmfty> <p>Allowed layer types are explicit; schedules (e.g., sliding/full alternation) live in config. This keeps the file readable and easy to tweak.</p><p><strong>Next:</strong> speedups come from kernels that don’t change semantics.</p> </div> </div> </div> </div> 
 <h3 id="community-kernels"><a href="#community-kernels"><a id="community-kernels"></a>Community Kernels</a></h3>
-<p>The same principle extends to normalization, activation, and other code paths. The model defines <strong>semantics</strong>; a kernel defines <strong>how</strong> to execute them faster. We annotate the module to borrow a community‑provided forward, keeping a <span class="glossary-term " style="" data-tooltip-id="tenet-s3h4h8h3m1" onmouseenter="window.showTenetTooltip(event, 'tenet-s3h4h8h3m1')" onmousemove="window.updateTenetTooltip(event, 'tenet-s3h4h8h3m1')" onmouseleave="window.hideTenetTooltip('tenet-s3h4h8h3m1')">consistent public surface</span> <div id="tenet-s3h4h8h3m1" class="glossary-tooltip" data-position="top"> <div class="glossary-tooltip__content"> <div class="glossary-tooltip__term">consistent-public-surface</div> <div class="glossary-tooltip__definition">Uniform naming, signatures, and conventions across all models for predictability.</div> </div> <div class="glossary-tooltip__arrow"></div> </div> <script>
+<p>The same principle extends to normalization, activation, and other code paths. The model defines <strong>semantics</strong>; a kernel defines <strong>how</strong> to execute them faster. We annotate the module to borrow a community‑provided forward, keeping a <span class="glossary-term " style="" data-tooltip-id="tenet-s24f7d2tlvk" onmouseenter="window.showTenetTooltip(event, 'tenet-s24f7d2tlvk')" onmousemove="window.updateTenetTooltip(event, 'tenet-s24f7d2tlvk')" onmouseleave="window.hideTenetTooltip('tenet-s24f7d2tlvk')">consistent public surface</span><span id="tenet-s24f7d2tlvk" class="glossary-tooltip" data-position="top"> <span class="glossary-tooltip__content"> <span class="glossary-tooltip__term">consistent-public-surface</span> <span class="glossary-tooltip__definition">Uniform naming, signatures, and conventions across all models for predictability.</span> </span> <span class="glossary-tooltip__arrow"></span> </span> <script>
 if (!window.tenetTooltipInitialized) {
   window.tenetTooltipInitialized = true;
 
   window.showTenetTooltip = function(event, id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'block';
+    tooltip.style.visibility = 'visible';
     tooltip.style.opacity = '1';
     tooltip.style.top = (event.clientY + 10) + 'px';
     tooltip.style.left = (event.clientX + 10) + 'px';
   };
 
+  window.updateTenetTooltip = function(event, id) {
+    const tooltip = document.getElementById(id);
+    if (!tooltip) return;
+    tooltip.style.top = (event.clientY + 10) + 'px';
+    tooltip.style.left = (event.clientX + 10) + 'px';
+  };
+
   window.hideTenetTooltip = function(id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'none';
+    tooltip.style.visibility = 'hidden';
     tooltip.style.opacity = '0';
   };
 }
@@ -1673,7 +1799,7 @@ if (!window.tenetTooltipInitialized) {
 <span class="line"></span></code></pre></div>
 <p>This also opens another contribution path: GPU specialists can contribute optimized kernels to the <a href="https://huggingface.co/kernels-community">Kernels Hub</a>, and have them immediately available to use in <code>transformers</code> and other libraries. You can check the <a href="https://huggingface.co/blog/hello-hf-kernels">kernel community blog post</a> to learn more about it!</p>
 <p>Even more resources have been added, like the formidable <a href="https://github.com/huggingface/kernel-builder">kernel builder</a> with its connected resources to <a href="https://github.com/huggingface/kernel-builder/blob/main/docs/writing-kernels.md">help you build kernels with it</a> and <a href="https://github.com/huggingface/kernel-builder/blob/main/docs/nix.md">with nix</a>.</p>
-<div class="crumbs"><p>Models define semantics; kernels define how to run them faster. Use decorations to borrow community forwards while keeping a consistent public surface.</p><p><strong>Next:</strong> what modularity looks like across the repo.</p></div>
+<div class="note note--info" data-astro-cid-qg6lmfty> <div class="note__layout" data-astro-cid-qg6lmfty>  <div class="note__body" data-astro-cid-qg6lmfty>  <div class="note__content" data-astro-cid-qg6lmfty> <p>Models define semantics; kernels define how to run them faster. Use decorations to borrow community forwards while keeping a consistent public surface.</p><p><strong>Next:</strong> what modularity looks like across the repo.</p> </div> </div> </div> </div> 
 <h2 id="a-modular-state"><a href="#a-modular-state">A Modular State</a></h2>
 <p>With <code>modular</code> transformers, we have a form of inheritance in our codebase. Some models become standards, and model contributors are given the opportunity to <em>define standards</em>. Pushing the boundaries of scientific knowledge can translate into the boundaries of engineering if this effort is made, and we’re striving for it.
 It’s hard to conceptualize very large libraries and how their components interact with each other, regardless of your cognitive abilities for abstractions.
@@ -1687,7 +1813,7 @@ So I wanted to take a look at the current <strong>state of modularity</strong> a
 <p>So what do we see?</p>
 <p>(Graph reading guide: nodes are models; edges are modular imports).</p>
 <p>Check out the <a href="https://huggingface.co/spaces/Molbap/transformers-modular-refactor">full viewer here</a> (tab “dependency graph”, hit “build graph”) for better manipulation and exploration.</p>
-<div class="wide"> <figure class="html-embed"><div class="html-embed__card is-frameless"><div id="frag-5autqbmguxy"><iframe
+<div class="wide"> <figure class="html-embed"><div class="html-embed__card is-frameless"><div id="frag-b1723qfa3gk"><iframe
 src="https://molbap-dependencies-1.hf.space"
 style="width:100%; height:680px; border:0"
 allow="clipboard-read; clipboard-write; fullscreen"
@@ -1695,7 +1821,7 @@ referrerpolicy="no-referrer-when-downgrade"
 ></iframe></div></div></figure> </div>
 <p>Let’s walk through some sections of this graph together.
 First, Llama is a basis and an influence for many models, and it is very visible.</p>
-<div class="ri-root" data-ri-root="ri_fm8e6evn6jg" data-has-caption data-astro-cid-6kov3kig> <figure class="" data-astro-cid-6kov3kig> <img src="/_astro/llama_center.CbQ5MyAc_ZraNCd.webp" alt="Llama in the center" data-zoomable="1" data-astro-cid-6kov3kig width="1030" height="1015" loading="lazy" decoding="async"> <figcaption data-astro-cid-6kov3kig> <span data-astro-cid-6kov3kig><strong>Figure 2:</strong> Llama as a central model influencing many other models in the dependency graph.</span> </figcaption> </figure> </div> <script>
+<div class="ri-root" data-ri-root="ri_1445klmzacc8" data-has-caption data-astro-cid-6kov3kig> <figure class="" data-astro-cid-6kov3kig> <img src="/_astro/llama_center.CbQ5MyAc_ZraNCd.webp" alt="Llama in the center" data-zoomable="1" data-astro-cid-6kov3kig width="1030" height="1015" loading="lazy" decoding="async"> <figcaption data-astro-cid-6kov3kig> <span data-astro-cid-6kov3kig><strong>Figure 2:</strong> Llama as a central model influencing many other models in the dependency graph.</span> </figcaption> </figure> </div> <script>
   (() => {
     const scriptEl = document.currentScript;
     const root = scriptEl ? scriptEl.previousElementSibling : null;
@@ -1864,7 +1990,7 @@ First, Llama is a basis and an influence for many models, and it is very visible
 </script> 
 <p>The models linked sometimes pull components from other models than <code>llama</code> of course. Radically different architectures such as mamba have spawned their own dependency subgraph.</p>
 <p>Audio models form sparser archipelagos, see for instance wav2vec2 which is a significant basis for a dozen of them.</p>
-<div class="ri-root" data-ri-root="ri_ulqlj4un8rp" data-has-caption data-astro-cid-6kov3kig> <figure class="" data-astro-cid-6kov3kig> <img src="/_astro/cluster_wave2vec2.BvHBUP61_Z2vdkmW.webp" alt="Wav2vec2 influence" data-zoomable="1" data-astro-cid-6kov3kig width="608" height="563" loading="lazy" decoding="async"> <figcaption data-astro-cid-6kov3kig> <span data-astro-cid-6kov3kig><strong>Figure 3:</strong> Cluster of audio architectures based on wav2vec2, forming a specialized archipelago.</span> </figcaption> </figure> </div> <script>
+<div class="ri-root" data-ri-root="ri_9maqwb8m6q4" data-has-caption data-astro-cid-6kov3kig> <figure class="" data-astro-cid-6kov3kig> <img src="/_astro/cluster_wave2vec2.BvHBUP61_Z2vdkmW.webp" alt="Wav2vec2 influence" data-zoomable="1" data-astro-cid-6kov3kig width="608" height="563" loading="lazy" decoding="async"> <figcaption data-astro-cid-6kov3kig> <span data-astro-cid-6kov3kig><strong>Figure 3:</strong> Cluster of audio architectures based on wav2vec2, forming a specialized archipelago.</span> </figcaption> </figure> </div> <script>
   (() => {
     const scriptEl = document.currentScript;
     const root = scriptEl ? scriptEl.previousElementSibling : null;
@@ -2033,7 +2159,7 @@ First, Llama is a basis and an influence for many models, and it is very visible
 </script> 
 <p>In the case of VLMs which have massively grown in popularity since 2024, there’s far too many vision-based architectures that are not yet defined as modulars of other existing archs. In other words, there is no strong reference point in terms of software for vision models.</p>
 <p>As you can see, there is a small <code>DETR</code> island:</p>
-<div class="ri-root" data-ri-root="ri_ny7zk1os7a" data-has-caption data-astro-cid-6kov3kig> <figure class="" data-astro-cid-6kov3kig> <img src="/_astro/detr_island.CSrqELWy_1IAFDR.webp" alt="DETR archipelago" data-zoomable="1" data-astro-cid-6kov3kig width="591" height="606" loading="lazy" decoding="async"> <figcaption data-astro-cid-6kov3kig> <span data-astro-cid-6kov3kig><strong>Figure 4:</strong> Small DETR archipelago for vision models, less centralized than Llama for text.</span> </figcaption> </figure> </div> <script>
+<div class="ri-root" data-ri-root="ri_f82ss0hm4q9" data-has-caption data-astro-cid-6kov3kig> <figure class="" data-astro-cid-6kov3kig> <img src="/_astro/detr_island.CSrqELWy_1IAFDR.webp" alt="DETR archipelago" data-zoomable="1" data-astro-cid-6kov3kig width="591" height="606" loading="lazy" decoding="async"> <figcaption data-astro-cid-6kov3kig> <span data-astro-cid-6kov3kig><strong>Figure 4:</strong> Small DETR archipelago for vision models, less centralized than Llama for text.</span> </figcaption> </figure> </div> <script>
   (() => {
     const scriptEl = document.currentScript;
     const root = scriptEl ? scriptEl.previousElementSibling : null;
@@ -2202,7 +2328,7 @@ First, Llama is a basis and an influence for many models, and it is very visible
 </script> 
 <p>There is also a little llava pocket, and so on, but it’s not comparable to the centrality observed for llama.</p>
 <p>Another problem is, this visualization only shows <code>modular</code> models. Several models still do NOT have a modular file. If we zoom out significantly, we can see them,  the red nodes are models that do not have a modular file yet.</p>
-<div class="ri-root" data-ri-root="ri_prp4zvqaxh" data-has-caption data-astro-cid-6kov3kig> <figure class="" data-astro-cid-6kov3kig> <img src="/_astro/big_picture_zoomout.BKwXtSkj_1bNS6U.webp" alt="Red nodes" data-zoomable="1" data-astro-cid-6kov3kig width="1043" height="972" loading="lazy" decoding="async"> <figcaption data-astro-cid-6kov3kig> <span data-astro-cid-6kov3kig><strong>Figure 5:</strong> Overview showing red nodes (models without modular files) to be modularized.</span> </figcaption> </figure> </div> <script>
+<div class="ri-root" data-ri-root="ri_mob9j97bstt" data-has-caption data-astro-cid-6kov3kig> <figure class="" data-astro-cid-6kov3kig> <img src="/_astro/big_picture_zoomout.BKwXtSkj_1bNS6U.webp" alt="Red nodes" data-zoomable="1" data-astro-cid-6kov3kig width="1043" height="972" loading="lazy" decoding="async"> <figcaption data-astro-cid-6kov3kig> <span data-astro-cid-6kov3kig><strong>Figure 5:</strong> Overview showing red nodes (models without modular files) to be modularized.</span> </figcaption> </figure> </div> <script>
   (() => {
     const scriptEl = document.currentScript;
     const root = scriptEl ? scriptEl.previousElementSibling : null;
@@ -2370,14 +2496,14 @@ First, Llama is a basis and an influence for many models, and it is very visible
   })();
 </script> 
 <p>Hence the next question, and how do we identify modularisable models?</p>
-<div class="crumbs"><p>Llama-lineage is a hub; several VLMs remain islands — engineering opportunity for shared parents.
-<strong>Next:</strong> timeline + similarity signals to spot modularisable candidates.</p></div>
+<div class="note note--info" data-astro-cid-qg6lmfty> <div class="note__layout" data-astro-cid-qg6lmfty>  <div class="note__body" data-astro-cid-qg6lmfty>  <div class="note__content" data-astro-cid-qg6lmfty> <p>Llama-lineage is a hub; several VLMs remain islands — engineering opportunity for shared parents.
+<strong>Next:</strong> timeline + similarity signals to spot modularisable candidates.</p> </div> </div> </div> </div> 
 <h3 id="many-models-but-not-enough-yet-are-alike"><a href="#many-models-but-not-enough-yet-are-alike">Many models, but not enough yet, are alike</a></h3>
 <p>I looked into Jaccard similarity, which we use to measure set differences, to find similarities across models. I know that code is more than a set of characters stringed together. We also tried code-embedding models that ranked candidates better in practice, but for this post we stick to the deterministic Jaccard index.</p>
 <p>It is interesting, for our comparison, to look at <em>when</em> we deployed the modular logic and what was its rippling effect on the library. Looking at the timeline makes it obvious: adding modular allowed to connect more and more models to solid reference points.</p>
 <p>Yet, we still have a lot of gaps to fill.</p>
 <p>Zoom out below - it’s full of models. You can click on a node to see its connections better, or use the text box to search for a model. You can use the <a href="https://huggingface.co/spaces/Molbap/transformers-modular-refactor">full viewer</a> (tab “timeline”, hit “build timeline”) for better exploration.</p>
-<figure class="html-embed"><div class="html-embed__card"><div id="frag-y2kybecguv">    <iframe
+<figure class="html-embed"><div class="html-embed__card"><div id="frag-4lvnnmk52m">    <iframe
   src="https://molbap-timeline-1.hf.space"
   style="width:100%; height:680px; border:0"
   allow="clipboard-read; clipboard-write; fullscreen"
@@ -2386,14 +2512,14 @@ First, Llama is a basis and an influence for many models, and it is very visible
 <p>Let’s look at a few highly connected models. Let’s start by the foundational work of <a href="https://arxiv.org/abs/2304.08485">Llava</a>.</p>
 <p><img src="/images/transformers/timeline_llava.png" alt="Llava in its timeline"/></p>
 <p>You see that <code>llava_video</code> is a red node, connected by a red edge to <code>llava</code>: it’s a candidate, something that we can <em>likely</em> remodularize, <a href="#backwards-compatibility">not touching the actual model</a> but being much more readable with <a href="#do-repeat-yourself">DRY*</a>.</p>
-<div class="wide"> <figure class="html-embed"><div class="html-embed__card is-frameless"><div id="frag-8g32f0mx2la">    <iframe
+<div class="wide"> <figure class="html-embed"><div class="html-embed__card is-frameless"><div id="frag-dt3p4vasw7">    <iframe
   src="https://molbap-timeline-1.hf.space"
   style="width:100%; height:680px; border:0"
   allow="clipboard-read; clipboard-write; fullscreen"
   referrerpolicy="no-referrer-when-downgrade"
 ></iframe></div></div></figure> </div>
 <p>Let’s look at a few highly connected models. Let’s start by the foundational work of <a href="https://arxiv.org/abs/2304.08485">Llava</a>.</p>
-<div class="ri-root" data-ri-root="ri_rwnefegiw8r" data-has-caption data-astro-cid-6kov3kig> <figure class="" data-astro-cid-6kov3kig> <img src="/_astro/timeline_llava.Bne5RSo9_Z26WEKX.webp" alt="Llava in its timeline" data-zoomable="1" data-astro-cid-6kov3kig width="1250" height="770" loading="lazy" decoding="async"> <figcaption data-astro-cid-6kov3kig> <span data-astro-cid-6kov3kig><strong>Figure 6:</strong> LLaVA and its variants in the timeline, with llava_video as a candidate for modularization.</span> </figcaption> </figure> </div> <script>
+<div class="ri-root" data-ri-root="ri_mhcdv4nmsit" data-has-caption data-astro-cid-6kov3kig> <figure class="" data-astro-cid-6kov3kig> <img src="/_astro/timeline_llava.Bne5RSo9_Z26WEKX.webp" alt="Llava in its timeline" data-zoomable="1" data-astro-cid-6kov3kig width="1250" height="770" loading="lazy" decoding="async"> <figcaption data-astro-cid-6kov3kig> <span data-astro-cid-6kov3kig><strong>Figure 6:</strong> LLaVA and its variants in the timeline, with llava_video as a candidate for modularization.</span> </figcaption> </figure> </div> <script>
   (() => {
     const scriptEl = document.currentScript;
     const root = scriptEl ? scriptEl.previousElementSibling : null;
@@ -2560,50 +2686,64 @@ First, Llama is a basis and an influence for many models, and it is very visible
     else window.addEventListener("load", initZoomIfNeeded, { once: true });
   })();
 </script> 
-<p>You see that <code>llava_video</code> is a red node, connected by a red edge to <code>llava</code>: it’s a candidate, something that we can <em>likely</em> remodularize, <span class="glossary-term " style="" data-tooltip-id="tenet-zyhdqoqw7t" onmouseenter="window.showTenetTooltip(event, 'tenet-zyhdqoqw7t')" onmousemove="window.updateTenetTooltip(event, 'tenet-zyhdqoqw7t')" onmouseleave="window.hideTenetTooltip('tenet-zyhdqoqw7t')">not touching the actual model</span> <div id="tenet-zyhdqoqw7t" class="glossary-tooltip" data-position="top"> <div class="glossary-tooltip__content"> <div class="glossary-tooltip__term">backwards-compatibility</div> <div class="glossary-tooltip__definition">Any artifact once on the hub must remain loadable. Breaking changes are unacceptable.</div> </div> <div class="glossary-tooltip__arrow"></div> </div> <script>
+<p>You see that <code>llava_video</code> is a red node, connected by a red edge to <code>llava</code>: it’s a candidate, something that we can <em>likely</em> remodularize, <span class="glossary-term " style="" data-tooltip-id="tenet-eueeprhlylq" onmouseenter="window.showTenetTooltip(event, 'tenet-eueeprhlylq')" onmousemove="window.updateTenetTooltip(event, 'tenet-eueeprhlylq')" onmouseleave="window.hideTenetTooltip('tenet-eueeprhlylq')">not touching the actual model</span><span id="tenet-eueeprhlylq" class="glossary-tooltip" data-position="top"> <span class="glossary-tooltip__content"> <span class="glossary-tooltip__term">backwards-compatibility</span> <span class="glossary-tooltip__definition">Any artifact once on the hub must remain loadable. Breaking changes are unacceptable.</span> </span> <span class="glossary-tooltip__arrow"></span> </span> <script>
 if (!window.tenetTooltipInitialized) {
   window.tenetTooltipInitialized = true;
 
   window.showTenetTooltip = function(event, id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'block';
+    tooltip.style.visibility = 'visible';
     tooltip.style.opacity = '1';
     tooltip.style.top = (event.clientY + 10) + 'px';
     tooltip.style.left = (event.clientX + 10) + 'px';
   };
 
+  window.updateTenetTooltip = function(event, id) {
+    const tooltip = document.getElementById(id);
+    if (!tooltip) return;
+    tooltip.style.top = (event.clientY + 10) + 'px';
+    tooltip.style.left = (event.clientX + 10) + 'px';
+  };
+
   window.hideTenetTooltip = function(id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'none';
+    tooltip.style.visibility = 'hidden';
     tooltip.style.opacity = '0';
   };
 }
-</script> but being much more readable with <span class="glossary-term " style="" data-tooltip-id="tenet-wfae1791doq" onmouseenter="window.showTenetTooltip(event, 'tenet-wfae1791doq')" onmousemove="window.updateTenetTooltip(event, 'tenet-wfae1791doq')" onmouseleave="window.hideTenetTooltip('tenet-wfae1791doq')">DRY*</span> <div id="tenet-wfae1791doq" class="glossary-tooltip" data-position="top"> <div class="glossary-tooltip__content"> <div class="glossary-tooltip__term">do-repeat-yourself</div> <div class="glossary-tooltip__definition">Strategic duplication can improve readability and maintainability when done thoughtfully.</div> </div> <div class="glossary-tooltip__arrow"></div> </div> <script>
+</script> but being much more readable with <span class="glossary-term " style="" data-tooltip-id="tenet-t90kgb3bxqi" onmouseenter="window.showTenetTooltip(event, 'tenet-t90kgb3bxqi')" onmousemove="window.updateTenetTooltip(event, 'tenet-t90kgb3bxqi')" onmouseleave="window.hideTenetTooltip('tenet-t90kgb3bxqi')">DRY*</span><span id="tenet-t90kgb3bxqi" class="glossary-tooltip" data-position="top"> <span class="glossary-tooltip__content"> <span class="glossary-tooltip__term">do-repeat-yourself</span> <span class="glossary-tooltip__definition">Strategic duplication can improve readability and maintainability when done thoughtfully.</span> </span> <span class="glossary-tooltip__arrow"></span> </span> <script>
 if (!window.tenetTooltipInitialized) {
   window.tenetTooltipInitialized = true;
 
   window.showTenetTooltip = function(event, id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'block';
+    tooltip.style.visibility = 'visible';
     tooltip.style.opacity = '1';
     tooltip.style.top = (event.clientY + 10) + 'px';
     tooltip.style.left = (event.clientX + 10) + 'px';
   };
 
+  window.updateTenetTooltip = function(event, id) {
+    const tooltip = document.getElementById(id);
+    if (!tooltip) return;
+    tooltip.style.top = (event.clientY + 10) + 'px';
+    tooltip.style.left = (event.clientX + 10) + 'px';
+  };
+
   window.hideTenetTooltip = function(id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'none';
+    tooltip.style.visibility = 'hidden';
     tooltip.style.opacity = '0';
   };
 }
 </script>.</p>
 <p>The same can be identified with the classical encoders family, centered on <code>BERT</code>:</p>
 <p>Here <code>roberta</code>, <code>xlm_roberta</code>, <code>ernie</code> are <code>modular</code>s of BERT, while models like <code>mobilebert</code> are likely candidates.</p>
-<div class="ri-root" data-ri-root="ri_a9jh8di7qbr" data-has-caption data-astro-cid-6kov3kig> <figure class="" data-astro-cid-6kov3kig> <img src="/_astro/classic_encoders.BSgQl9lp_3OtlT.webp" alt="Classical encoders" data-zoomable="1" data-astro-cid-6kov3kig width="1274" height="749" loading="lazy" decoding="async"> <figcaption data-astro-cid-6kov3kig> <span data-astro-cid-6kov3kig><strong>Figure 7:</strong> Family of classical encoders centered on BERT, with several models already modularized.</span> </figcaption> </figure> </div> <script>
+<div class="ri-root" data-ri-root="ri_8h24f1ku9ub" data-has-caption data-astro-cid-6kov3kig> <figure class="" data-astro-cid-6kov3kig> <img src="/_astro/classic_encoders.BSgQl9lp_3OtlT.webp" alt="Classical encoders" data-zoomable="1" data-astro-cid-6kov3kig width="1274" height="749" loading="lazy" decoding="async"> <figcaption data-astro-cid-6kov3kig> <span data-astro-cid-6kov3kig><strong>Figure 7:</strong> Family of classical encoders centered on BERT, with several models already modularized.</span> </figcaption> </figure> </div> <script>
   (() => {
     const scriptEl = document.currentScript;
     const root = scriptEl ? scriptEl.previousElementSibling : null;
@@ -2770,37 +2910,44 @@ if (!window.tenetTooltipInitialized) {
     else window.addEventListener("load", initZoomIfNeeded, { once: true });
   })();
 </script> 
-<div class="crumbs"><p>Similarity metrics (Jaccard index or embeddings) surfaces likely parents; the timeline shows consolidation after modular landed. Red nodes/edges = candidates (e.g., <code>llava_video</code> → <code>llava</code>) for refactors that preserve behavior.</p><p><strong>Next:</strong> concrete VLM choices that avoid leaky abstractions.</p></div>
+<div class="note note--info" data-astro-cid-qg6lmfty> <div class="note__layout" data-astro-cid-qg6lmfty>  <div class="note__body" data-astro-cid-qg6lmfty>  <div class="note__content" data-astro-cid-qg6lmfty> <p>Similarity metrics (Jaccard index or embeddings) surfaces likely parents; the timeline shows consolidation after modular landed. Red nodes/edges = candidates (e.g., <code>llava_video</code> → <code>llava</code>) for refactors that preserve behavior.</p><p><strong>Next:</strong> concrete VLM choices that avoid leaky abstractions.</p> </div> </div> </div> </div> 
 <h3 id="vlm-improvements-avoiding-abstraction"><a href="#vlm-improvements-avoiding-abstraction">VLM improvements, avoiding abstraction</a></h3>
 <p>We don’t yet have a cookbook for common VLM patterns (image token scatter, multi‑tower encoders, cross‑attention bridges). This is one of the main improvement points where we can work.</p>
 <p>For instance, we thought of abstracting away the mixing of <code>inputs_embeds</code>, the tensor fed into an LLM decoder in 95% of the existing VLMs. It would have looked like something like</p>
 <div class="code-card"><button class="code-copy button--ghost" type="button" aria-label="Copy code"><svg viewBox="0 0 24 24" aria-hidden="true" focusable="false"><path d="M16 1H4c-1.1 0-2 .9-2 2v12h2V3h12V1zm3 4H8c-1.1 0-2 .9-2 2v14c0 1.1.9 2 2 2h11c1.1 0 2-.9 2-2V7c0-1.1-.9-2-2-2zm0 16H8V7h11v14z"></path></svg></button><pre class="astro-code astro-code-themes github-light github-dark" style="--shiki-light:#24292e;--shiki-dark:#e1e4e8;--shiki-light-bg:#fff;--shiki-dark-bg:#24292e;overflow-x:auto" tabindex="0" data-language="python"><code><span class="line"><span style="--shiki-light:#D73A49;--shiki-dark:#F97583">class</span><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0"> InputsEmbeddingMixerMixin</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8">(</span><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0">nn</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8">.</span><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0">Module</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8">):</span></span>
 <span class="line"><span style="--shiki-light:#6A737D;--shiki-dark:#6A737D">    #</span></span>
 <span class="line"></span></code></pre></div>
-<p>But this is <span class="glossary-term " style="" data-tooltip-id="tenet-slh7hv7t3vc" onmouseenter="window.showTenetTooltip(event, 'tenet-slh7hv7t3vc')" onmousemove="window.updateTenetTooltip(event, 'tenet-slh7hv7t3vc')" onmouseleave="window.hideTenetTooltip('tenet-slh7hv7t3vc')">not an abstraction</span> <div id="tenet-slh7hv7t3vc" class="glossary-tooltip" data-position="top"> <div class="glossary-tooltip__content"> <div class="glossary-tooltip__term">standardize-dont-abstract</div> <div class="glossary-tooltip__definition">Model-specific logic belongs in the model file, not hidden behind abstractions.</div> </div> <div class="glossary-tooltip__arrow"></div> </div> <script>
+<p>But this is <span class="glossary-term " style="" data-tooltip-id="tenet-4anl1xbq9aj" onmouseenter="window.showTenetTooltip(event, 'tenet-4anl1xbq9aj')" onmousemove="window.updateTenetTooltip(event, 'tenet-4anl1xbq9aj')" onmouseleave="window.hideTenetTooltip('tenet-4anl1xbq9aj')">not an abstraction</span><span id="tenet-4anl1xbq9aj" class="glossary-tooltip" data-position="top"> <span class="glossary-tooltip__content"> <span class="glossary-tooltip__term">standardize-dont-abstract</span> <span class="glossary-tooltip__definition">Model-specific logic belongs in the model file, not hidden behind abstractions.</span> </span> <span class="glossary-tooltip__arrow"></span> </span> <script>
 if (!window.tenetTooltipInitialized) {
   window.tenetTooltipInitialized = true;
 
   window.showTenetTooltip = function(event, id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'block';
+    tooltip.style.visibility = 'visible';
     tooltip.style.opacity = '1';
     tooltip.style.top = (event.clientY + 10) + 'px';
     tooltip.style.left = (event.clientX + 10) + 'px';
   };
 
+  window.updateTenetTooltip = function(event, id) {
+    const tooltip = document.getElementById(id);
+    if (!tooltip) return;
+    tooltip.style.top = (event.clientY + 10) + 'px';
+    tooltip.style.left = (event.clientX + 10) + 'px';
+  };
+
   window.hideTenetTooltip = function(id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'none';
+    tooltip.style.visibility = 'hidden';
     tooltip.style.opacity = '0';
   };
 }
 </script>. Embedding mixin is part of the model, removing it would break it. A user opening <a href="https://github.com/huggingface/transformers/blob/b3bd815786c36f4e6c3791fae0a96cac86658b32/src/transformers/models/qwen2_5_vl/modeling_qwen2_5_vl.py#L1358"><code>modeling_qwen2.5_vl</code></a> (check out the <a href="https://huggingface.co/collections/Qwen/qwen25-vl-6795ffac22b334a837c0f9a5">Qwen2.5VL collection</a>) should not have to go to another file to understand how it works.</p>
 <p>What is the current state of these “abstractions” across the codebase?
 You will see all the imports around a modeling file, here <a href="https://huggingface.co/google/gemma-3n-E4B-it">Gemma3n</a>.</p>
-<div class="ri-root" data-ri-root="ri_mo02stgyay" data-has-caption data-astro-cid-6kov3kig> <figure class="" data-astro-cid-6kov3kig> <img src="/_astro/still_graph_bloat.BII6Am4a_Z2d5KVT.webp" alt="Gemma3n graph" data-zoomable="1" data-astro-cid-6kov3kig width="2580" height="1207" loading="lazy" decoding="async"> <figcaption data-astro-cid-6kov3kig> <span data-astro-cid-6kov3kig><strong>Figure 8:</strong> Gemma3n import graph showing dependency complexity, with GenerationMixin very central.</span> </figcaption> </figure> </div> <script>
+<div class="ri-root" data-ri-root="ri_6kbqnjxpbit" data-has-caption data-astro-cid-6kov3kig> <figure class="" data-astro-cid-6kov3kig> <img src="/_astro/still_graph_bloat.BII6Am4a_Z2d5KVT.webp" alt="Gemma3n graph" data-zoomable="1" data-astro-cid-6kov3kig width="2580" height="1207" loading="lazy" decoding="async"> <figcaption data-astro-cid-6kov3kig> <span data-astro-cid-6kov3kig><strong>Figure 8:</strong> Gemma3n import graph showing dependency complexity, with GenerationMixin very central.</span> </figcaption> </figure> </div> <script>
   (() => {
     const scriptEl = document.currentScript;
     const root = scriptEl ? scriptEl.previousElementSibling : null;
@@ -3010,34 +3157,41 @@ That means every decision we make to abstract something else has to be extremely
 <span class="line"></span>
 <span class="line"><span style="--shiki-light:#D73A49;--shiki-dark:#F97583">        return</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8"> special_image_mask, special_video_mask</span></span>
 <span class="line"></span></code></pre></div>
-<p>But this is <em>within</em> the modeling file, not in the <code>PreTrainedModel</code> base class. It will not move away from it, because it’d break the <span class="glossary-term " style="" data-tooltip-id="tenet-sixn0w2maq" onmouseenter="window.showTenetTooltip(event, 'tenet-sixn0w2maq')" onmousemove="window.updateTenetTooltip(event, 'tenet-sixn0w2maq')" onmouseleave="window.hideTenetTooltip('tenet-sixn0w2maq')">One model, one file tenet.</span> <div id="tenet-sixn0w2maq" class="glossary-tooltip" data-position="top"> <div class="glossary-tooltip__content"> <div class="glossary-tooltip__term">one-model-one-file</div> <div class="glossary-tooltip__definition">All inference and training core logic visible, top‑to‑bottom, in a single file.</div> </div> <div class="glossary-tooltip__arrow"></div> </div> <script>
+<p>But this is <em>within</em> the modeling file, not in the <code>PreTrainedModel</code> base class. It will not move away from it, because it’d break the <span class="glossary-term " style="" data-tooltip-id="tenet-xxo4wgd6wpk" onmouseenter="window.showTenetTooltip(event, 'tenet-xxo4wgd6wpk')" onmousemove="window.updateTenetTooltip(event, 'tenet-xxo4wgd6wpk')" onmouseleave="window.hideTenetTooltip('tenet-xxo4wgd6wpk')">One model, one file tenet.</span><span id="tenet-xxo4wgd6wpk" class="glossary-tooltip" data-position="top"> <span class="glossary-tooltip__content"> <span class="glossary-tooltip__term">one-model-one-file</span> <span class="glossary-tooltip__definition">All inference and training core logic visible, top‑to‑bottom, in a single file.</span> </span> <span class="glossary-tooltip__arrow"></span> </span> <script>
 if (!window.tenetTooltipInitialized) {
   window.tenetTooltipInitialized = true;
 
   window.showTenetTooltip = function(event, id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'block';
+    tooltip.style.visibility = 'visible';
     tooltip.style.opacity = '1';
     tooltip.style.top = (event.clientY + 10) + 'px';
     tooltip.style.left = (event.clientX + 10) + 'px';
   };
 
+  window.updateTenetTooltip = function(event, id) {
+    const tooltip = document.getElementById(id);
+    if (!tooltip) return;
+    tooltip.style.top = (event.clientY + 10) + 'px';
+    tooltip.style.left = (event.clientX + 10) + 'px';
+  };
+
   window.hideTenetTooltip = function(id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'none';
+    tooltip.style.visibility = 'hidden';
     tooltip.style.opacity = '0';
   };
 }
 </script></p>
 <p>What do we conclude? Going forward, we should aim for VLMs to have a form of centrality similar to that of <code>Llama</code> for text-only models. This centrality should not be achieved at the cost of abstracting and hiding away crucial inner workings of said models.</p>
-<div class="crumbs"><p>Keep VLM embedding mix in the modeling file (semantics), standardize safe helpers (e.g., placeholder masking), don’t migrate behavior to <code>PreTrainedModel</code>.
-<strong>Next:</strong> pipeline-level wins that came from PyTorch-first choices (fast processors).</p></div>
+<div class="note note--info" data-astro-cid-qg6lmfty> <div class="note__layout" data-astro-cid-qg6lmfty>  <div class="note__body" data-astro-cid-qg6lmfty>  <div class="note__content" data-astro-cid-qg6lmfty> <p>Keep VLM embedding mix in the modeling file (semantics), standardize safe helpers (e.g., placeholder masking), don’t migrate behavior to <code>PreTrainedModel</code>.
+<strong>Next:</strong> pipeline-level wins that came from PyTorch-first choices (fast processors).</p> </div> </div> </div> </div> 
 <h3 id="on-image-processing-and-processors"><a href="#on-image-processing-and-processors">On image processing and processors</a></h3>
 <p>Deciding to become a <code>torch</code>-first library meant relieving a tremendous amount of support for <code>jax </code> and <code>TensorFlow</code>, and it also meant that we could be more lenient about the amount of torch-dependent utilities that we were able to accept. One of these is the <em>fast processing</em> of images. Where inputs were once minimally assumed to be ndarrays, enforcing native <code>torch</code> and <code>torchvision</code> inputs allowed us to massively improve processing speed for each model.</p>
 <p>The gains in performance are immense, up to 20x speedup for most models when using compiled torchvision ops. Furthermore, lets us run the whole pipeline solely on GPU.</p>
-<div class="ri-root" data-ri-root="ri_jnrbiya92x8" data-has-caption data-astro-cid-6kov3kig> <figure class="" data-astro-cid-6kov3kig> <img src="/_astro/fast_image_processors.D3x5vY3o_2cacGa.webp" alt="Fast Image Processors Performance" data-zoomable="1" data-astro-cid-6kov3kig width="2251" height="2409" loading="lazy" decoding="async"> <figcaption data-astro-cid-6kov3kig> <span data-astro-cid-6kov3kig><strong>Figure 9:</strong> Performance gains of fast image processors, up to 20x acceleration with compiled torchvision.</span> </figcaption> </figure> </div> <script>
+<div class="ri-root" data-ri-root="ri_rx173sswef" data-has-caption data-astro-cid-6kov3kig> <figure class="" data-astro-cid-6kov3kig> <img src="/_astro/fast_image_processors.D3x5vY3o_2cacGa.webp" alt="Fast Image Processors Performance" data-zoomable="1" data-astro-cid-6kov3kig width="2251" height="2409" loading="lazy" decoding="async"> <figcaption data-astro-cid-6kov3kig> <span data-astro-cid-6kov3kig><strong>Figure 9:</strong> Performance gains of fast image processors, up to 20x acceleration with compiled torchvision.</span> </figcaption> </figure> </div> <script>
   (() => {
     const scriptEl = document.currentScript;
     const root = scriptEl ? scriptEl.previousElementSibling : null;
@@ -3204,17 +3358,17 @@ if (!window.tenetTooltipInitialized) {
     else window.addEventListener("load", initZoomIfNeeded, { once: true });
   })();
 </script> 
-<div class="crumbs"><p>PyTorch-first lets processors assume torch/torchvision and run the whole pipeline on GPU; big per-model speedups.</p><p><strong>Next:</strong> how this lowers friction for contributors and downstream users.</p></div>
+<div class="note note--info" data-astro-cid-qg6lmfty> <div class="note__layout" data-astro-cid-qg6lmfty>  <div class="note__body" data-astro-cid-qg6lmfty>  <div class="note__content" data-astro-cid-qg6lmfty> <p>PyTorch-first lets processors assume torch/torchvision and run the whole pipeline on GPU; big per-model speedups.</p><p><strong>Next:</strong> how this lowers friction for contributors and downstream users.</p> </div> </div> </div> </div> 
 <h2 id="reduce-barrier-to-entrycontribution"><a href="#reduce-barrier-to-entrycontribution">Reduce barrier to entry/contribution</a></h2>
 <p>This is an overall objective: there’s no <code>transformers</code> without its community.</p>
 <p>Having a framework means forcing users into it. It restrains flexibility and creativity, which are the fertile soil for new ideas to grow.</p>
 <p>Among the most valuable contributions to <code>transformers</code> is of course the addition of new models. Very recently, <a href="https://huggingface.co/blog/welcome-openai-gpt-oss">OpenAI added GPT-OSS</a>, which prompted the addition of many new features to the library in order to support <a href="https://huggingface.co/openai/gpt-oss-120b">their model</a>.</p>
 <p>These additions are immediately available for other models to use.</p>
 <p>Another important advantage is the ability to fine-tune and pipeline these models into many other libraries and tools. Check here on the hub how many finetunes are registered for <a href="https://huggingface.co/models?other=base_model:finetune:openai/gpt-oss-120b">gpt-oss 120b</a>, despite its size!</p>
-<div class="crumbs"><p>The shape of a contribution: add a model (or variant) with a small modular shard; the community and serving stacks pick it up immediately. Popularity trends (encoders/embeddings) guide where we invest.</p><p><strong>Next:</strong> power tools enabled by a consistent API.</p></div>
+<div class="note note--info" data-astro-cid-qg6lmfty> <div class="note__layout" data-astro-cid-qg6lmfty>  <div class="note__body" data-astro-cid-qg6lmfty>  <div class="note__content" data-astro-cid-qg6lmfty> <p>The shape of a contribution: add a model (or variant) with a small modular shard; the community and serving stacks pick it up immediately. Popularity trends (encoders/embeddings) guide where we invest.</p><p><strong>Next:</strong> power tools enabled by a consistent API.</p> </div> </div> </div> </div> 
 <h3 id="-models-popularity"><a href="#-models-popularity"><a id="encoders-ftw"></a> Models popularity</a></h3>
 <p>Talking about dependencies, we can take a look at the number of downloads as a measure of popularity. One thing we see is the prominence of encoders, despite the apparent prevalence of decoder LLMs. The reason is that encoders are used to generate embeddings, which have multiple downstream uses. Just check out <a href="https://huggingface.co/blog/embeddinggemma">EmbeddingGemma</a> for a modern recap. Hence, it is vital to keep the encoders portion of the library viable, usable, fine-tunable.</p>
-<div><figure class="html-embed"><div class="html-embed__card"><div id="frag-leyu459dlea"><div class="d3-model-popularity"></div>
+<div><figure class="html-embed"><div class="html-embed__card"><div id="frag-ijhi82q1p9j"><div class="d3-model-popularity"></div>
 <style>
   .d3-model-popularity .controls {
     margin-top: 0;
@@ -3790,36 +3944,43 @@ if (!window.tenetTooltipInitialized) {
 </script>
 </div></div></figure></div>
 <p>As the codebase grows, we need to maintain it in coordination with our friend <a href="https://huggingface.co/sentence-transformers">Sentence Transformers codebase</a>. Retrieval use-cases, smart databases, FAISS-based indexing rely on it, and thus indirectly on transformers.</p>
-<p>In that regard, we DO want to be a modular toolbox, being <span class="glossary-term " style="" data-tooltip-id="tenet-21zirrhepxi" onmouseenter="window.showTenetTooltip(event, 'tenet-21zirrhepxi')" onmousemove="window.updateTenetTooltip(event, 'tenet-21zirrhepxi')" onmouseleave="window.hideTenetTooltip('tenet-21zirrhepxi')">minimal</span> <div id="tenet-21zirrhepxi" class="glossary-tooltip" data-position="top"> <div class="glossary-tooltip__content"> <div class="glossary-tooltip__term">minimal-user-api</div> <div class="glossary-tooltip__definition">Config, model, preprocessing; from_pretrained, save_pretrained, push_to_hub. Least amount of codepaths.</div> </div> <div class="glossary-tooltip__arrow"></div> </div> <script>
+<p>In that regard, we DO want to be a modular toolbox, being <span class="glossary-term " style="" data-tooltip-id="tenet-qi4jgueuog" onmouseenter="window.showTenetTooltip(event, 'tenet-qi4jgueuog')" onmousemove="window.updateTenetTooltip(event, 'tenet-qi4jgueuog')" onmouseleave="window.hideTenetTooltip('tenet-qi4jgueuog')">minimal</span><span id="tenet-qi4jgueuog" class="glossary-tooltip" data-position="top"> <span class="glossary-tooltip__content"> <span class="glossary-tooltip__term">minimal-user-api</span> <span class="glossary-tooltip__definition">Config, model, preprocessing; from_pretrained, save_pretrained, push_to_hub. Least amount of codepaths.</span> </span> <span class="glossary-tooltip__arrow"></span> </span> <script>
 if (!window.tenetTooltipInitialized) {
   window.tenetTooltipInitialized = true;
 
   window.showTenetTooltip = function(event, id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'block';
+    tooltip.style.visibility = 'visible';
     tooltip.style.opacity = '1';
     tooltip.style.top = (event.clientY + 10) + 'px';
     tooltip.style.left = (event.clientX + 10) + 'px';
   };
 
+  window.updateTenetTooltip = function(event, id) {
+    const tooltip = document.getElementById(id);
+    if (!tooltip) return;
+    tooltip.style.top = (event.clientY + 10) + 'px';
+    tooltip.style.left = (event.clientX + 10) + 'px';
+  };
+
   window.hideTenetTooltip = function(id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'none';
+    tooltip.style.visibility = 'hidden';
     tooltip.style.opacity = '0';
   };
 }
 </script> enough and well documented enough so any ML/AI developer can use <code>transformers</code> without having to think about it. We aim to reduce the cognitive load brought about by model development, not increase it.</p>
 <p>So, how do these design choices, these “tenets” influence development of models and overall usage of transformers?</p>
-<div class="crumbs"><p>Encoders remain critical for embeddings and retrieval; maintaining them well benefits the broader ecosystem (e.g., Sentence Transformers, FAISS).</p><p><strong>Next:</strong> dev tools that leverage unified attention APIs and PyTorch-only internals.</p></div>
+<div class="note note--info" data-astro-cid-qg6lmfty> <div class="note__layout" data-astro-cid-qg6lmfty>  <div class="note__body" data-astro-cid-qg6lmfty>  <div class="note__content" data-astro-cid-qg6lmfty> <p>Encoders remain critical for embeddings and retrieval; maintaining them well benefits the broader ecosystem (e.g., Sentence Transformers, FAISS).</p><p><strong>Next:</strong> dev tools that leverage unified attention APIs and PyTorch-only internals.</p> </div> </div> </div> </div> 
 <h2 id="a-surgical-toolbox-for-model-development"><a href="#a-surgical-toolbox-for-model-development">A surgical toolbox for model development</a></h2>
 <p>Transformers provides many tools that can help you add a new architecture, understand the inner workings of a model, as well as the library itself.</p>
 <h3 id="attention-visualisation"><a href="#attention-visualisation">Attention visualisation</a></h3>
 <p>All models have the same API for attention computation, thanks to <a href="#external-attention-classes">the externalisation of attention classes</a>.</p>
 <p>This uniformity allows us to build cool tools to visualize the inner workings of the attention mechanism.</p>
 <p>One particular piece of machinery is the <code>attention mask</code>. Here you see the famous bidirectional attention pattern for the whole prefix (text + image) in PaliGemma and all Gemma2+ models, contrasting with the usual “causal-only” models.</p>
-<figure class="html-embed"><div class="html-embed__card is-frameless"><div id="frag-ad4cdirj61j"><!-- Minimal HTML fragment: terminal-style ASCII attention masks -->
+<figure class="html-embed"><div class="html-embed__card is-frameless"><div id="frag-v0j9g591b1s"><!-- Minimal HTML fragment: terminal-style ASCII attention masks -->
 <div style="max-width: 940px; margin: 16px 0; border:1px solid #2a2f3a; border-radius:8px; background:#0b0f19; font-family: ui-monospace, SFMono-Regular, Menlo, Monaco, Consolas, 'Liberation Mono', 'Courier New', monospace; color:#e5e7eb;">
     <div style="display:flex; align-items:center; gap:8px; padding:8px 10px; border-bottom:1px solid #1f2430; background:#111827; border-top-left-radius:8px; border-top-right-radius:8px;">
       <span style="width:10px; height:10px; background:#ef4444; border-radius:50%; display:inline-block;"></span>
@@ -3864,31 +4025,38 @@ if (!window.tenetTooltipInitialized) {
     </div>
   </div>
   </div></div></figure>
-<div class="crumbs"><p>Uniform attention APIs enable cross-model diagnostics (e.g., PaliGemma prefix bidirectionality vs causal).</p><p><strong>Next:</strong> whole-model tracing for ports and regressions.</p></div>
+<div class="note note--info" data-astro-cid-qg6lmfty> <div class="note__layout" data-astro-cid-qg6lmfty>  <div class="note__body" data-astro-cid-qg6lmfty>  <div class="note__content" data-astro-cid-qg6lmfty> <p>Uniform attention APIs enable cross-model diagnostics (e.g., PaliGemma prefix bidirectionality vs causal).</p><p><strong>Next:</strong> whole-model tracing for ports and regressions.</p> </div> </div> </div> </div> 
 <h3 id="logging-entire-model-activations"><a href="#logging-entire-model-activations">Logging entire model activations</a></h3>
 <p>Because everything is PyTorch, we can easily <a href="https://huggingface.co/docs/transformers/internal/model_debugging_utils">debug any model</a> when we want to add it to transformers. We now have a power-user tool for porting or adding models, that wraps a forward pass, intercepts every submodule call, and logs shapes, dtypes, and sample statistics of inputs/outputs to nested JSON.</p>
-<p>It just works with PyTorch models and is especially useful when aligning outputs with a reference implementation, to match our <span class="glossary-term " style="" data-tooltip-id="tenet-vc5drnxawh" onmouseenter="window.showTenetTooltip(event, 'tenet-vc5drnxawh')" onmousemove="window.updateTenetTooltip(event, 'tenet-vc5drnxawh')" onmouseleave="window.hideTenetTooltip('tenet-vc5drnxawh')">Source of Truth guideline</span> <div id="tenet-vc5drnxawh" class="glossary-tooltip" data-position="top"> <div class="glossary-tooltip__content"> <div class="glossary-tooltip__term">source-of-truth</div> <div class="glossary-tooltip__definition">Model implementations should be reliable, reproducible, and faithful to original performances.</div> </div> <div class="glossary-tooltip__arrow"></div> </div> <script>
+<p>It just works with PyTorch models and is especially useful when aligning outputs with a reference implementation, to match our <span class="glossary-term " style="" data-tooltip-id="tenet-er2ghf48bft" onmouseenter="window.showTenetTooltip(event, 'tenet-er2ghf48bft')" onmousemove="window.updateTenetTooltip(event, 'tenet-er2ghf48bft')" onmouseleave="window.hideTenetTooltip('tenet-er2ghf48bft')">Source of Truth guideline</span><span id="tenet-er2ghf48bft" class="glossary-tooltip" data-position="top"> <span class="glossary-tooltip__content"> <span class="glossary-tooltip__term">source-of-truth</span> <span class="glossary-tooltip__definition">Model implementations should be reliable, reproducible, and faithful to original performances.</span> </span> <span class="glossary-tooltip__arrow"></span> </span> <script>
 if (!window.tenetTooltipInitialized) {
   window.tenetTooltipInitialized = true;
 
   window.showTenetTooltip = function(event, id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'block';
+    tooltip.style.visibility = 'visible';
     tooltip.style.opacity = '1';
     tooltip.style.top = (event.clientY + 10) + 'px';
     tooltip.style.left = (event.clientX + 10) + 'px';
   };
 
+  window.updateTenetTooltip = function(event, id) {
+    const tooltip = document.getElementById(id);
+    if (!tooltip) return;
+    tooltip.style.top = (event.clientY + 10) + 'px';
+    tooltip.style.left = (event.clientX + 10) + 'px';
+  };
+
   window.hideTenetTooltip = function(id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'none';
+    tooltip.style.visibility = 'hidden';
     tooltip.style.opacity = '0';
   };
 }
 </script>.</p>
-<div class="wide"> <div class="ri-root" data-ri-root="ri_3pv5h75z5gr" data-has-caption data-astro-cid-6kov3kig> <figure class="" data-astro-cid-6kov3kig> <img src="/_astro/model_debugger.DouWEpKv_Z2338YE.webp" alt="Model debugger interface" data-zoomable="1" data-astro-cid-6kov3kig width="2053" height="1016" loading="lazy" decoding="async"> <figcaption data-astro-cid-6kov3kig> <span data-astro-cid-6kov3kig><strong>Figure 10:</strong> Model debugger interface intercepting calls and logging statistics in nested JSON.</span> </figcaption> </figure> </div> <script>
+<div class="wide"> <div class="ri-root" data-ri-root="ri_j6stw3cuel" data-has-caption data-astro-cid-6kov3kig> <figure class="" data-astro-cid-6kov3kig> <img src="/_astro/model_debugger.DouWEpKv_Z2338YE.webp" alt="Model debugger interface" data-zoomable="1" data-astro-cid-6kov3kig width="2053" height="1016" loading="lazy" decoding="async"> <figcaption data-astro-cid-6kov3kig> <span data-astro-cid-6kov3kig><strong>Figure 10:</strong> Model debugger interface intercepting calls and logging statistics in nested JSON.</span> </figcaption> </figure> </div> <script>
   (() => {
     const scriptEl = document.currentScript;
     const root = scriptEl ? scriptEl.previousElementSibling : null;
@@ -4055,30 +4223,37 @@ if (!window.tenetTooltipInitialized) {
     else window.addEventListener("load", initZoomIfNeeded, { once: true });
   })();
 </script>  </div>
-<div class="crumbs"><p>Forward interception and nested JSON logging align ports to reference implementations, reinforcing “Source of Truth.”</p><p><strong>Next:</strong> CUDA warmup reduces load-time without touching modeling semantics.</p></div>
+<div class="note note--info" data-astro-cid-qg6lmfty> <div class="note__layout" data-astro-cid-qg6lmfty>  <div class="note__body" data-astro-cid-qg6lmfty>  <div class="note__content" data-astro-cid-qg6lmfty> <p>Forward interception and nested JSON logging align ports to reference implementations, reinforcing “Source of Truth.”</p><p><strong>Next:</strong> CUDA warmup reduces load-time without touching modeling semantics.</p> </div> </div> </div> </div> 
 <h3 id="cooking-faster-cuda-warmups"><a href="#cooking-faster-cuda-warmups">Cooking faster CUDA warmups</a></h3>
-<p>Having a clean <em>external</em> API allows us to work on the <span class="glossary-term " style="" data-tooltip-id="tenet-qxu1gsp64zh" onmouseenter="window.showTenetTooltip(event, 'tenet-qxu1gsp64zh')" onmousemove="window.updateTenetTooltip(event, 'tenet-qxu1gsp64zh')" onmouseleave="window.hideTenetTooltip('tenet-qxu1gsp64zh')">true inner workings of transformers</span> <div id="tenet-qxu1gsp64zh" class="glossary-tooltip" data-position="top"> <div class="glossary-tooltip__content"> <div class="glossary-tooltip__term">code-is-product</div> <div class="glossary-tooltip__definition">Optimize for reading, diffing, and tweaking. Code quality matters as much as functionality.</div> </div> <div class="glossary-tooltip__arrow"></div> </div> <script>
+<p>Having a clean <em>external</em> API allows us to work on the <span class="glossary-term " style="" data-tooltip-id="tenet-8jeyv5epfja" onmouseenter="window.showTenetTooltip(event, 'tenet-8jeyv5epfja')" onmousemove="window.updateTenetTooltip(event, 'tenet-8jeyv5epfja')" onmouseleave="window.hideTenetTooltip('tenet-8jeyv5epfja')">true inner workings of transformers</span><span id="tenet-8jeyv5epfja" class="glossary-tooltip" data-position="top"> <span class="glossary-tooltip__content"> <span class="glossary-tooltip__term">code-is-product</span> <span class="glossary-tooltip__definition">Optimize for reading, diffing, and tweaking. Code quality matters as much as functionality.</span> </span> <span class="glossary-tooltip__arrow"></span> </span> <script>
 if (!window.tenetTooltipInitialized) {
   window.tenetTooltipInitialized = true;
 
   window.showTenetTooltip = function(event, id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'block';
+    tooltip.style.visibility = 'visible';
     tooltip.style.opacity = '1';
     tooltip.style.top = (event.clientY + 10) + 'px';
     tooltip.style.left = (event.clientX + 10) + 'px';
   };
 
+  window.updateTenetTooltip = function(event, id) {
+    const tooltip = document.getElementById(id);
+    if (!tooltip) return;
+    tooltip.style.top = (event.clientY + 10) + 'px';
+    tooltip.style.left = (event.clientX + 10) + 'px';
+  };
+
   window.hideTenetTooltip = function(id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'none';
+    tooltip.style.visibility = 'hidden';
     tooltip.style.opacity = '0';
   };
 }
 </script>. One of a few recent additions was the <em>CUDA warmup</em> via <code>caching_allocator_warmup</code>, which dramatically improved loading times by pre-allocating GPU memory to avoid malloc bottlenecks during model loading. It can achieve a 7x speedup factor for an 8B model, or 6x for a 32B one, as you can check in <a href="https://github.com/huggingface/transformers/pull/36380">the PR</a>!</p>
-<div class="wide"> <figure class="html-embed"><figcaption class="html-embed__title" style="text-align:center">Mem allocation patterns during model loading</figcaption><div class="html-embed__card"><div id="frag-pps6k0rfrrh"><div class="d3-warmup-demo"></div>
+<div class="wide"> <figure class="html-embed"><figcaption class="html-embed__title" style="text-align:center">Mem allocation patterns during model loading</figcaption><div class="html-embed__card"><div id="frag-foo33i56iho"><div class="d3-warmup-demo"></div>
 <style>
   .d3-warmup-demo {
     font-family: 'Segoe UI', Tahoma, Geneva, Verdana, sans-serif;
@@ -4705,7 +4880,7 @@ if (!window.tenetTooltipInitialized) {
 </script>
 </div></div></figure> </div>
 <p>It’s hard to overstate how much of a lifesaver that is when you’re trying to load a model as fast as possible, as it’s the narrowest bottleneck for your iteration speed.</p>
-<div class="crumbs"><p>Pre-allocating GPU memory removes malloc spikes (e.g., 7× for 8B, 6× for 32B in the referenced PR).</p><p><strong>Next:</strong> consistent interfaces allow transformers-serve.</p></div>
+<div class="note note--info" data-astro-cid-qg6lmfty> <div class="note__layout" data-astro-cid-qg6lmfty>  <div class="note__body" data-astro-cid-qg6lmfty>  <div class="note__content" data-astro-cid-qg6lmfty> <p>Pre-allocating GPU memory removes malloc spikes (e.g., 7× for 8B, 6× for 32B in the referenced PR).</p><p><strong>Next:</strong> consistent interfaces allow transformers-serve.</p> </div> </div> </div> </div> 
 <h3 id="transformers-serve-and-continuous-batching"><a href="#transformers-serve-and-continuous-batching">Transformers-serve and continuous batching</a></h3>
 <p>Having all these models readily available and sharing the same interface allowed us to implement transformers-serve, a CLI tool to expose models through a standard OpenAI http API.</p>
 <div class="code-card"><button class="code-copy button--ghost" type="button" aria-label="Copy code"><svg viewBox="0 0 24 24" aria-hidden="true" focusable="false"><path d="M16 1H4c-1.1 0-2 .9-2 2v12h2V3h12V1zm3 4H8c-1.1 0-2 .9-2 2v14c0 1.1.9 2 2 2h11c1.1 0 2-.9 2-2V7c0-1.1-.9-2-2-2zm0 16H8V7h11v14z"></path></svg></button><pre class="astro-code astro-code-themes github-light github-dark" style="--shiki-light:#24292e;--shiki-dark:#e1e4e8;--shiki-light-bg:#fff;--shiki-dark-bg:#24292e;overflow-x:auto" tabindex="0" data-language="bash"><code><span class="line"><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0">transformers</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF"> serve</span></span>
@@ -4722,7 +4897,7 @@ if (!window.tenetTooltipInitialized) {
 <li>Run evaluations efficiently, again without having to spend a lot of time engineering your infrastructure.</li>
 </ul>
 <p>For model deployment, check <a href="https://huggingface.co/docs/inference-providers/en/index">Inference Providers</a> or roll your solution using any of the excellent serving libraries.</p>
-<div class="crumbs"><p>OpenAI-compatible surface + continuous batching; kernels/backends slot in because the modeling API stayed stable.</p><p><strong>Next:</strong> reuse across vLLM/SGLang relies on the same consistency.</p></div>
+<div class="note note--info" data-astro-cid-qg6lmfty> <div class="note__layout" data-astro-cid-qg6lmfty>  <div class="note__body" data-astro-cid-qg6lmfty>  <div class="note__content" data-astro-cid-qg6lmfty> <p>OpenAI-compatible surface + continuous batching; kernels/backends slot in because the modeling API stayed stable.</p><p><strong>Next:</strong> reuse across vLLM/SGLang relies on the same consistency.</p> </div> </div> </div> </div> 
 <h2 id="community-reusability"><a href="#community-reusability">Community reusability</a></h2>
 <p>The transformers-serve CLI built on transformers, for sure, but the library is made first and foremost to be <em>reused</em> at large by the open-source ecosystem.</p>
 <p>Adding a model to transformers means:</p>
@@ -4731,28 +4906,35 @@ if (!window.tenetTooltipInitialized) {
 <li>having it immediately usable in vLLM, <a href="https://huggingface.co/blog/transformers-backend-sglang">SGLang</a>, and so on without additional code. In the case of vLLM, transformers was added as a backend to run models on vLLM, which optimizes throughput/latency on top of <em>existing</em> transformers architectures <a href="https://blog.vllm.ai/2025/04/11/transformers-backend.html">as seen in this great vLLM x HF blog post.</a></li>
 <li>being the reference code for implementations in MLX, llama.cpp and other libraries.</li>
 </ul>
-<p>This further cements the need for a <span class="glossary-term " style="" data-tooltip-id="tenet-dg8oof5xy0q" onmouseenter="window.showTenetTooltip(event, 'tenet-dg8oof5xy0q')" onmousemove="window.updateTenetTooltip(event, 'tenet-dg8oof5xy0q')" onmouseleave="window.hideTenetTooltip('tenet-dg8oof5xy0q')">consistent public surface</span> <div id="tenet-dg8oof5xy0q" class="glossary-tooltip" data-position="top"> <div class="glossary-tooltip__content"> <div class="glossary-tooltip__term">consistent-public-surface</div> <div class="glossary-tooltip__definition">Uniform naming, signatures, and conventions across all models for predictability.</div> </div> <div class="glossary-tooltip__arrow"></div> </div> <script>
+<p>This further cements the need for a <span class="glossary-term " style="" data-tooltip-id="tenet-a3q2oj2fej8" onmouseenter="window.showTenetTooltip(event, 'tenet-a3q2oj2fej8')" onmousemove="window.updateTenetTooltip(event, 'tenet-a3q2oj2fej8')" onmouseleave="window.hideTenetTooltip('tenet-a3q2oj2fej8')">consistent public surface</span><span id="tenet-a3q2oj2fej8" class="glossary-tooltip" data-position="top"> <span class="glossary-tooltip__content"> <span class="glossary-tooltip__term">consistent-public-surface</span> <span class="glossary-tooltip__definition">Uniform naming, signatures, and conventions across all models for predictability.</span> </span> <span class="glossary-tooltip__arrow"></span> </span> <script>
 if (!window.tenetTooltipInitialized) {
   window.tenetTooltipInitialized = true;
 
   window.showTenetTooltip = function(event, id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'block';
+    tooltip.style.visibility = 'visible';
     tooltip.style.opacity = '1';
     tooltip.style.top = (event.clientY + 10) + 'px';
     tooltip.style.left = (event.clientX + 10) + 'px';
   };
 
+  window.updateTenetTooltip = function(event, id) {
+    const tooltip = document.getElementById(id);
+    if (!tooltip) return;
+    tooltip.style.top = (event.clientY + 10) + 'px';
+    tooltip.style.left = (event.clientX + 10) + 'px';
+  };
+
   window.hideTenetTooltip = function(id) {
     const tooltip = document.getElementById(id);
     if (!tooltip) return;
-    tooltip.style.display = 'none';
+    tooltip.style.visibility = 'hidden';
     tooltip.style.opacity = '0';
   };
 }
 </script>: we are a backend and a reference, and there’s more software than us to handle serving. At the time of writing, more effort is done in that direction. We already have compatible configs for VLMs for vLLM (say that three times fast), check <a href="https://github.com/huggingface/transformers/pull/40696/files">here for GLM4 video support</a>, and here for <a href="https://github.com/huggingface/transformers/pull/40132">MoE support</a>, for instance.</p>
-<div class="crumbs"><p>Being a good backend consumer requires a consistent public surface; modular shards and configs make that stability practical.</p><p><strong>Next:</strong> what changes in v5 without breaking the promise of visible semantics.</p></div>
+<div class="note note--info" data-astro-cid-qg6lmfty> <div class="note__layout" data-astro-cid-qg6lmfty>  <div class="note__body" data-astro-cid-qg6lmfty>  <div class="note__content" data-astro-cid-qg6lmfty> <p>Being a good backend consumer requires a consistent public surface; modular shards and configs make that stability practical.</p><p><strong>Next:</strong> what changes in v5 without breaking the promise of visible semantics.</p> </div> </div> </div> </div> 
 <h2 id="what-is-coming-next"><a href="#what-is-coming-next">What is coming next</a></h2>
 <p>The next major version of <code>transformers</code> is just around the corner (and will have another blog post to its name when it comes out). When v5 is released, we aim to keep <a href="#backwards-compatibility">backwards compatibility</a> as solid as possible. The changes we make now are in service of that goal.</p>
 <p>We will lean further into a modular toolbox, not a framework. You should not be forced to rewrite modeling code. It’s better when a model can inherit from <code>PreTrainedModel</code> and opt into Tensor Parallel, <code>from_pretrained</code>, sharding, <code>push_to_hub</code>, loss plumbing, and external stacks like PEFT/TRL/SGLang/vLLM.</p> </main> </section> <footer class="footer"> <div class="footer-inner"> <section class="citation-block"> <h3>Citation</h3> <p>For attribution, cite this work as</p> <pre class="citation short">Pablo Montalvo, Lysandre Debut, Pedro Cuenca, Yoni Gozlan (2025). &quot;Maintain the unmaintainable: 1M python loc, 400+ models&quot;.</pre> <p>BibTeX citation</p> <pre class="citation long">@misc{montalvo2025_maintain_the_unmaintaina,
@@ -4760,8 +4942,7 @@ if (!window.tenetTooltipInitialized) {
   author={Pablo Montalvo and Lysandre Debut and Pedro Cuenca and Yoni Gozlan},
   year={2025},
   
-}</pre> </section> <section class="acknowledgements-block"> <h3>Acknowledgements</h3> <p>Special thanks to all the reviewers on this! <a href='https://huggingface.co/reach-vb '>Vaibhav Srivastav</a>, <a href='https://huggingface.co/cyrilvallez '>Cyril Vallez</a>, <a href='https://huggingface.co/yonigozlan'>Yoni Gozlan</a> also for his excellent work on fast image processors, <a href='https://huggingface.co/ArthurZ'>Arthur Zucker</a> for his guidance, and of course the wonderful <a href='https://huggingface.co/tfrere'>Thibaud Frere</a> for designing this template and helping me out with it!
-Most importantly: thanks to the entire Open-Source community, sincerely.</p> </section>   <section class="references-block">  </section> </div> </footer> <script>
+}</pre> </section> <section class="acknowledgements-block"> <h3>Acknowledgements</h3> <p>Special thanks to all the reviewers on this! <a href='https://huggingface.co/reach-vb '>Vaibhav Srivastav</a>, <a href='https://huggingface.co/cyrilvallez '>Cyril Vallez</a>, <a href='https://huggingface.co/yonigozlan'>Yoni Gozlan</a> also for his excellent work on fast image processors, <a href='https://huggingface.co/ArthurZ'>Arthur Zucker</a> for his guidance, and of course the wonderful <a href='https://huggingface.co/tfrere'>Thibaud Frere</a> for designing this template and helping me out with it!<br><br>Most importantly: thanks to the entire Open-Source community, sincerely.</p> </section>   <section class="references-block">  </section> </div> </footer> <script>
   (() => {
     const getFooter = () => document.currentScript?.closest('footer') || document.querySelector('footer.footer');
     const footer = getFooter();
diff --git a/app/dist/index.html.gz b/app/dist/index.html.gz
index eb43d02446303e051a0142785c04d96e2466fc94..77074f9c9f542806aed9087f1ec3f43108e0eb53 100644
--- a/app/dist/index.html.gz
+++ b/app/dist/index.html.gz
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:37ea57936a220876b21506e54a0fd9b034c0a8f35cd65c3a062de0f2d65a576d
-size 63568
+oid sha256:32a6ba4be66d4b9424a952b557d1564e5adf5d7a6d2c70927200c1d3bfb7b86f
+size 64076
diff --git a/app/src/content/article.mdx b/app/src/content/article.mdx
index a0c0fb9a13e1a9b5c7b373f37c5141521dcd49ef..45ea6328e1b951a480b58e5ed7bce68e9731834d 100644
--- a/app/src/content/article.mdx
+++ b/app/src/content/article.mdx
@@ -21,9 +21,7 @@ affiliations:
 published: "October 2, 2025"
 tags: [transformers, engineering, design-philosophy]
 tableOfContentsAutoCollapse: true
-acknowledgements: "Special thanks to all the reviewers on this! <a href='https://huggingface.co/reach-vb '>Vaibhav Srivastav</a>, <a href='https://huggingface.co/cyrilvallez '>Cyril Vallez</a>, <a href='https://huggingface.co/yonigozlan'>Yoni Gozlan</a> also for his excellent work on fast image processors, <a href='https://huggingface.co/ArthurZ'>Arthur Zucker</a> for his guidance, and of course the wonderful <a href='https://huggingface.co/tfrere'>Thibaud Frere</a> for designing this template and helping me out with it!
-
-Most importantly: thanks to the entire Open-Source community, sincerely."
+acknowledgements: "Special thanks to all the reviewers on this! <a href='https://huggingface.co/reach-vb '>Vaibhav Srivastav</a>, <a href='https://huggingface.co/cyrilvallez '>Cyril Vallez</a>, <a href='https://huggingface.co/yonigozlan'>Yoni Gozlan</a> also for his excellent work on fast image processors, <a href='https://huggingface.co/ArthurZ'>Arthur Zucker</a> for his guidance, and of course the wonderful <a href='https://huggingface.co/tfrere'>Thibaud Frere</a> for designing this template and helping me out with it!<br><br>Most importantly: thanks to the entire Open-Source community, sincerely."
 ---
 
 import HtmlEmbed from "../components/HtmlEmbed.astro";
@@ -209,7 +207,8 @@ class GlmForCausalLM(LlamaForCausalLM):
     pass
 ```
 </div>
-<div style="overflow-y: auto;">
+<div style="max-height: 400px; overflow-y: auto !important; overflow-x: hidden;">
+
 #### modeling_glm.py (auto-expanded)
 
 ```python
