diff --git "a/rotary/impls/hf_kernels_rotary.html" "b/rotary/impls/hf_kernels_rotary.html"
--- "a/rotary/impls/hf_kernels_rotary.html"
+++ "b/rotary/impls/hf_kernels_rotary.html"
@@ -4,6 +4,11 @@
     <meta charset="UTF-8">
     <meta name="viewport" content="width=device-width, initial-scale=1.0">
     <title>hf_kernels_rotary</title>
+
+    <link rel="preconnect" href="https://fonts.googleapis.com">
+    <link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
+    <link href="https://fonts.googleapis.com/css2?family=JetBrains+Mono:ital,wght@0,100..800;1,100..800&display=swap" rel="stylesheet">
+
     <script>
 // Iframe-friendly navigation router
         (function() {
@@ -87,1320 +92,1542 @@
     </script>
     <style>
 :root[data-theme="light"] {
-            --bg-primary: #ffffff;
-            --bg-secondary: #f6f8fa;
-            --bg-tertiary: #f8f9fa;
-            --bg-code: #f8f9fa;
-            --bg-error: #fdf2f2;
-            --bg-artifact: #e6f3ff;
-            --bg-artifact-hover: #d0e7ff;
-            
-            --text-primary: #333;
-            --text-secondary: #656d76;
-            --text-error: #c53030;
-            --text-link: #0969da;
-            
-            --border-primary: #e1e5e9;
-            --border-error: #e53e3e;
-            --border-cell-failed: #d73a49;
-            
-            --shadow: rgba(0, 0, 0, 0.1);
-        }
-
-        :root[data-theme="dark"] {
-            --bg-primary: #0a0a0a;
-            --bg-secondary: #121212;
-            --bg-tertiary: #181818;
-            --bg-code: #0d0d0d;
-            --bg-error: #1a0f0f;
-            --bg-artifact: #151515;
-            --bg-artifact-hover: #1a1a1a;
-            
-            --text-primary: #e0e0e0;
-            --text-secondary: #888888;
-            --text-error: #ff6b6b;
-            --text-link: #64b5f6;
-            
-            --border-primary: #2a2a2a;
-            --border-error: #ff6b6b;
-            --border-cell-failed: #ff6b6b;
-            
-            --shadow: rgba(255, 255, 255, 0.05);
-        }
-        /* Monocolor UI theme: black/white background, all text/borders single blue */
-        :root[data-ui="monocolor"] { --mono-color: #0a66ff; }
-        :root[data-ui="monocolor"][data-theme="light"] {
-            --bg-primary: #ffffff;
-        }
-        :root[data-ui="monocolor"][data-theme="dark"] {
-            --bg-primary: #000000;
-        }
-        :root[data-ui="monocolor"] {
-            --bg-secondary: var(--bg-primary);
-            --bg-tertiary: var(--bg-primary);
-            --bg-code: var(--bg-primary);
-            --bg-error: var(--bg-primary);
-            --bg-artifact: var(--bg-primary);
-            --bg-artifact-hover: var(--bg-primary);
-
-            --text-primary: var(--mono-color);
-            --text-secondary: var(--mono-color);
-            --text-error: var(--mono-color);
-            --text-link: var(--mono-color);
-
-            --border-primary: var(--mono-color);
-            --border-error: var(--mono-color);
-            --border-cell-failed: var(--mono-color);
-
-            --shadow: none;
-        }
-        :root[data-ui="monocolor"] a { color: var(--mono-color); }
-        :root[data-ui="monocolor"] .menu-button,
-        :root[data-ui="monocolor"] .theme-toggle,
-        :root[data-ui="monocolor"] .reset-toggle,
-        :root[data-ui="monocolor"] .back-button { background: var(--bg-primary); color: var(--mono-color); border-color: var(--mono-color); }
-        :root[data-ui="monocolor"] .menu-button:hover,
-        :root[data-ui="monocolor"] .theme-toggle:hover,
-        :root[data-ui="monocolor"] .reset-toggle:hover,
-        :root[data-ui="monocolor"] .back-button:hover { background: var(--bg-primary); color: var(--mono-color); border-color: var(--mono-color); }
-        :root[data-ui="monocolor"] .menu-dropdown { background: var(--bg-primary); border-color: var(--mono-color); box-shadow: none; }
-        :root[data-ui="monocolor"] .menu-item { color: var(--mono-color); border-bottom-color: var(--mono-color); }
-        :root[data-ui="monocolor"] .system-info { background: var(--bg-primary); border-color: var(--mono-color); }
-        :root[data-ui="monocolor"] .cell { border-color: var(--mono-color); background: var(--bg-primary); }
-        :root[data-ui="monocolor"] .cell-header { background: var(--bg-primary); border-bottom-color: var(--mono-color); }
-        :root[data-ui="monocolor"] .artifact { background: var(--bg-primary); border-color: var(--mono-color); color: var(--mono-color); }
-        :root[data-ui="monocolor"] .artifact:hover { background: var(--bg-primary); }
-        :root[data-ui="monocolor"] .artifact-preview img,
-        :root[data-ui="monocolor"] .artifact-preview svg { border-color: var(--mono-color); }
-        :root[data-ui="monocolor"] .status-widget { background: var(--bg-primary); border-color: var(--mono-color); color: var(--mono-color); }
-        :root[data-ui="monocolor"] .minimap,
-        :root[data-ui="monocolor"] .file-explorer,
-        :root[data-ui="monocolor"] .tools-widget {
-            background: var(--bg-primary);
-            border-color: var(--mono-color);
-            color: var(--mono-color);
-        }
-        :root[data-ui="monocolor"] .cell-code {
-            background: var(--bg-primary);
-            border-bottom-color: var(--mono-color);
-        }
-        :root[data-ui="monocolor"] .tools-title,
-        :root[data-ui="monocolor"] .file-explorer-section-title,
-        :root[data-ui="monocolor"] .minimap-title { color: var(--mono-color); border-bottom-color: var(--mono-color); }
-        :root[data-ui="monocolor"] .tool-button { background: var(--bg-primary); border-color: var(--mono-color); color: var(--mono-color); }
-        :root[data-ui="monocolor"] .tool-button.active { border-color: var(--mono-color); }
-        :root[data-ui="monocolor"] .file-explorer-item,
-        :root[data-ui="monocolor"] .minimap-item { color: var(--mono-color); }
-        /* Force Pygments code to mono blue on mono bg */
-        :root[data-ui="monocolor"] .highlight { background: var(--bg-primary) !important; color: var(--mono-color) !important; }
-        :root[data-ui="monocolor"] .highlight *,
-        :root[data-ui="monocolor"] .highlight .hll { color: var(--mono-color) !important; background: transparent !important; border-color: var(--mono-color) !important; }
-        /* Default code font + metrics (overridable via frontmatter) */
-        :root { --code-font-size: 0.95rem; --code-line-height: 1.5; --code-pad-y: 0.75rem; }
-        /* Minimal UI theme overrides base variables for a flatter, 90s look */
-        :root[data-ui="none"] {
-            --bg-primary: #ffffff;
-            --bg-secondary: transparent;
-            --bg-tertiary: transparent;
-            --bg-code: #f9f9f9;
-            --bg-error: #fff0f0;
-            --bg-artifact: #f0f7ff;
-            --bg-artifact-hover: #e5f1ff;
-
-            --text-primary: #000000;
-            --text-secondary: #222222;
-            --text-error: #a00000;
-            --text-link: #0000ee;
-
-            --border-primary: #cccccc;
-            --border-error: #cc0000;
-            --border-cell-failed: #cc0000;
-
-            --shadow: none;
-        }
-        html {
-            overscroll-behavior: none;
-        }
-        body {
-            font-family: 'Cascadia Mono', 'Cascadia Code', 'JetBrains Mono', 'SF Mono', Monaco, 'Consolas', monospace;
-            line-height: 1.4;
-            max-width: 1000px;
-            margin: 0 auto;
-            padding: 15px;
-            color: var(--text-primary);
-            background: var(--bg-primary);
-            transition: background-color 0.2s ease, color 0.2s ease;
-            overscroll-behavior: none;
-        }
-        /* Minimal "none" UI theme overrides */
-        :root[data-ui="none"] body {
-            font-family: 'Times New Roman', Times, serif;
-            line-height: 1.5;
-            max-width: 860px;
-            padding: 12px;
-            background: #ffffff;
-            color: #000000;
-            transition: none;
-        }
-        
-        /* Two panel layout removed */
-        
-        .controls {
-            position: fixed;
-            top: 20px;
-            right: 20px;
-            display: flex;
-            flex-direction: column;
-            align-items: flex-end;
-            gap: 0.25rem;
-            z-index: 1000;
-        }
-        .controls-buttons { display: flex; gap: 0.5rem; }
-        
-        .menu-button {
-            position: relative;
-            background: var(--bg-secondary);
-            border: 1px solid var(--border-primary);
-            padding: 8px 12px;
-            border-radius: 2px;
-            color: var(--text-secondary);
-            cursor: pointer;
-            font-family: inherit;
-            font-size: 0.9rem;
-            user-select: none;
-        }
-        /* Keep default control styling when widgets are enabled, even in minimal UI */
-        :root[data-ui="none"][data-widgets="on"] .menu-button,
-        :root[data-ui="none"][data-widgets="on"] .theme-toggle,
-        :root[data-ui="none"][data-widgets="on"] .reset-toggle,
-        :root[data-ui="none"][data-widgets="on"] .back-button {
-            background: #f6f6f6;
-            border: 1px solid #cccccc;
-            color: #222222;
-        }
-        
-        .menu-button:hover {
-            color: var(--text-primary);
-            background: var(--bg-tertiary);
-        }
-        /* Controls state indicator (top-right) */
-        /* Status widget (bottom-right) */
-        .status-widget {
-            position: fixed;
-            right: 20px;
-            bottom: 20px;
-            width: auto;
-            max-width: 260px;
-            background: var(--bg-secondary);
-            border: 1px solid var(--border-primary);
-            border-radius: 2px;
-            padding: 6px 8px;
-            font-size: 0.8rem;
-            color: var(--text-secondary);
-            z-index: 100;
-        }
-        .status-widget strong { color: var(--text-primary); }
-        :root[data-ui="none"][data-widgets="on"] .status-widget { background: #f6f6f6; border-color: #ccc; color: #222; }
-        :root[data-ui="none"][data-widgets="on"] .menu-button:hover,
-        :root[data-ui="none"][data-widgets="on"] .theme-toggle:hover,
-        :root[data-ui="none"][data-widgets="on"] .reset-toggle:hover,
-        :root[data-ui="none"][data-widgets="on"] .back-button:hover {
-            background: #ededed;
-            border-color: #bbbbbb;
-            color: #000000;
-        }
-        
-        .menu-dropdown {
-            position: absolute;
-            top: 100%;
-            right: 0;
-            background: var(--bg-secondary);
-            border: 1px solid var(--border-primary);
-            border-radius: 4px;
-            box-shadow: 0 4px 12px var(--shadow);
-            min-width: 160px;
-            opacity: 0;
-            visibility: hidden;
-            transform: translateY(-8px);
-            transition: all 0.2s ease;
-            z-index: 1001;
-            margin-top: 4px;
-        }
-        :root[data-ui="none"][data-widgets="on"] .menu-dropdown { background: #ffffff; border: 1px solid #cccccc; box-shadow: none; }
-        
-        .menu-button.active .menu-dropdown {
-            opacity: 1;
-            visibility: visible;
-            transform: translateY(0);
-        }
-        
-        .menu-item {
-            display: block;
-            padding: 8px 12px;
-            color: var(--text-secondary);
-            text-decoration: none;
-            font-size: 0.85rem;
-            border-bottom: 1px solid var(--border-primary);
-            cursor: pointer;
-        }
-        :root[data-ui="none"] .menu-item { color: #000; border-bottom: 1px solid #eee; }
-        
-        .menu-item:last-child {
-            border-bottom: none;
-        }
-        
-        .menu-item:hover {
-            background: var(--bg-tertiary);
-            color: var(--text-primary);
-        }
-        
-        .menu-checkbox {
-            display: inline-block;
-            width: 16px;
-            font-family: monospace;
-            color: var(--text-link);
-        }
-        
-        .theme-toggle,
-        .reset-toggle,
-        .back-button {
-            background: var(--bg-secondary);
-            border: 1px solid var(--border-primary);
-            padding: 8px 12px;
-            border-radius: 4px;
-            color: var(--text-secondary);
-            cursor: pointer;
-            font-family: inherit;
-            font-size: 0.9rem;
-            user-select: none;
-        }
-
-        .back-button {
-            text-decoration: none;
-            display: inline-block;
-        }
-
-        .theme-toggle:hover,
-        .reset-toggle:hover,
-        .back-button:hover {
-            color: var(--text-primary);
-            background: var(--bg-tertiary);
-        }
-        
-        .system-info {
-            background: var(--bg-secondary);
-            border: 1px solid var(--border-primary);
-            border-radius: 4px;
-            padding: 8px 12px;
-            margin-bottom: 16px;
-            font-size: 0.85em;
-            color: var(--text-secondary);
-        }
-        
-        .system-info-header {
-            font-weight: 600;
-            color: var(--text-primary);
-            margin-bottom: 2px;
-        }
-        
-        .system-info-content {
-            font-family: monospace;
-        }
-        
-        .theme-toggle, .reset-toggle {
-            background: var(--bg-secondary);
-            border: 1px solid var(--border-primary);
-            border-radius: 2px;
-            /* padding: 0.4rem 0.6rem; */
-            cursor: pointer;
-            font-family: inherit;
-            font-size: 0.8rem;
-            color: var(--text-secondary);
-            user-select: none;
-            transition: all 0.2s ease;
-            text-transform: lowercase;
-            letter-spacing: 0;
-        }
-        
-        .theme-toggle:hover, .reset-toggle:hover {
-            background: var(--bg-tertiary);
-            border-color: var(--text-secondary);
-            color: var(--text-primary);
-        }
-        
-        .minimap {
-            position: fixed;
-            bottom: 20px;
-            right: 20px;
-            width: 220px;
-            max-height: 400px;
-            background: var(--bg-secondary);
-            border: 1px solid var(--border-primary);
-            border-radius: 2px;
-            padding: 0.5rem;
-            font-size: 0.7rem;
-            overflow-y: auto;
-            z-index: 100;
-            opacity: 0.9;
-            transition: opacity 0.2s ease;
-        }
-        /* Hide widgets and controls when disabled via frontmatter */
-        :root[data-widgets="off"] .controls,
-        :root[data-widgets="off"] .minimap,
-        :root[data-widgets="off"] .file-explorer,
-        :root[data-widgets="off"] .tools-widget,
-        :root[data-widgets="off"] .status-widget { display: none !important; }
-        
-        .file-explorer {
-            position: fixed;
-            bottom: 20px; /* default; JS will stack */
-            right: 20px;
-            left: auto;
-            top: auto;
-            width: 220px;
-            max-height: 400px;
-            background: var(--bg-secondary);
-            border: 1px solid var(--border-primary);
-            border-radius: 2px;
-            padding: 0.5rem;
-            font-size: 0.7rem;
-            overflow-y: auto;
-            z-index: 100;
-            opacity: 0.9;
-            transition: opacity 0.2s ease;
-        }
-
-        /* Drawing overlay */
-        .draw-overlay {
-            position: fixed;
-            top: 0;
-            left: 0;
-            width: 100vw;
-            height: 100vh;
-            z-index: 80; /* under widgets (100) and controls (1000) */
-            display: block;
-            pointer-events: none; /* enabled only when a tool is active */
-        }
-
-        /* Tools widget */
-        .tools-widget {
-            position: fixed;
-            bottom: 20px; /* default; JS will stack */
-            right: 20px;
-            left: auto;
-            top: auto;
-            width: 220px;
-            background: var(--bg-secondary);
-            border: 1px solid var(--border-primary);
-            border-radius: 2px;
-            padding: 0.5rem;
-            font-size: 0.7rem;
-            z-index: 100;
-            opacity: 0.95;
-        }
-        .tools-title {
-            font-weight: bold;
-            color: var(--text-secondary);
-            margin-bottom: 0.5rem;
-            padding-bottom: 0.25rem;
-            border-bottom: 1px solid var(--border-primary);
-            cursor: grab;
-            user-select: none;
-        }
-        .tools-row { display: flex; gap: 0.4rem; flex-wrap: wrap; }
-        .tool-button {
-            background: var(--bg-tertiary);
-            border: 1px solid var(--border-primary);
-            border-radius: 2px;
-            padding: 0.25rem 0.4rem;
-            cursor: pointer;
-            color: var(--text-secondary);
-            font-family: inherit;
-            font-size: 0.75rem;
-            user-select: none;
-        }
-        .tool-button:hover { color: var(--text-primary); }
-        .tool-button.active { color: var(--text-primary); border-color: var(--text-secondary); background: var(--bg-secondary); }
-        
-        .minimap:hover, .file-explorer:hover {
-            opacity: 1;
-        }
-        
-        .minimap-title {
-            font-weight: bold;
-            color: var(--text-secondary);
-            margin-bottom: 0.5rem;
-            padding-bottom: 0.25rem;
-            border-bottom: 1px solid var(--border-primary);
-            cursor: grab; /* drag handle */
-            user-select: none;
-        }
-        
-        .minimap-item {
-            display: block;
-            color: var(--text-secondary);
-            text-decoration: none;
-            padding: 0.15rem 0;
-            border-left: 2px solid transparent;
-            padding-left: 0.5rem;
-            transition: all 0.2s ease;
-            cursor: pointer;
-        }
-        
-        .minimap-item:hover {
-            color: var(--text-primary);
-            border-left-color: var(--text-secondary);
-        }
-        
-        .minimap-item.active {
-            color: var(--text-primary);
-            border-left-color: var(--text-link);
-        }
-        
-        .minimap-heading {
-            font-weight: normal;
-        }
-        
-        .minimap-heading.h1 { padding-left: 0.5rem; }
-        .minimap-heading.h2 { padding-left: 1rem; }
-        .minimap-heading.h3 { padding-left: 1.5rem; }
-        .minimap-heading.h4 { padding-left: 2rem; }
-        .minimap-heading.h5 { padding-left: 2.5rem; }
-        .minimap-heading.h6 { padding-left: 3rem; }
-        
-        .minimap-cell {
-            color: var(--text-link);
-            opacity: 0.8;
-            font-style: italic;
-        }
-        
-        .minimap-cell:hover {
-            opacity: 1;
-        }
-        
-        .file-explorer-title {
-            font-weight: bold;
-            color: var(--text-secondary);
-            margin-bottom: 0.5rem;
-            padding-bottom: 0.25rem;
-            border-bottom: 1px solid var(--border-primary);
-            cursor: grab; /* drag handle */
-            user-select: none;
-        }
-        
-        .file-explorer-section {
-            margin-bottom: 0.75rem;
-        }
-        
-        .file-explorer-section-title {
-            font-weight: bold;
-            color: var(--text-secondary);
-            font-size: 0.65rem;
-            margin-bottom: 0.25rem;
-            text-transform: uppercase;
-            letter-spacing: 0.5px;
-        }
-        
-        .file-explorer-item {
-            display: block;
-            color: var(--text-secondary);
-            text-decoration: none;
-            padding: 0.1rem 0;
-            margin-left: 0.5rem;
-            transition: color 0.2s ease;
-            cursor: pointer;
-            font-family: monospace;
-        }
-        
-        .file-explorer-item:hover {
-            color: var(--text-primary);
-        }
-        
-        .file-explorer-item.script {
-            color: var(--text-link);
-        }
-        
-        .file-explorer-item.artifact {
-            color: var(--text-secondary);
-            opacity: 0.8;
-        }
-        
-
-        /* Hide widgets on smaller screens */
-        @media (max-width: 768px) {
-            .minimap, .file-explorer, .tools-widget {
-                display: none;
-            }
-        }
-        
-        .cell {
-            margin: 1rem 0;
-            border: 1px solid var(--border-primary);
-            border-radius: 2px;
-            overflow: hidden;
-            background: var(--bg-secondary);
-        }
-        :root[data-ui="none"] .cell { margin: 1em 0; border: none; background: transparent; }
-        .cell-header {
-            background: var(--bg-secondary);
-            padding: 0.5rem 1rem;
-            border-bottom: 1px solid var(--border-primary);
-            font-family: inherit;
-            font-size: 0.85rem;
-        }
-        :root[data-ui="none"] .cell-header { background: transparent; border: none; padding: 0; font-weight: bold; }
-        :root[data-ui="none"] .cell-content { padding: 0; }
-        :root[data-ui="none"] .copy-button,
-        :root[data-ui="none"] .collapse-indicators,
-        :root[data-ui="none"] .cell-meta,
-        :root[data-ui="none"] .cell-outputs-header { display: none !important; }
-        :root[data-ui="none"] pre, :root[data-ui="none"] code { font-family: Menlo, Monaco, 'Courier New', monospace; }
-        :root[data-ui="none"] .code-content pre { background: #f9f9f9; border: 1px solid #ddd; padding: 8px; }
-        :root[data-ui="none"] .output { background: transparent; border: none; padding: 0.25em 0; }
-            color: var(--text-secondary);
-            cursor: pointer;
-            user-select: none;
-            transition: background-color 0.2s ease;
-        }
-        .cell-header:hover {
-            background: var(--bg-tertiary);
-        }
-        .collapse-indicators {
-            color: var(--text-secondary);
-            font-size: 0.8rem;
-            opacity: 0.7;
-        }
-        .collapse-indicators span:hover {
-            color: var(--text-primary);
-            opacity: 1;
-        }
-        .cell-code {
-            display: block;
-            background: var(--bg-code);
-        }
-        .cell-code.collapsed {
-            display: none;
-        }
-        .cell-code pre {
-            margin: 0;
-            padding: 0.75rem;
-            background: var(--bg-code);
-            overflow-x: auto;
-            color: var(--text-primary);
-        }
-        .cell-output {
-            padding: 0.75rem;
-            /* background: var(--bg-primary); */
-            background: var(--bg-secondary);
-        }
-        .cell-output.collapsed {
-            display: none;
-        }
-        .cell-stdout {
-            background: var(--bg-tertiary);
-            padding: 0.75rem;
-            border-radius: 1px;
-            /* margin: 0.25rem 0; */
-            font-family: inherit;
-            font-size: 0.9rem;
-            white-space: pre-wrap;
-            color: var(--text-primary);
-        }
-
-        .cell-stdout {
-            background: var(--bg-tertiary);
-            padding: 0.75rem;
-            border-radius: 1px;
-            font-family: inherit;
-            font-size: 0.9rem;
-            color: var(--text-primary);
-
-            /* key bits */
-            overflow: auto;          /* show scrollbars when needed */
-            max-width: 100%;         /* respects whatever layout width you give it */
-        }
-
-        .cell-stdout .stdout-text {
-            margin: 0;               /* reset pre default margin */
-            white-space: pre;        /* keep line breaks, NO wrapping */
-            display: inline-block;   /* shrink-to-content */
-            min-width: max-content;  /* allow very long lines to define intrinsic width */
-            font-family: ui-monospace, SFMono-Regular, Menlo, Consolas, monospace;
-            tab-size: 2;
-        }
-
-        .cell-stderr {
-            background: var(--bg-error);
-            border-left: 2px solid var(--border-error);
-            padding: 1rem;
-            margin: 0.5rem 0;
-            font-family: inherit;
-            font-size: 0.9rem;
-            color: var(--text-error);
-            white-space: pre-wrap;
-        }
-        .uv-install-logs {
-            margin: 0.5rem 0;
-        }
-        .uv-logs-header {
-            cursor: pointer;
-            padding: 0.75rem;
-            border-left: 3px solid var(--border-color);
-            font-family: inherit;
-            font-size: 0.85rem;
-            color: var(--text-secondary);
-            user-select: none;
-        }
-        .uv-logs-content {
-            background: var(--bg-secondary);
-            padding: 1rem;
-            border-left: 3px solid var(--border-color);
-            white-space: pre-wrap;
-            font-family: monospace;
-            font-size: 0.85rem;
-            color: var(--text-secondary);
-            overflow-x: auto;
-        }
-        .cell-artifacts {
-            margin: 1rem 0;
-        }
-        .cell-artifacts h4 {
-            margin: 0 0 0.5rem 0;
-            color: var(--text-secondary);
-            font-size: 0.9rem;
-        }
-        .artifact {
-            display: inline-block;
-            background: var(--bg-artifact);
-            padding: 0.25rem 0.5rem;
-            border-radius: 1px;
-            margin: 0.25rem 0.5rem 0.25rem 0;
-            font-family: inherit;
-            font-size: 0.8rem;
-            color: var(--text-link);
-            text-decoration: none;
-            transition: background-color 0.2s ease;
-            border: 1px solid var(--border-primary);
-        }
-        .artifact:hover {
-            background: var(--bg-artifact-hover);
-        }
-        .artifact-preview {
-            margin-top: 1rem;
-        }
-        .artifact-preview img {
-            max-width: 100%;
-            height: auto;
-            border: 1px solid var(--border-primary);
-            border-radius: 1px;
-        }
-        .artifact-preview svg {
-            max-width: 100%;
-            height: auto;
-            border: 1px solid var(--border-primary);
-            border-radius: 1px;
-            display: block;
-        }
-        /* Style SVG text elements */
-        .artifact-preview svg g {
-            fill: var(--text-primary) !important;
-        }
-        /* Auto-theme SVG elements */
-        .artifact-preview svg {
-            background: transparent;
-        }
-        /* Invert SVG images in dark mode */
-        :root[data-theme="dark"] .artifact-preview img[src$=".svg"] {
-            filter: invert(0.9) hue-rotate(180deg);
-        }
-        /* Keep SVG images readable in monocolor mode */
-        :root[data-ui="monocolor"] .artifact-preview img[src$=".svg"] {
-            filter: none;
-        }
-        /* CSV table styling */
-        .artifact-csv {
-            margin-top: 1rem;
-            overflow-x: auto;
-        }
-        .csv-table {
-            width: 100%;
-            border-collapse: collapse;
-            font-size: 0.9rem;
-            background: var(--bg-secondary);
-            border: 1px solid var(--border-primary);
-            border-radius: 1px;
-        }
-        .csv-table th,
-        .csv-table td {
-            padding: 0.5rem 0.75rem;
-            text-align: left;
-            border: 1px solid var(--border-primary);
-        }
-        .csv-table th {
-            background: var(--bg-tertiary);
-            font-weight: 600;
-            color: var(--text-primary);
-        }
-        .csv-table tbody tr:hover {
-            background: var(--bg-artifact-hover);
-        }
-        .artifact-csv-error {
-            margin-top: 1rem;
-            padding: 1rem;
-            background: var(--bg-error);
-            color: var(--text-error);
-            border: 1px solid var(--border-error);
-            border-radius: 1px;
-        }
-        .cell-failed {
-            border-color: var(--border-cell-failed);
-        }
-        .cell-failed .cell-header {
-            background: var(--bg-error);
-            color: var(--text-error);
-        }
-        .cell-commented {
-            opacity: 0.6;
-            border-style: dashed;
-        }
-        .cell-commented .cell-header {
-            background: var(--bg-secondary);
-            color: var(--text-secondary);
-            font-style: italic;
-        }
-        .run-btn {
-            background: var(--bg-tertiary);
-            border: 1px solid var(--border-primary);
-            padding: 2px 6px;
-            border-radius: 2px;
-            color: var(--text-secondary);
-            cursor: pointer;
-            font-size: 0.75em;
-            font-family: inherit;
-            margin-left: 4px;
-        }
-        .run-btn:hover {
-            color: var(--text-primary);
-            background: var(--bg-primary);
-        }
-        .run-btn:disabled {
-            opacity: 0.6;
-            cursor: not-allowed;
-        }
-        .copy-btn {
-            background: var(--bg-tertiary);
-            border: 1px solid var(--border-primary);
-            padding: 2px 6px;
-            border-radius: 2px;
-            color: var(--text-secondary);
-            cursor: pointer;
-            font-size: 0.75em;
-            font-family: inherit;
-            margin-left: 4px;
-        }
-        .copy-btn:hover {
-            color: var(--text-primary);
-            background: var(--bg-primary);
-        }
-        .copy-btn:disabled {
-            opacity: 0.6;
-            cursor: not-allowed;
-        }
-        .copy-btn.copied {
-            color: #4caf50;
-            background: var(--bg-primary);
-            border-color: #4caf50;
-            transition: all 0.2s ease;
-        }
-        .raw-btn {
-            background: var(--bg-tertiary);
-            border: 1px solid var(--border-primary);
-            padding: 2px 6px;
-            border-radius: 2px;
-            color: var(--text-secondary);
-            cursor: pointer;
-            font-size: 0.75em;
-            font-family: inherit;
-            margin-left: 4px;
-            text-decoration: none;
-            display: inline-block;
-        }
-        .raw-btn:hover {
-            color: var(--text-primary);
-            background: var(--bg-primary);
-            text-decoration: none;
-        }
-        .github-btn {
-            background: var(--bg-tertiary);
-            border: 1px solid var(--border-primary);
-            padding: 2px 6px;
-            border-radius: 2px;
-            color: var(--text-secondary);
-            cursor: pointer;
-            font-size: 0.75em;
-            font-family: inherit;
-            margin-left: 4px;
-            text-decoration: none;
-            display: inline-block;
-        }
-        .github-btn:hover {
-            color: var(--text-primary);
-            background: var(--bg-primary);
-            text-decoration: none;
-        }
-        .hf-btn {
-            background: var(--bg-tertiary);
-            border: 1px solid var(--border-primary);
-            padding: 2px 6px;
-            border-radius: 2px;
-            color: var(--text-secondary);
-            cursor: pointer;
-            font-size: 0.75em;
-            font-family: inherit;
-            margin-left: 4px;
-            text-decoration: none;
-            display: inline-block;
-        }
-        .hf-btn:hover {
-            color: var(--text-primary);
-            background: var(--bg-primary);
-            text-decoration: none;
-        }
-        .output-stale {
-            opacity: 0.5;
-            position: relative;
-        }
-        .output-stale::after {
-            content: '⏳ updating...';
-            position: absolute;
-            top: 8px;
-            right: 8px;
-            background: var(--bg-secondary);
-            padding: 4px 8px;
-            border-radius: 2px;
-            font-size: 0.75em;
-            color: var(--text-secondary);
-            border: 1px solid var(--border-primary);
-        }
-        h1, h2, h3, h4, h5, h6 {
-            margin-top: 1.5rem;
-            margin-bottom: 0.75rem;
-            color: var(--text-primary);
-        }
-        h1 {
-            margin-top: 0;
-            margin-bottom: 1rem;
-        }
-        p {
-            margin: 0.75rem 0;
-            color: var(--text-primary);
-        }
-        a {
-            color: var(--text-link);
-        }
-        img {
-            max-width: 100%;
-            height: auto;
-            border-radius: 1px;
-            box-shadow: none;
-        }
-        pre, code {
-            font-family: 'Cascadia Mono', 'Cascadia Code', 'JetBrains Mono', 'SF Mono', Monaco, 'Consolas', monospace;
-            font-size: var(--code-font-size);
-        }
-        .code-wrap { position: relative; }
-        .code-line-highlight { display: none; position: absolute; left: 0; right: 0; height: 1.5em; background: rgba(255, 235, 170, 0.35); pointer-events: none; border-left: 3px solid #f4c542; }
-        .line-number { cursor: pointer; text-decoration: none; color: var(--text-secondary); padding: 0 0.25rem; }
-        .line-number.selected { background: rgba(255, 235, 170, 0.4); color: var(--text-primary); }
-        
-        /* Line numbers */
-        .highlight-with-lines {
-            display: flex;
-        }
-        .line-numbers {
-            background: var(--bg-tertiary);
-            padding: var(--code-pad-y) 0.5rem;
-            font-family: 'Cascadia Mono', 'Cascadia Code', 'JetBrains Mono', 'SF Mono', Monaco, 'Consolas', monospace;
-            font-size: var(--code-font-size);
-            line-height: var(--code-line-height);
-            color: var(--text-secondary);
-            user-select: none;
-            text-align: right;
-            border-right: 1px solid var(--border-primary);
-        }
-        .line-numbers .line-number {
-            display: block;
-            line-height: var(--code-line-height);
-        }
-        .highlight-with-lines .highlight {
-            flex: 1;
-        }
-        .highlight .hll { background-color: transparent; } /* don't conflict with our highlight */
-        .highlight pre {
-            white-space: pre;
-            margin: 0;
-            padding: var(--code-pad-y) 0.75rem;
-            line-height: var(--code-line-height);
-        }
-        
-        /* Collapsed code styling */
-        .cell-code.collapsed {
-            display: none;
-        }
-        .cell-code.expanded {
-            display: block;
-        }
-        
-        .cell-code {
-            display: block;
-            border-bottom: 1px solid var(--border-primary);
-        }
-        
-        
-        pre { line-height: 125%; }
-td.linenos .normal { color: inherit; background-color: transparent; padding-left: 5px; padding-right: 5px; }
-span.linenos { color: inherit; background-color: transparent; padding-left: 5px; padding-right: 5px; }
-td.linenos .special { color: #000000; background-color: #ffffc0; padding-left: 5px; padding-right: 5px; }
-span.linenos.special { color: #000000; background-color: #ffffc0; padding-left: 5px; padding-right: 5px; }
-[data-theme="light"] .highlight .hll { background-color: #ffffcc }
-[data-theme="light"] .highlight { background: #f8f8f8; }
-[data-theme="light"] .highlight .c { color: #3D7B7B; font-style: italic } /* Comment */
-[data-theme="light"] .highlight .err { border: 1px solid #F00 } /* Error */
-[data-theme="light"] .highlight .k { color: #008000; font-weight: bold } /* Keyword */
-[data-theme="light"] .highlight .o { color: #666 } /* Operator */
-[data-theme="light"] .highlight .ch { color: #3D7B7B; font-style: italic } /* Comment.Hashbang */
-[data-theme="light"] .highlight .cm { color: #3D7B7B; font-style: italic } /* Comment.Multiline */
-[data-theme="light"] .highlight .cp { color: #9C6500 } /* Comment.Preproc */
-[data-theme="light"] .highlight .cpf { color: #3D7B7B; font-style: italic } /* Comment.PreprocFile */
-[data-theme="light"] .highlight .c1 { color: #3D7B7B; font-style: italic } /* Comment.Single */
-[data-theme="light"] .highlight .cs { color: #3D7B7B; font-style: italic } /* Comment.Special */
-[data-theme="light"] .highlight .gd { color: #A00000 } /* Generic.Deleted */
-[data-theme="light"] .highlight .ge { font-style: italic } /* Generic.Emph */
-[data-theme="light"] .highlight .ges { font-weight: bold; font-style: italic } /* Generic.EmphStrong */
-[data-theme="light"] .highlight .gr { color: #E40000 } /* Generic.Error */
-[data-theme="light"] .highlight .gh { color: #000080; font-weight: bold } /* Generic.Heading */
-[data-theme="light"] .highlight .gi { color: #008400 } /* Generic.Inserted */
-[data-theme="light"] .highlight .go { color: #717171 } /* Generic.Output */
-[data-theme="light"] .highlight .gp { color: #000080; font-weight: bold } /* Generic.Prompt */
-[data-theme="light"] .highlight .gs { font-weight: bold } /* Generic.Strong */
-[data-theme="light"] .highlight .gu { color: #800080; font-weight: bold } /* Generic.Subheading */
-[data-theme="light"] .highlight .gt { color: #04D } /* Generic.Traceback */
-[data-theme="light"] .highlight .kc { color: #008000; font-weight: bold } /* Keyword.Constant */
-[data-theme="light"] .highlight .kd { color: #008000; font-weight: bold } /* Keyword.Declaration */
-[data-theme="light"] .highlight .kn { color: #008000; font-weight: bold } /* Keyword.Namespace */
-[data-theme="light"] .highlight .kp { color: #008000 } /* Keyword.Pseudo */
-[data-theme="light"] .highlight .kr { color: #008000; font-weight: bold } /* Keyword.Reserved */
-[data-theme="light"] .highlight .kt { color: #B00040 } /* Keyword.Type */
-[data-theme="light"] .highlight .m { color: #666 } /* Literal.Number */
-[data-theme="light"] .highlight .s { color: #BA2121 } /* Literal.String */
-[data-theme="light"] .highlight .na { color: #687822 } /* Name.Attribute */
-[data-theme="light"] .highlight .nb { color: #008000 } /* Name.Builtin */
-[data-theme="light"] .highlight .nc { color: #00F; font-weight: bold } /* Name.Class */
-[data-theme="light"] .highlight .no { color: #800 } /* Name.Constant */
-[data-theme="light"] .highlight .nd { color: #A2F } /* Name.Decorator */
-[data-theme="light"] .highlight .ni { color: #717171; font-weight: bold } /* Name.Entity */
-[data-theme="light"] .highlight .ne { color: #CB3F38; font-weight: bold } /* Name.Exception */
-[data-theme="light"] .highlight .nf { color: #00F } /* Name.Function */
-[data-theme="light"] .highlight .nl { color: #767600 } /* Name.Label */
-[data-theme="light"] .highlight .nn { color: #00F; font-weight: bold } /* Name.Namespace */
-[data-theme="light"] .highlight .nt { color: #008000; font-weight: bold } /* Name.Tag */
-[data-theme="light"] .highlight .nv { color: #19177C } /* Name.Variable */
-[data-theme="light"] .highlight .ow { color: #A2F; font-weight: bold } /* Operator.Word */
-[data-theme="light"] .highlight .w { color: #BBB } /* Text.Whitespace */
-[data-theme="light"] .highlight .mb { color: #666 } /* Literal.Number.Bin */
-[data-theme="light"] .highlight .mf { color: #666 } /* Literal.Number.Float */
-[data-theme="light"] .highlight .mh { color: #666 } /* Literal.Number.Hex */
-[data-theme="light"] .highlight .mi { color: #666 } /* Literal.Number.Integer */
-[data-theme="light"] .highlight .mo { color: #666 } /* Literal.Number.Oct */
-[data-theme="light"] .highlight .sa { color: #BA2121 } /* Literal.String.Affix */
-[data-theme="light"] .highlight .sb { color: #BA2121 } /* Literal.String.Backtick */
-[data-theme="light"] .highlight .sc { color: #BA2121 } /* Literal.String.Char */
-[data-theme="light"] .highlight .dl { color: #BA2121 } /* Literal.String.Delimiter */
-[data-theme="light"] .highlight .sd { color: #BA2121; font-style: italic } /* Literal.String.Doc */
-[data-theme="light"] .highlight .s2 { color: #BA2121 } /* Literal.String.Double */
-[data-theme="light"] .highlight .se { color: #AA5D1F; font-weight: bold } /* Literal.String.Escape */
-[data-theme="light"] .highlight .sh { color: #BA2121 } /* Literal.String.Heredoc */
-[data-theme="light"] .highlight .si { color: #A45A77; font-weight: bold } /* Literal.String.Interpol */
-[data-theme="light"] .highlight .sx { color: #008000 } /* Literal.String.Other */
-[data-theme="light"] .highlight .sr { color: #A45A77 } /* Literal.String.Regex */
-[data-theme="light"] .highlight .s1 { color: #BA2121 } /* Literal.String.Single */
-[data-theme="light"] .highlight .ss { color: #19177C } /* Literal.String.Symbol */
-[data-theme="light"] .highlight .bp { color: #008000 } /* Name.Builtin.Pseudo */
-[data-theme="light"] .highlight .fm { color: #00F } /* Name.Function.Magic */
-[data-theme="light"] .highlight .vc { color: #19177C } /* Name.Variable.Class */
-[data-theme="light"] .highlight .vg { color: #19177C } /* Name.Variable.Global */
-[data-theme="light"] .highlight .vi { color: #19177C } /* Name.Variable.Instance */
-[data-theme="light"] .highlight .vm { color: #19177C } /* Name.Variable.Magic */
-[data-theme="light"] .highlight .il { color: #666 } /* Literal.Number.Integer.Long */
-
-pre { line-height: 125%; }
-td.linenos .normal { color: inherit; background-color: transparent; padding-left: 5px; padding-right: 5px; }
-span.linenos { color: inherit; background-color: transparent; padding-left: 5px; padding-right: 5px; }
-td.linenos .special { color: #000000; background-color: #ffffc0; padding-left: 5px; padding-right: 5px; }
-span.linenos.special { color: #000000; background-color: #ffffc0; padding-left: 5px; padding-right: 5px; }
-[data-theme="dark"] .highlight .hll { background-color: #49483e }
-[data-theme="dark"] .highlight { background: #272822; color: #F8F8F2 }
-[data-theme="dark"] .highlight .c { color: #959077 } /* Comment */
-[data-theme="dark"] .highlight .err { color: #ED007E; background-color: #1E0010 } /* Error */
-[data-theme="dark"] .highlight .esc { color: #F8F8F2 } /* Escape */
-[data-theme="dark"] .highlight .g { color: #F8F8F2 } /* Generic */
-[data-theme="dark"] .highlight .k { color: #66D9EF } /* Keyword */
-[data-theme="dark"] .highlight .l { color: #AE81FF } /* Literal */
-[data-theme="dark"] .highlight .n { color: #F8F8F2 } /* Name */
-[data-theme="dark"] .highlight .o { color: #FF4689 } /* Operator */
-[data-theme="dark"] .highlight .x { color: #F8F8F2 } /* Other */
-[data-theme="dark"] .highlight .p { color: #F8F8F2 } /* Punctuation */
-[data-theme="dark"] .highlight .ch { color: #959077 } /* Comment.Hashbang */
-[data-theme="dark"] .highlight .cm { color: #959077 } /* Comment.Multiline */
-[data-theme="dark"] .highlight .cp { color: #959077 } /* Comment.Preproc */
-[data-theme="dark"] .highlight .cpf { color: #959077 } /* Comment.PreprocFile */
-[data-theme="dark"] .highlight .c1 { color: #959077 } /* Comment.Single */
-[data-theme="dark"] .highlight .cs { color: #959077 } /* Comment.Special */
-[data-theme="dark"] .highlight .gd { color: #FF4689 } /* Generic.Deleted */
-[data-theme="dark"] .highlight .ge { color: #F8F8F2; font-style: italic } /* Generic.Emph */
-[data-theme="dark"] .highlight .ges { color: #F8F8F2; font-weight: bold; font-style: italic } /* Generic.EmphStrong */
-[data-theme="dark"] .highlight .gr { color: #F8F8F2 } /* Generic.Error */
-[data-theme="dark"] .highlight .gh { color: #F8F8F2 } /* Generic.Heading */
-[data-theme="dark"] .highlight .gi { color: #A6E22E } /* Generic.Inserted */
-[data-theme="dark"] .highlight .go { color: #66D9EF } /* Generic.Output */
-[data-theme="dark"] .highlight .gp { color: #FF4689; font-weight: bold } /* Generic.Prompt */
-[data-theme="dark"] .highlight .gs { color: #F8F8F2; font-weight: bold } /* Generic.Strong */
-[data-theme="dark"] .highlight .gu { color: #959077 } /* Generic.Subheading */
-[data-theme="dark"] .highlight .gt { color: #F8F8F2 } /* Generic.Traceback */
-[data-theme="dark"] .highlight .kc { color: #66D9EF } /* Keyword.Constant */
-[data-theme="dark"] .highlight .kd { color: #66D9EF } /* Keyword.Declaration */
-[data-theme="dark"] .highlight .kn { color: #FF4689 } /* Keyword.Namespace */
-[data-theme="dark"] .highlight .kp { color: #66D9EF } /* Keyword.Pseudo */
-[data-theme="dark"] .highlight .kr { color: #66D9EF } /* Keyword.Reserved */
-[data-theme="dark"] .highlight .kt { color: #66D9EF } /* Keyword.Type */
-[data-theme="dark"] .highlight .ld { color: #E6DB74 } /* Literal.Date */
-[data-theme="dark"] .highlight .m { color: #AE81FF } /* Literal.Number */
-[data-theme="dark"] .highlight .s { color: #E6DB74 } /* Literal.String */
-[data-theme="dark"] .highlight .na { color: #A6E22E } /* Name.Attribute */
-[data-theme="dark"] .highlight .nb { color: #F8F8F2 } /* Name.Builtin */
-[data-theme="dark"] .highlight .nc { color: #A6E22E } /* Name.Class */
-[data-theme="dark"] .highlight .no { color: #66D9EF } /* Name.Constant */
-[data-theme="dark"] .highlight .nd { color: #A6E22E } /* Name.Decorator */
-[data-theme="dark"] .highlight .ni { color: #F8F8F2 } /* Name.Entity */
-[data-theme="dark"] .highlight .ne { color: #A6E22E } /* Name.Exception */
-[data-theme="dark"] .highlight .nf { color: #A6E22E } /* Name.Function */
-[data-theme="dark"] .highlight .nl { color: #F8F8F2 } /* Name.Label */
-[data-theme="dark"] .highlight .nn { color: #F8F8F2 } /* Name.Namespace */
-[data-theme="dark"] .highlight .nx { color: #A6E22E } /* Name.Other */
-[data-theme="dark"] .highlight .py { color: #F8F8F2 } /* Name.Property */
-[data-theme="dark"] .highlight .nt { color: #FF4689 } /* Name.Tag */
-[data-theme="dark"] .highlight .nv { color: #F8F8F2 } /* Name.Variable */
-[data-theme="dark"] .highlight .ow { color: #FF4689 } /* Operator.Word */
-[data-theme="dark"] .highlight .pm { color: #F8F8F2 } /* Punctuation.Marker */
-[data-theme="dark"] .highlight .w { color: #F8F8F2 } /* Text.Whitespace */
-[data-theme="dark"] .highlight .mb { color: #AE81FF } /* Literal.Number.Bin */
-[data-theme="dark"] .highlight .mf { color: #AE81FF } /* Literal.Number.Float */
-[data-theme="dark"] .highlight .mh { color: #AE81FF } /* Literal.Number.Hex */
-[data-theme="dark"] .highlight .mi { color: #AE81FF } /* Literal.Number.Integer */
-[data-theme="dark"] .highlight .mo { color: #AE81FF } /* Literal.Number.Oct */
-[data-theme="dark"] .highlight .sa { color: #E6DB74 } /* Literal.String.Affix */
-[data-theme="dark"] .highlight .sb { color: #E6DB74 } /* Literal.String.Backtick */
-[data-theme="dark"] .highlight .sc { color: #E6DB74 } /* Literal.String.Char */
-[data-theme="dark"] .highlight .dl { color: #E6DB74 } /* Literal.String.Delimiter */
-[data-theme="dark"] .highlight .sd { color: #E6DB74 } /* Literal.String.Doc */
-[data-theme="dark"] .highlight .s2 { color: #E6DB74 } /* Literal.String.Double */
-[data-theme="dark"] .highlight .se { color: #AE81FF } /* Literal.String.Escape */
-[data-theme="dark"] .highlight .sh { color: #E6DB74 } /* Literal.String.Heredoc */
-[data-theme="dark"] .highlight .si { color: #E6DB74 } /* Literal.String.Interpol */
-[data-theme="dark"] .highlight .sx { color: #E6DB74 } /* Literal.String.Other */
-[data-theme="dark"] .highlight .sr { color: #E6DB74 } /* Literal.String.Regex */
-[data-theme="dark"] .highlight .s1 { color: #E6DB74 } /* Literal.String.Single */
-[data-theme="dark"] .highlight .ss { color: #E6DB74 } /* Literal.String.Symbol */
-[data-theme="dark"] .highlight .bp { color: #F8F8F2 } /* Name.Builtin.Pseudo */
-[data-theme="dark"] .highlight .fm { color: #A6E22E } /* Name.Function.Magic */
-[data-theme="dark"] .highlight .vc { color: #F8F8F2 } /* Name.Variable.Class */
-[data-theme="dark"] .highlight .vg { color: #F8F8F2 } /* Name.Variable.Global */
-[data-theme="dark"] .highlight .vi { color: #F8F8F2 } /* Name.Variable.Instance */
-[data-theme="dark"] .highlight .vm { color: #F8F8F2 } /* Name.Variable.Magic */
-[data-theme="dark"] .highlight .il { color: #AE81FF } /* Literal.Number.Integer.Long */
-
-        /* Ensure our code metrics override Pygments defaults */
-        .highlight pre {
-            white-space: pre;
-            margin: 0;
-            padding: var(--code-pad-y) 0.75rem !important;
-            line-height: var(--code-line-height) !important;
-            font-size: var(--code-font-size) !important;
-            font-family: 'Cascadia Mono', 'Cascadia Code', 'JetBrains Mono', 'SF Mono', Monaco, 'Consolas', monospace !important;
-            border: none;
-        }
-        .line-numbers { line-height: var(--code-line-height) !important; }
-        .line-numbers .line-number { line-height: var(--code-line-height) !important; }
-
-        /* Custom CSS from frontmatter */
-        
-
-        
-        
-        
-        /* Cursor for tools */
-        body[data-tool="arrow"] .main-content { 
-            cursor: url('data:image/svg+xml;utf8,<svg xmlns="http://www.w3.org/2000/svg" width="24" height="24" viewBox="0 0 24 24" fill="none" stroke="%23e53935" stroke-width="2"><path d="M12 19l7-7 3 3-7 7-3-3z"/><path d="M18 13l-1.5-7.5L2 2l3.5 14.5L13 18l5-5z"/><path d="M2 2l7.586 7.586"/><circle cx="11" cy="11" r="2"/></svg>') 12 12, crosshair;
-        }
-        body[data-tool="pen"] .main-content { 
-            cursor: url('data:image/svg+xml;utf8,<svg xmlns="http://www.w3.org/2000/svg" width="24" height="24" viewBox="0 0 24 24" fill="none" stroke="%23e53935" stroke-width="2"><path d="M12 19l7-7 3 3-7 7-3-3z"/><path d="M18 13l-1.5-7.5L2 2l3.5 14.5L13 18l5-5z"/><circle cx="4" cy="20" r="2" fill="%23e53935"/></svg>') 4 20, pointer;
-        }
-        body[data-tool="eraser"] .main-content { 
-            cursor: url('data:image/svg+xml;utf8,<svg xmlns="http://www.w3.org/2000/svg" width="24" height="24" viewBox="0 0 24 24" fill="none" stroke="%23e53935" stroke-width="2"><path d="M20 20H7l-7-7 7-7h13v14z"/><path d="M13 13l7-7"/><path d="M13 13L9 9"/></svg>') 12 12, auto;
-        }
-
-        /* Color picker styles */
-        .tools-section-title {
-            font-weight: bold;
-            color: var(--text-secondary);
-            font-size: 0.65rem;
-            margin: 0.75rem 0 0.5rem 0;
-            text-transform: uppercase;
-            letter-spacing: 0.5px;
-        }
-        .color-row {
-            display: grid;
-            grid-template-columns: repeat(6, 1fr);
-            gap: 0.25rem;
-            margin-bottom: 0.5rem;
-        }
-        .color-swatch {
-            width: 18px;
-            height: 18px;
-            border: 2px solid var(--border-primary);
-            border-radius: 3px;
-            cursor: pointer;
-            transition: all 0.2s ease;
-            position: relative;
-        }
-        .color-swatch:hover {
-            transform: scale(1.1);
-            border-color: var(--text-secondary);
-        }
-        .color-swatch.selected {
-            border-color: var(--text-primary);
-            box-shadow: 0 0 0 2px var(--text-link);
-        }
-        .color-swatch.selected::after {
-            content: '✓';
-            position: absolute;
-            top: 50%;
-            left: 50%;
-            transform: translate(-50%, -50%);
-            color: white;
-            font-size: 10px;
-            font-weight: bold;
-            text-shadow: 1px 1px 1px black;
-        }
-        .color-input {
-            width: 24px;
-            height: 24px;
-            border: 2px solid var(--border-primary);
-            border-radius: 3px;
-            cursor: pointer;
-            background: none;
-            padding: 0;
-            grid-column: span 2;
-            justify-self: center;
-        }
-        .color-input:hover {
-            border-color: var(--text-secondary);
-        }
-        
-        /* Thickness slider styles */
-        .thickness-row {
-            display: flex;
-            align-items: center;
-            gap: 0.5rem;
-            margin-top: 0.75rem;
-        }
-        .thickness-slider {
-            flex: 1;
-            -webkit-appearance: none;
-            appearance: none;
-            height: 4px;
-            background: var(--border-primary);
-            border-radius: 2px;
-            outline: none;
-            opacity: 0.7;
-            transition: opacity 0.2s;
-        }
-        .thickness-slider:hover {
-            opacity: 1;
-        }
-        .thickness-slider::-webkit-slider-thumb {
-            -webkit-appearance: none;
-            appearance: none;
-            width: 12px;
-            height: 12px;
-            background: var(--text-link);
-            border-radius: 50%;
-            cursor: pointer;
-        }
-        .thickness-slider::-moz-range-thumb {
-            width: 12px;
-            height: 12px;
-            background: var(--text-link);
-            border-radius: 50%;
-            cursor: pointer;
-            border: none;
-        }
-        .thickness-value {
-            font-size: 0.7rem;
-            color: var(--text-secondary);
-            min-width: 20px;
-            text-align: right;
-        }
-
-        .highlight {
-            background: none !important;
-        }
-        
-        /* Loading animations */
-        .loading-spinner {
-            display: inline-block;
-            width: 16px;
-            height: 16px;
-            border: 2px solid var(--border-primary);
-            border-radius: 50%;
-            border-top-color: var(--text-link);
-            animation: spin 1s linear infinite;
-            margin-right: 8px;
-            vertical-align: middle;
-        }
-        
-        @keyframes spin {
-            to { transform: rotate(360deg); }
-        }
-        
-        .loading-skeleton {
-            display: inline-block;
-            background: var(--bg-tertiary);
-            background: linear-gradient(
-                90deg,
-                var(--bg-tertiary) 25%,
-                var(--bg-secondary) 50%,
-                var(--bg-tertiary) 75%
-            );
-            background-size: 200% 100%;
-            animation: loading-shimmer 2s ease-in-out infinite;
-            border-radius: 2px;
-            height: 1em;
-            width: 80px;
-            vertical-align: middle;
-        }
-        
-        @keyframes loading-shimmer {
-            0% { background-position: -200% 0; }
-            100% { background-position: 200% 0; }
-        }
-        
-        /* Loading state for cell output */
-        .cell-output:has(.loading-spinner) {
-            opacity: 0.7;
-            background: var(--bg-secondary);
-            /* border-left: 3px solid var(--text-link); */
-        }
+    --bg-primary: #ffffff;
+    --bg-secondary: #f6f8fa;
+    --bg-tertiary: #f8f9fa;
+    --bg-code: #f8f9fa;
+    --bg-error: #fdf2f2;
+    --bg-artifact: #e6f3ff;
+    --bg-artifact-hover: #d0e7ff;
+
+    --text-primary: #333;
+    --text-secondary: #656d76;
+    --text-error: #c53030;
+    --text-link: #0969da;
+
+    --border-primary: #e1e5e9;
+    --border-error: #e53e3e;
+    --border-cell-failed: #d73a49;
+
+    --shadow: rgba(0, 0, 0, 0.1);
+}
+
+:root[data-theme="dark"] {
+    --bg-primary: #0a0a0a;
+    --bg-secondary: #121212;
+    --bg-tertiary: #181818;
+    --bg-code: #0d0d0d;
+    --bg-error: #1a0f0f;
+    --bg-artifact: #151515;
+    --bg-artifact-hover: #1a1a1a;
+
+    --text-primary: #e0e0e0;
+    --text-secondary: #888888;
+    --text-error: #ff6b6b;
+    --text-link: #64b5f6;
+
+    --border-primary: #2a2a2a;
+    --border-error: #ff6b6b;
+    --border-cell-failed: #ff6b6b;
+
+    --shadow: rgba(255, 255, 255, 0.05);
+}
+
+/* Monocolor UI theme: black/white background, all text/borders single blue */
+:root[data-ui="monocolor"] {
+    --mono-color: #0a66ff;
+}
+
+:root[data-ui="monocolor"][data-theme="light"] {
+    --bg-primary: #ffffff;
+}
+
+:root[data-ui="monocolor"][data-theme="dark"] {
+    --bg-primary: #000000;
+}
+
+:root[data-ui="monocolor"] {
+    --bg-secondary: var(--bg-primary);
+    --bg-tertiary: var(--bg-primary);
+    --bg-code: var(--bg-primary);
+    --bg-error: var(--bg-primary);
+    --bg-artifact: var(--bg-primary);
+    --bg-artifact-hover: var(--bg-primary);
+
+    --text-primary: var(--mono-color);
+    --text-secondary: var(--mono-color);
+    --text-error: var(--mono-color);
+    --text-link: var(--mono-color);
+
+    --border-primary: var(--mono-color);
+    --border-error: var(--mono-color);
+    --border-cell-failed: var(--mono-color);
+
+    --shadow: none;
+}
+
+:root[data-ui="monocolor"] a {
+    color: var(--mono-color);
+}
+
+:root[data-ui="monocolor"] .menu-button,
+:root[data-ui="monocolor"] .theme-toggle,
+:root[data-ui="monocolor"] .reset-toggle,
+:root[data-ui="monocolor"] .back-button {
+    background: var(--bg-primary);
+    color: var(--mono-color);
+    border-color: var(--mono-color);
+}
+
+:root[data-ui="monocolor"] .menu-button:hover,
+:root[data-ui="monocolor"] .theme-toggle:hover,
+:root[data-ui="monocolor"] .reset-toggle:hover,
+:root[data-ui="monocolor"] .back-button:hover {
+    background: var(--bg-primary);
+    color: var(--mono-color);
+    border-color: var(--mono-color);
+}
+
+:root[data-ui="monocolor"] .menu-dropdown {
+    background: var(--bg-primary);
+    border-color: var(--mono-color);
+    box-shadow: none;
+}
+
+:root[data-ui="monocolor"] .menu-item {
+    color: var(--mono-color);
+    border-bottom-color: var(--mono-color);
+}
+
+:root[data-ui="monocolor"] .system-info {
+    background: var(--bg-primary);
+    border-color: var(--mono-color);
+}
+
+:root[data-ui="monocolor"] .cell {
+    border-color: var(--mono-color);
+    background: var(--bg-primary);
+}
+
+:root[data-ui="monocolor"] .cell-header {
+    background: var(--bg-primary);
+    border-bottom-color: var(--mono-color);
+}
+
+:root[data-ui="monocolor"] .artifact {
+    background: var(--bg-primary);
+    border-color: var(--mono-color);
+    color: var(--mono-color);
+}
+
+:root[data-ui="monocolor"] .artifact:hover {
+    background: var(--bg-primary);
+}
+
+:root[data-ui="monocolor"] .artifact-preview img,
+:root[data-ui="monocolor"] .artifact-preview svg {
+    border-color: var(--mono-color);
+}
+
+:root[data-ui="monocolor"] .status-widget {
+    background: var(--bg-primary);
+    border-color: var(--mono-color);
+    color: var(--mono-color);
+}
+
+:root[data-ui="monocolor"] .minimap,
+:root[data-ui="monocolor"] .file-explorer,
+:root[data-ui="monocolor"] .tools-widget {
+    background: var(--bg-primary);
+    border-color: var(--mono-color);
+    color: var(--mono-color);
+}
+
+:root[data-ui="monocolor"] .cell-code {
+    background: var(--bg-primary);
+    border-bottom-color: var(--mono-color);
+}
+
+:root[data-ui="monocolor"] .tools-title,
+:root[data-ui="monocolor"] .file-explorer-section-title,
+:root[data-ui="monocolor"] .minimap-title {
+    color: var(--mono-color);
+    border-bottom-color: var(--mono-color);
+}
+
+:root[data-ui="monocolor"] .tool-button {
+    background: var(--bg-primary);
+    border-color: var(--mono-color);
+    color: var(--mono-color);
+}
+
+:root[data-ui="monocolor"] .tool-button.active {
+    border-color: var(--mono-color);
+}
+
+:root[data-ui="monocolor"] .file-explorer-item,
+:root[data-ui="monocolor"] .minimap-item {
+    color: var(--mono-color);
+}
+
+/* Force Pygments code to mono blue on mono bg */
+:root[data-ui="monocolor"] .highlight {
+    background: var(--bg-primary) !important;
+    color: var(--mono-color) !important;
+}
+
+:root[data-ui="monocolor"] .highlight *,
+:root[data-ui="monocolor"] .highlight .hll {
+    color: var(--mono-color) !important;
+    background: transparent !important;
+    border-color: var(--mono-color) !important;
+}
+
+/* Default code font + metrics (overridable via frontmatter) */
+:root {
+    --code-font-size: 0.95rem;
+    --code-line-height: 1.5;
+    --code-pad-y: 0.75rem;
+}
+
+/* Minimal UI theme overrides base variables for a flatter, 90s look */
+:root[data-ui="none"] {
+    --bg-primary: #ffffff;
+    --bg-secondary: transparent;
+    --bg-tertiary: transparent;
+    --bg-code: #f9f9f9;
+    --bg-error: #fff0f0;
+    --bg-artifact: #f0f7ff;
+    --bg-artifact-hover: #e5f1ff;
+
+    --text-primary: #000000;
+    --text-secondary: #222222;
+    --text-error: #a00000;
+    --text-link: #0000ee;
+
+    --border-primary: #cccccc;
+    --border-error: #cc0000;
+    --border-cell-failed: #cc0000;
+
+    --shadow: none;
+}
+
+html {
+    overscroll-behavior: none;
+}
+
+body {
+    font-family: 'Cascadia Mono', 'Cascadia Code', 'JetBrains Mono', 'SF Mono', Monaco, 'Consolas', monospace;
+    line-height: 1.4;
+    max-width: 1000px;
+    margin: 0 auto;
+    padding: 15px;
+    color: var(--text-primary);
+    background: var(--bg-primary);
+    transition: background-color 0.2s ease, color 0.2s ease;
+    overscroll-behavior: none;
+}
+
+/* Minimal "none" UI theme overrides */
+:root[data-ui="none"] body {
+    font-family: 'Times New Roman', Times, serif;
+    line-height: 1.5;
+    max-width: 860px;
+    padding: 12px;
+    background: #ffffff;
+    color: #000000;
+    transition: none;
+}
+
+/* Two panel layout removed */
+
+.controls {
+    position: fixed;
+    top: 20px;
+    right: 20px;
+    display: flex;
+    flex-direction: column;
+    align-items: flex-end;
+    gap: 0.25rem;
+    z-index: 1000;
+}
+
+.controls-buttons {
+    display: flex;
+    gap: 0.5rem;
+}
+
+.menu-button {
+    position: relative;
+    background: var(--bg-secondary);
+    border: 1px solid var(--border-primary);
+    padding: 8px 12px;
+    border-radius: 2px;
+    color: var(--text-secondary);
+    cursor: pointer;
+    font-family: inherit;
+    font-size: 0.9rem;
+    user-select: none;
+}
+
+/* Keep default control styling when widgets are enabled, even in minimal UI */
+:root[data-ui="none"][data-widgets="on"] .menu-button,
+:root[data-ui="none"][data-widgets="on"] .theme-toggle,
+:root[data-ui="none"][data-widgets="on"] .reset-toggle,
+:root[data-ui="none"][data-widgets="on"] .back-button {
+    background: #f6f6f6;
+    border: 1px solid #cccccc;
+    color: #222222;
+}
+
+.menu-button:hover {
+    color: var(--text-primary);
+    background: var(--bg-tertiary);
+}
+
+/* Controls state indicator (top-right) */
+/* Status widget (bottom-right) */
+.status-widget {
+    position: fixed;
+    right: 20px;
+    bottom: 20px;
+    width: auto;
+    max-width: 260px;
+    background: var(--bg-secondary);
+    border: 1px solid var(--border-primary);
+    border-radius: 2px;
+    padding: 6px 8px;
+    font-size: 0.8rem;
+    color: var(--text-secondary);
+    z-index: 100;
+}
+
+.status-widget strong {
+    color: var(--text-primary);
+}
+
+:root[data-ui="none"][data-widgets="on"] .status-widget {
+    background: #f6f6f6;
+    border-color: #ccc;
+    color: #222;
+}
+
+:root[data-ui="none"][data-widgets="on"] .menu-button:hover,
+:root[data-ui="none"][data-widgets="on"] .theme-toggle:hover,
+:root[data-ui="none"][data-widgets="on"] .reset-toggle:hover,
+:root[data-ui="none"][data-widgets="on"] .back-button:hover {
+    background: #ededed;
+    border-color: #bbbbbb;
+    color: #000000;
+}
+
+.menu-dropdown {
+    position: absolute;
+    top: 100%;
+    right: 0;
+    background: var(--bg-secondary);
+    border: 1px solid var(--border-primary);
+    border-radius: 4px;
+    box-shadow: 0 4px 12px var(--shadow);
+    min-width: 160px;
+    opacity: 0;
+    visibility: hidden;
+    transform: translateY(-8px);
+    transition: all 0.2s ease;
+    z-index: 1001;
+    margin-top: 4px;
+}
+
+:root[data-ui="none"][data-widgets="on"] .menu-dropdown {
+    background: #ffffff;
+    border: 1px solid #cccccc;
+    box-shadow: none;
+}
+
+.menu-button.active .menu-dropdown {
+    opacity: 1;
+    visibility: visible;
+    transform: translateY(0);
+}
+
+.menu-item {
+    display: block;
+    padding: 8px 12px;
+    color: var(--text-secondary);
+    text-decoration: none;
+    font-size: 0.85rem;
+    border-bottom: 1px solid var(--border-primary);
+    cursor: pointer;
+}
+
+:root[data-ui="none"] .menu-item {
+    color: #000;
+    border-bottom: 1px solid #eee;
+}
+
+.menu-item:last-child {
+    border-bottom: none;
+}
+
+.menu-item:hover {
+    background: var(--bg-tertiary);
+    color: var(--text-primary);
+}
+
+.menu-checkbox {
+    display: inline-block;
+    width: 16px;
+    font-family: monospace;
+    color: var(--text-link);
+}
+
+.theme-toggle,
+.reset-toggle,
+.back-button {
+    background: var(--bg-secondary);
+    border: 1px solid var(--border-primary);
+    padding: 8px 12px;
+    border-radius: 4px;
+    color: var(--text-secondary);
+    cursor: pointer;
+    font-family: inherit;
+    font-size: 0.9rem;
+    user-select: none;
+}
+
+.back-button {
+    text-decoration: none;
+    display: inline-block;
+}
+
+.theme-toggle:hover,
+.reset-toggle:hover,
+.back-button:hover {
+    color: var(--text-primary);
+    background: var(--bg-tertiary);
+}
+
+.system-info {
+    background: var(--bg-secondary);
+    border: 1px solid var(--border-primary);
+    border-radius: 4px;
+    padding: 8px 12px;
+    margin-bottom: 16px;
+    font-size: 0.85em;
+    color: var(--text-secondary);
+}
+
+.system-info-header {
+    font-weight: 600;
+    color: var(--text-primary);
+    margin-bottom: 2px;
+}
+
+.system-info-content {
+    font-family: monospace;
+}
+
+.theme-toggle,
+.reset-toggle {
+    background: var(--bg-secondary);
+    border: 1px solid var(--border-primary);
+    border-radius: 2px;
+    /* padding: 0.4rem 0.6rem; */
+    cursor: pointer;
+    font-family: inherit;
+    font-size: 0.8rem;
+    color: var(--text-secondary);
+    user-select: none;
+    transition: all 0.2s ease;
+    text-transform: lowercase;
+    letter-spacing: 0;
+}
+
+.theme-toggle:hover,
+.reset-toggle:hover {
+    background: var(--bg-tertiary);
+    border-color: var(--text-secondary);
+    color: var(--text-primary);
+}
+
+.minimap {
+    position: fixed;
+    bottom: 20px;
+    right: 20px;
+    width: 220px;
+    max-height: 400px;
+    background: var(--bg-secondary);
+    border: 1px solid var(--border-primary);
+    border-radius: 2px;
+    padding: 0.5rem;
+    font-size: 0.7rem;
+    overflow-y: auto;
+    z-index: 100;
+    opacity: 0.9;
+    transition: opacity 0.2s ease;
+}
+
+/* Hide widgets and controls when disabled via frontmatter */
+:root[data-widgets="off"] .controls,
+:root[data-widgets="off"] .minimap,
+:root[data-widgets="off"] .file-explorer,
+:root[data-widgets="off"] .tools-widget,
+:root[data-widgets="off"] .status-widget {
+    display: none !important;
+}
+
+.file-explorer {
+    position: fixed;
+    bottom: 20px;
+    /* default; JS will stack */
+    right: 20px;
+    left: auto;
+    top: auto;
+    width: 220px;
+    max-height: 400px;
+    background: var(--bg-secondary);
+    border: 1px solid var(--border-primary);
+    border-radius: 2px;
+    padding: 0.5rem;
+    font-size: 0.7rem;
+    overflow-y: auto;
+    z-index: 100;
+    opacity: 0.9;
+    transition: opacity 0.2s ease;
+}
+
+/* Drawing overlay */
+.draw-overlay {
+    position: fixed;
+    top: 0;
+    left: 0;
+    width: 100vw;
+    height: 100vh;
+    z-index: 80;
+    /* under widgets (100) and controls (1000) */
+    display: block;
+    pointer-events: none;
+    /* enabled only when a tool is active */
+}
+
+/* Tools widget */
+.tools-widget {
+    position: fixed;
+    bottom: 20px;
+    /* default; JS will stack */
+    right: 20px;
+    left: auto;
+    top: auto;
+    width: 220px;
+    background: var(--bg-secondary);
+    border: 1px solid var(--border-primary);
+    border-radius: 2px;
+    padding: 0.5rem;
+    font-size: 0.7rem;
+    z-index: 100;
+    opacity: 0.95;
+}
+
+.tools-title {
+    font-weight: bold;
+    color: var(--text-secondary);
+    margin-bottom: 0.5rem;
+    padding-bottom: 0.25rem;
+    border-bottom: 1px solid var(--border-primary);
+    cursor: grab;
+    user-select: none;
+}
+
+.tools-row {
+    display: flex;
+    gap: 0.4rem;
+    flex-wrap: wrap;
+}
+
+.tool-button {
+    background: var(--bg-tertiary);
+    border: 1px solid var(--border-primary);
+    border-radius: 2px;
+    padding: 0.25rem 0.4rem;
+    cursor: pointer;
+    color: var(--text-secondary);
+    font-family: inherit;
+    font-size: 0.75rem;
+    user-select: none;
+}
+
+.tool-button:hover {
+    color: var(--text-primary);
+}
+
+.tool-button.active {
+    color: var(--text-primary);
+    border-color: var(--text-secondary);
+    background: var(--bg-secondary);
+}
+
+.minimap:hover,
+.file-explorer:hover {
+    opacity: 1;
+}
+
+.minimap-title {
+    font-weight: bold;
+    color: var(--text-secondary);
+    margin-bottom: 0.5rem;
+    padding-bottom: 0.25rem;
+    border-bottom: 1px solid var(--border-primary);
+    cursor: grab;
+    /* drag handle */
+    user-select: none;
+}
+
+.minimap-item {
+    display: block;
+    color: var(--text-secondary);
+    text-decoration: none;
+    padding: 0.15rem 0;
+    border-left: 2px solid transparent;
+    padding-left: 0.5rem;
+    transition: all 0.2s ease;
+    cursor: pointer;
+}
+
+.minimap-item:hover {
+    color: var(--text-primary);
+    border-left-color: var(--text-secondary);
+}
+
+.minimap-item.active {
+    color: var(--text-primary);
+    border-left-color: var(--text-link);
+}
+
+.minimap-heading {
+    font-weight: normal;
+}
+
+.minimap-heading.h1 {
+    padding-left: 0.5rem;
+}
+
+.minimap-heading.h2 {
+    padding-left: 1rem;
+}
+
+.minimap-heading.h3 {
+    padding-left: 1.5rem;
+}
+
+.minimap-heading.h4 {
+    padding-left: 2rem;
+}
+
+.minimap-heading.h5 {
+    padding-left: 2.5rem;
+}
+
+.minimap-heading.h6 {
+    padding-left: 3rem;
+}
+
+.minimap-cell {
+    color: var(--text-link);
+    opacity: 0.8;
+    font-style: italic;
+}
+
+.minimap-cell:hover {
+    opacity: 1;
+}
+
+.file-explorer-title {
+    font-weight: bold;
+    color: var(--text-secondary);
+    margin-bottom: 0.5rem;
+    padding-bottom: 0.25rem;
+    border-bottom: 1px solid var(--border-primary);
+    cursor: grab;
+    /* drag handle */
+    user-select: none;
+}
+
+.file-explorer-section {
+    margin-bottom: 0.75rem;
+}
+
+.file-explorer-section-title {
+    font-weight: bold;
+    color: var(--text-secondary);
+    font-size: 0.65rem;
+    margin-bottom: 0.25rem;
+    text-transform: uppercase;
+    letter-spacing: 0.5px;
+}
+
+.file-explorer-item {
+    display: block;
+    color: var(--text-secondary);
+    text-decoration: none;
+    padding: 0.1rem 0;
+    margin-left: 0.5rem;
+    transition: color 0.2s ease;
+    cursor: pointer;
+    font-family: monospace;
+}
+
+.file-explorer-item:hover {
+    color: var(--text-primary);
+}
+
+.file-explorer-item.script {
+    color: var(--text-link);
+}
+
+.file-explorer-item.artifact {
+    color: var(--text-secondary);
+    opacity: 0.8;
+}
+
+
+/* Hide widgets on smaller screens */
+@media (max-width: 768px) {
+
+    .minimap,
+    .file-explorer,
+    .tools-widget {
+        display: none;
+    }
+}
+
+.cell {
+    margin: 1rem 0;
+    border: 1px solid var(--border-primary);
+    border-radius: 2px;
+    overflow: hidden;
+    background: var(--bg-secondary);
+}
+
+:root[data-ui="none"] .cell {
+    margin: 1em 0;
+    border: none;
+    background: transparent;
+}
+
+.cell-header {
+    background: var(--bg-secondary);
+    padding: 0.5rem 1rem;
+    border-bottom: 1px solid var(--border-primary);
+    font-family: inherit;
+    font-size: 0.85rem;
+}
+
+:root[data-ui="none"] .cell-header {
+    background: transparent;
+    border: none;
+    padding: 0;
+    font-weight: bold;
+}
+
+:root[data-ui="none"] .cell-content {
+    padding: 0;
+}
+
+:root[data-ui="none"] .copy-button,
+:root[data-ui="none"] .collapse-indicators,
+:root[data-ui="none"] .cell-meta,
+:root[data-ui="none"] .cell-outputs-header {
+    display: none !important;
+}
+
+:root[data-ui="none"] pre,
+:root[data-ui="none"] code {
+    font-family: Menlo, Monaco, 'Courier New', monospace;
+}
+
+:root[data-ui="none"] .code-content pre {
+    background: #f9f9f9;
+    border: 1px solid #ddd;
+    padding: 8px;
+}
+
+:root[data-ui="none"] .output {
+    background: transparent;
+    border: none;
+    padding: 0.25em 0;
+}
+
+color: var(--text-secondary);
+cursor: pointer;
+user-select: none;
+transition: background-color 0.2s ease;
+}
+
+.cell-header:hover {
+    background: var(--bg-tertiary);
+}
+
+.collapse-indicators {
+    color: var(--text-secondary);
+    font-size: 0.8rem;
+    opacity: 0.7;
+}
+
+.collapse-indicators span:hover {
+    color: var(--text-primary);
+    opacity: 1;
+}
+
+.cell-code {
+    display: block;
+    background: var(--bg-code);
+}
+
+.cell-code.collapsed {
+    display: none;
+}
+
+.cell-code pre {
+    margin: 0;
+    padding: 0.75rem;
+    background: var(--bg-code);
+    overflow-x: auto;
+    color: var(--text-primary);
+}
+
+.cell-output {
+    padding: 0.75rem;
+    /* background: var(--bg-primary); */
+    background: var(--bg-secondary);
+}
+
+.cell-output.collapsed {
+    display: none;
+}
+
+.cell-stdout {
+    background: var(--bg-tertiary);
+    padding: 0.75rem;
+    border-radius: 1px;
+    /* margin: 0.25rem 0; */
+    font-family: inherit;
+    font-size: 0.9rem;
+    white-space: pre-wrap;
+    color: var(--text-primary);
+}
+
+.cell-stdout {
+    background: var(--bg-tertiary);
+    padding: 0.75rem;
+    border-radius: 1px;
+    font-family: inherit;
+    font-size: 0.9rem;
+    color: var(--text-primary);
+
+    /* key bits */
+    overflow: auto;
+    /* show scrollbars when needed */
+    max-width: 100%;
+    /* respects whatever layout width you give it */
+}
+
+.cell-stdout .stdout-text {
+    margin: 0;
+    /* reset pre default margin */
+    white-space: pre;
+    /* keep line breaks, NO wrapping */
+    display: inline-block;
+    /* shrink-to-content */
+    min-width: max-content;
+    /* allow very long lines to define intrinsic width */
+    font-family: ui-monospace, SFMono-Regular, Menlo, Consolas, monospace;
+    tab-size: 2;
+}
+
+.cell-stderr {
+    background: var(--bg-error);
+    border-left: 2px solid var(--border-error);
+    padding: 1rem;
+    margin: 0.5rem 0;
+    font-family: inherit;
+    font-size: 0.9rem;
+    color: var(--text-error);
+    white-space: pre-wrap;
+}
+
+.uv-install-logs {
+    margin: 0.5rem 0;
+}
+
+.uv-logs-header {
+    cursor: pointer;
+    padding: 0.75rem;
+    border-left: 3px solid var(--border-color);
+    font-family: inherit;
+    font-size: 0.85rem;
+    color: var(--text-secondary);
+    user-select: none;
+}
+
+.uv-logs-content {
+    background: var(--bg-secondary);
+    padding: 1rem;
+    border-left: 3px solid var(--border-color);
+    white-space: pre-wrap;
+    font-family: monospace;
+    font-size: 0.85rem;
+    color: var(--text-secondary);
+    overflow-x: auto;
+}
+
+.cell-artifacts {
+    margin: 1rem 0;
+}
+
+.cell-artifacts h4 {
+    margin: 0 0 0.5rem 0;
+    color: var(--text-secondary);
+    font-size: 0.9rem;
+}
+
+.artifact {
+    display: inline-block;
+    background: var(--bg-artifact);
+    padding: 0.25rem 0.5rem;
+    border-radius: 1px;
+    margin: 0.25rem 0.5rem 0.25rem 0;
+    font-family: inherit;
+    font-size: 0.8rem;
+    color: var(--text-link);
+    text-decoration: none;
+    transition: background-color 0.2s ease;
+    border: 1px solid var(--border-primary);
+}
+
+.artifact:hover {
+    background: var(--bg-artifact-hover);
+}
+
+.artifact-preview {
+    margin-top: 1rem;
+}
+
+.artifact-preview img {
+    max-width: 100%;
+    height: auto;
+    border: 1px solid var(--border-primary);
+    border-radius: 1px;
+}
+
+.artifact-preview svg {
+    max-width: 100%;
+    height: auto;
+    border: 1px solid var(--border-primary);
+    border-radius: 1px;
+    display: block;
+}
+
+/* Style SVG text elements */
+.artifact-preview svg g {
+    fill: var(--text-primary) !important;
+}
+
+/* Auto-theme SVG elements */
+.artifact-preview svg {
+    background: transparent;
+}
+
+/* Invert SVG images in dark mode */
+:root[data-theme="dark"] .artifact-preview img[src$=".svg"] {
+    filter: invert(0.9) hue-rotate(180deg);
+}
+
+/* Keep SVG images readable in monocolor mode */
+:root[data-ui="monocolor"] .artifact-preview img[src$=".svg"] {
+    filter: none;
+}
+
+/* CSV table styling */
+.artifact-csv {
+    margin-top: 1rem;
+    overflow-x: auto;
+}
+
+.csv-table {
+    width: 100%;
+    border-collapse: collapse;
+    font-size: 0.9rem;
+    background: var(--bg-secondary);
+    border: 1px solid var(--border-primary);
+    border-radius: 1px;
+}
+
+.csv-table th,
+.csv-table td {
+    padding: 0.5rem 0.75rem;
+    text-align: left;
+    border: 1px solid var(--border-primary);
+}
+
+.csv-table th {
+    background: var(--bg-tertiary);
+    font-weight: 600;
+    color: var(--text-primary);
+}
+
+.csv-table tbody tr:hover {
+    background: var(--bg-artifact-hover);
+}
+
+.artifact-csv-error {
+    margin-top: 1rem;
+    padding: 1rem;
+    background: var(--bg-error);
+    color: var(--text-error);
+    border: 1px solid var(--border-error);
+    border-radius: 1px;
+}
+
+.cell-failed {
+    border-color: var(--border-cell-failed);
+}
+
+.cell-failed .cell-header {
+    background: var(--bg-error);
+    color: var(--text-error);
+}
+
+.cell-commented {
+    opacity: 0.6;
+    border-style: dashed;
+}
+
+.cell-commented .cell-header {
+    background: var(--bg-secondary);
+    color: var(--text-secondary);
+    font-style: italic;
+}
+
+.run-btn {
+    background: var(--bg-tertiary);
+    border: 1px solid var(--border-primary);
+    padding: 2px 6px;
+    border-radius: 2px;
+    color: var(--text-secondary);
+    cursor: pointer;
+    font-size: 0.75em;
+    font-family: inherit;
+    margin-left: 4px;
+}
+
+.run-btn:hover {
+    color: var(--text-primary);
+    background: var(--bg-primary);
+}
+
+.run-btn:disabled {
+    opacity: 0.6;
+    cursor: not-allowed;
+}
+
+.copy-btn {
+    background: var(--bg-tertiary);
+    border: 1px solid var(--border-primary);
+    padding: 2px 6px;
+    border-radius: 2px;
+    color: var(--text-secondary);
+    cursor: pointer;
+    font-size: 0.75em;
+    font-family: inherit;
+    margin-left: 4px;
+}
+
+.copy-btn:hover {
+    color: var(--text-primary);
+    background: var(--bg-primary);
+}
+
+.copy-btn:disabled {
+    opacity: 0.6;
+    cursor: not-allowed;
+}
+
+.copy-btn.copied {
+    color: #4caf50;
+    background: var(--bg-primary);
+    border-color: #4caf50;
+    transition: all 0.2s ease;
+}
+
+.raw-btn {
+    background: var(--bg-tertiary);
+    border: 1px solid var(--border-primary);
+    padding: 2px 6px;
+    border-radius: 2px;
+    color: var(--text-secondary);
+    cursor: pointer;
+    font-size: 0.75em;
+    font-family: inherit;
+    margin-left: 4px;
+    text-decoration: none;
+    display: inline-block;
+}
+
+.raw-btn:hover {
+    color: var(--text-primary);
+    background: var(--bg-primary);
+    text-decoration: none;
+}
+
+.github-btn {
+    background: var(--bg-tertiary);
+    border: 1px solid var(--border-primary);
+    padding: 2px 6px;
+    border-radius: 2px;
+    color: var(--text-secondary);
+    cursor: pointer;
+    font-size: 0.75em;
+    font-family: inherit;
+    margin-left: 4px;
+    text-decoration: none;
+    display: inline-block;
+}
+
+.github-btn:hover {
+    color: var(--text-primary);
+    background: var(--bg-primary);
+    text-decoration: none;
+}
+
+.hf-btn {
+    background: var(--bg-tertiary);
+    border: 1px solid var(--border-primary);
+    padding: 2px 6px;
+    border-radius: 2px;
+    color: var(--text-secondary);
+    cursor: pointer;
+    font-size: 0.75em;
+    font-family: inherit;
+    margin-left: 4px;
+    text-decoration: none;
+    display: inline-block;
+}
+
+.hf-btn:hover {
+    color: var(--text-primary);
+    background: var(--bg-primary);
+    text-decoration: none;
+}
+
+.output-stale {
+    opacity: 0.5;
+    position: relative;
+}
+
+.output-stale::after {
+    content: '⏳ updating...';
+    position: absolute;
+    top: 8px;
+    right: 8px;
+    background: var(--bg-secondary);
+    padding: 4px 8px;
+    border-radius: 2px;
+    font-size: 0.75em;
+    color: var(--text-secondary);
+    border: 1px solid var(--border-primary);
+}
+
+h1,
+h2,
+h3,
+h4,
+h5,
+h6 {
+    margin-top: 1.5rem;
+    margin-bottom: 0.75rem;
+    color: var(--text-primary);
+}
+
+h1 {
+    margin-top: 0;
+    margin-bottom: 1rem;
+}
+
+p {
+    margin: 0.75rem 0;
+    color: var(--text-primary);
+}
+
+a {
+    color: var(--text-link);
+}
+
+img {
+    max-width: 100%;
+    height: auto;
+    border-radius: 1px;
+    box-shadow: none;
+}
+
+pre,
+code {
+    font-family: 'Cascadia Mono', 'Cascadia Code', 'JetBrains Mono', 'SF Mono', Monaco, 'Consolas', monospace;
+    font-size: var(--code-font-size);
+}
+
+.code-wrap {
+    position: relative;
+}
+
+.code-line-highlight {
+    display: none;
+    position: absolute;
+    left: 0;
+    right: 0;
+    height: 1.5em;
+    background: rgba(255, 235, 170, 0.35);
+    pointer-events: none;
+    border-left: 3px solid #f4c542;
+}
+
+.line-number {
+    cursor: pointer;
+    text-decoration: none;
+    color: var(--text-secondary);
+    padding: 0 0.25rem;
+}
+
+.line-number.selected {
+    background: rgba(255, 235, 170, 0.4);
+    color: var(--text-primary);
+}
+
+/* Line numbers */
+.highlight-with-lines {
+    display: flex;
+}
+
+.line-numbers {
+    background: var(--bg-tertiary);
+    padding: var(--code-pad-y) 0.5rem;
+    font-family: 'Cascadia Mono', 'Cascadia Code', 'JetBrains Mono', 'SF Mono', Monaco, 'Consolas', monospace;
+    font-size: var(--code-font-size);
+    line-height: var(--code-line-height);
+    color: var(--text-secondary);
+    user-select: none;
+    text-align: right;
+    border-right: 1px solid var(--border-primary);
+}
+
+.line-numbers .line-number {
+    display: block;
+    line-height: var(--code-line-height);
+}
+
+.highlight-with-lines .highlight {
+    flex: 1;
+}
+
+.highlight .hll {
+    background-color: transparent;
+}
+
+/* don't conflict with our highlight */
+.highlight pre {
+    white-space: pre;
+    margin: 0;
+    padding: var(--code-pad-y) 0.75rem;
+    line-height: var(--code-line-height);
+}
+
+/* Collapsed code styling */
+.cell-code.collapsed {
+    display: none;
+}
+
+.cell-code.expanded {
+    display: block;
+}
+
+    {
+    % if config.collapse_code %
+}
+
+.cell-code {
+    display: none;
+}
+
+    {
+    % else %
+}
+
+.cell-code {
+    display: block;
+    border-bottom: 1px solid var(--border-primary);
+}
+
+    {
+    % endif %
+}
+
+    {
+        {
+        pygments_css
+    }
+}
+
+/* Ensure our code metrics override Pygments defaults */
+.highlight pre {
+    white-space: pre;
+    margin: 0;
+    padding: var(--code-pad-y) 0.75rem !important;
+    line-height: var(--code-line-height) !important;
+    font-size: var(--code-font-size) !important;
+    font-family: 'Cascadia Mono', 'Cascadia Code', 'JetBrains Mono', 'SF Mono', Monaco, 'Consolas', monospace !important;
+    border: none;
+}
+
+.line-numbers {
+    line-height: var(--code-line-height) !important;
+}
+
+.line-numbers .line-number {
+    line-height: var(--code-line-height) !important;
+}
+
+/* Custom CSS from frontmatter */
+    {
+        {
+        config.custom_css
+    }
+}
+
+    {
+    # Override code font size from frontmatter (accept number as px) #
+}
+
+    {
+    % if config.code_font_size is not none %
+}
+
+    {
+    % if config.code_font_size is string %
+}
+
+:root {
+    --code-font-size: {
+            {
+            config.code_font_size
+        }
+    }
+
+    ;
+}
+
+    {
+    % else %
+}
+
+:root {
+    --code-font-size: {
+            {
+            config.code_font_size
+        }
+    }
+
+    px;
+}
+
+    {
+    % endif %
+}
+
+    {
+    % endif %
+}
+
+/* Cursor for tools */
+body[data-tool="arrow"] .main-content {
+    cursor: url('data:image/svg+xml;utf8,<svg xmlns="http://www.w3.org/2000/svg" width="24" height="24" viewBox="0 0 24 24" fill="none" stroke="%23e53935" stroke-width="2"><path d="M12 19l7-7 3 3-7 7-3-3z"/><path d="M18 13l-1.5-7.5L2 2l3.5 14.5L13 18l5-5z"/><path d="M2 2l7.586 7.586"/><circle cx="11" cy="11" r="2"/></svg>') 12 12, crosshair;
+}
+
+body[data-tool="pen"] .main-content {
+    cursor: url('data:image/svg+xml;utf8,<svg xmlns="http://www.w3.org/2000/svg" width="24" height="24" viewBox="0 0 24 24" fill="none" stroke="%23e53935" stroke-width="2"><path d="M12 19l7-7 3 3-7 7-3-3z"/><path d="M18 13l-1.5-7.5L2 2l3.5 14.5L13 18l5-5z"/><circle cx="4" cy="20" r="2" fill="%23e53935"/></svg>') 4 20, pointer;
+}
+
+body[data-tool="eraser"] .main-content {
+    cursor: url('data:image/svg+xml;utf8,<svg xmlns="http://www.w3.org/2000/svg" width="24" height="24" viewBox="0 0 24 24" fill="none" stroke="%23e53935" stroke-width="2"><path d="M20 20H7l-7-7 7-7h13v14z"/><path d="M13 13l7-7"/><path d="M13 13L9 9"/></svg>') 12 12, auto;
+}
+
+/* Color picker styles */
+.tools-section-title {
+    font-weight: bold;
+    color: var(--text-secondary);
+    font-size: 0.65rem;
+    margin: 0.75rem 0 0.5rem 0;
+    text-transform: uppercase;
+    letter-spacing: 0.5px;
+}
+
+.color-row {
+    display: grid;
+    grid-template-columns: repeat(6, 1fr);
+    gap: 0.25rem;
+    margin-bottom: 0.5rem;
+}
+
+.color-swatch {
+    width: 18px;
+    height: 18px;
+    border: 2px solid var(--border-primary);
+    border-radius: 3px;
+    cursor: pointer;
+    transition: all 0.2s ease;
+    position: relative;
+}
+
+.color-swatch:hover {
+    transform: scale(1.1);
+    border-color: var(--text-secondary);
+}
+
+.color-swatch.selected {
+    border-color: var(--text-primary);
+    box-shadow: 0 0 0 2px var(--text-link);
+}
+
+.color-swatch.selected::after {
+    content: '✓';
+    position: absolute;
+    top: 50%;
+    left: 50%;
+    transform: translate(-50%, -50%);
+    color: white;
+    font-size: 10px;
+    font-weight: bold;
+    text-shadow: 1px 1px 1px black;
+}
+
+.color-input {
+    width: 24px;
+    height: 24px;
+    border: 2px solid var(--border-primary);
+    border-radius: 3px;
+    cursor: pointer;
+    background: none;
+    padding: 0;
+    grid-column: span 2;
+    justify-self: center;
+}
+
+.color-input:hover {
+    border-color: var(--text-secondary);
+}
+
+/* Thickness slider styles */
+.thickness-row {
+    display: flex;
+    align-items: center;
+    gap: 0.5rem;
+    margin-top: 0.75rem;
+}
+
+.thickness-slider {
+    flex: 1;
+    -webkit-appearance: none;
+    appearance: none;
+    height: 4px;
+    background: var(--border-primary);
+    border-radius: 2px;
+    outline: none;
+    opacity: 0.7;
+    transition: opacity 0.2s;
+}
+
+.thickness-slider:hover {
+    opacity: 1;
+}
+
+.thickness-slider::-webkit-slider-thumb {
+    -webkit-appearance: none;
+    appearance: none;
+    width: 12px;
+    height: 12px;
+    background: var(--text-link);
+    border-radius: 50%;
+    cursor: pointer;
+}
+
+.thickness-slider::-moz-range-thumb {
+    width: 12px;
+    height: 12px;
+    background: var(--text-link);
+    border-radius: 50%;
+    cursor: pointer;
+    border: none;
+}
+
+.thickness-value {
+    font-size: 0.7rem;
+    color: var(--text-secondary);
+    min-width: 20px;
+    text-align: right;
+}
+
+.highlight {
+    background: none !important;
+}
+
+/* Loading animations */
+.loading-spinner {
+    display: inline-block;
+    width: 16px;
+    height: 16px;
+    border: 2px solid var(--border-primary);
+    border-radius: 50%;
+    border-top-color: var(--text-link);
+    animation: spin 1s linear infinite;
+    margin-right: 8px;
+    vertical-align: middle;
+}
+
+@keyframes spin {
+    to {
+        transform: rotate(360deg);
+    }
+}
+
+.loading-skeleton {
+    display: inline-block;
+    background: var(--bg-tertiary);
+    background: linear-gradient(90deg,
+            var(--bg-tertiary) 25%,
+            var(--bg-secondary) 50%,
+            var(--bg-tertiary) 75%);
+    background-size: 200% 100%;
+    animation: loading-shimmer 2s ease-in-out infinite;
+    border-radius: 2px;
+    height: 1em;
+    width: 80px;
+    vertical-align: middle;
+}
+
+@keyframes loading-shimmer {
+    0% {
+        background-position: -200% 0;
+    }
+
+    100% {
+        background-position: 200% 0;
+    }
+}
+
+/* Loading state for cell output */
+.cell-output:has(.loading-spinner) {
+    opacity: 0.7;
+    background: var(--bg-secondary);
+    /* border-left: 3px solid var(--text-link); */
+}
     </style>
     <script>
 // --- Drag utilities ---
@@ -3879,7 +4106,7 @@ span.linenos.special { color: #000000; background-color: #ffffc0; padding-left:
 <span onclick="toggleOutput('nv')" style="cursor: pointer;">▼ output</span>
  <span id="uv-indicator-nv" style="cursor: default; opacity: 0.3;">▶ uv-logs</span>
 </span> | 
-Cell: nv | 0.23s
+Cell: nv | 0.21s
  | <button class="run-btn" onclick="runCell('nv')">▶ run</button>
 <button class="copy-btn" onclick="copyCell('nv')">Copy</button>
 <a href="cells/nv.py" target="_blank" class="raw-btn">Raw</a>
@@ -3895,7 +4122,7 @@ Cell: nv | 0.23s
 </div>
 </div>
 <div id="output-nv" class="cell-output">
-<div class="cell-stdout"><pre class="stdout-text">Wed Oct 29 15:50:24 2025       
+<div class="cell-stdout"><pre class="stdout-text">Thu Oct 30 15:52:23 2025       
 +-----------------------------------------------------------------------------------------+
 | NVIDIA-SMI 570.195.03             Driver Version: 570.195.03     CUDA Version: 12.8     |
 |-----------------------------------------+------------------------+----------------------+
@@ -3904,7 +4131,7 @@ Cell: nv | 0.23s
 |                                         |                        |               MIG M. |
 |=========================================+========================+======================|
 |   0  NVIDIA L40S                    On  |   00000000:4D:00.0 Off |                    0 |
-| N/A   29C    P0             88W /  350W |       0MiB /  46068MiB |      0%      Default |
+| N/A   30C    P0             76W /  350W |       0MiB /  46068MiB |     11%      Default |
 |                                         |                        |                  N/A |
 +-----------------------------------------+------------------------+----------------------+
 
@@ -3928,7 +4155,7 @@ Cell: nv | 0.23s
 <span onclick="toggleOutput('benchmark')" style="cursor: pointer;">▼ output</span>
  <span id="uv-indicator-benchmark" onclick="toggleUvLogsFromHeader('benchmark')" style="cursor: pointer;">▶ uv-logs</span>
 </span> | 
-Cell: benchmark | 4.45s
+Cell: benchmark | 8.39s
  | <button class="run-btn" onclick="runCell('benchmark')">▶ run</button>
 <button class="copy-btn" onclick="copyCell('benchmark')">Copy</button>
 <a href="cells/benchmark.py" target="_blank" class="raw-btn">Raw</a>
@@ -3998,22 +4225,22 @@ PROFILE TRACE: hf_kernels_rotary | cuda_B1_S128_H8_D64_R32
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     415.167us      1791.98%     415.167us     415.167us             1  
-                                      hf_kernels_rotary        12.17%     252.218us        99.63%       2.065ms       2.065ms       0.000us         0.00%      24.448us      24.448us             1  
-                          _rotary_dba7d1e::apply_rotary         2.75%      56.920us         5.09%     105.521us      17.587us      16.128us        69.61%      16.128us       2.688us             6  
+                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     437.951us      1890.33%     437.951us     437.951us             1  
+                                      hf_kernels_rotary        12.22%     256.435us        99.67%       2.092ms       2.092ms       0.000us         0.00%      24.448us      24.448us             1  
+                          _rotary_dba7d1e::apply_rotary         2.70%      56.773us         5.22%     109.533us      18.255us      16.128us        69.61%      16.128us       2.688us             6  
 void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      16.128us        69.61%      16.128us       2.688us             6  
-                                            aten::clone         2.00%      41.539us        79.52%       1.648ms     274.716us       0.000us         0.00%       8.320us       1.387us             6  
-                                            aten::copy_         1.86%      38.603us        74.72%       1.549ms     258.116us       7.040us        30.39%       8.320us       1.387us             6  
+                                            aten::clone         2.06%      43.312us        79.20%       1.663ms     277.110us       0.000us         0.00%       8.320us       1.387us             6  
+                                            aten::copy_         2.16%      45.349us        74.16%       1.557ms     259.469us       7.040us        30.39%       8.320us       1.387us             6  
                          Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us       7.040us        30.39%       7.040us       1.173us             6  
-                                Activity Buffer Request        69.22%       1.435ms        69.22%       1.435ms       1.435ms       1.280us         5.52%       1.280us       1.280us             1  
-                                    aten::empty_strided         2.80%      58.062us         2.80%      58.062us       9.677us       0.000us         0.00%       0.000us       0.000us             6  
-                                        cudaMemcpyAsync         3.64%      75.429us         3.64%      75.429us      12.571us       0.000us         0.00%       0.000us       0.000us             6  
-                                            aten::slice         2.13%      44.231us         2.84%      58.952us       4.913us       0.000us         0.00%       0.000us       0.000us            12  
-                                       aten::as_strided         0.71%      14.721us         0.71%      14.721us       1.227us       0.000us         0.00%       0.000us       0.000us            12  
-                                       cudaLaunchKernel         2.34%      48.601us         2.34%      48.601us       8.100us       0.000us         0.00%       0.000us       0.000us             6  
-                                  cudaDeviceSynchronize         0.37%       7.760us         0.37%       7.760us       7.760us       0.000us         0.00%       0.000us       0.000us             1  
+                                Activity Buffer Request        68.35%       1.435ms        68.35%       1.435ms       1.435ms       1.280us         5.52%       1.280us       1.280us             1  
+                                    aten::empty_strided         2.98%      62.532us         2.98%      62.532us      10.422us       0.000us         0.00%       0.000us       0.000us             6  
+                                        cudaMemcpyAsync         3.65%      76.672us         3.65%      76.672us      12.779us       0.000us         0.00%       0.000us       0.000us             6  
+                                            aten::slice         2.33%      48.990us         3.04%      63.719us       5.310us       0.000us         0.00%       0.000us       0.000us            12  
+                                       aten::as_strided         0.70%      14.729us         0.70%      14.729us       1.227us       0.000us         0.00%       0.000us       0.000us            12  
+                                       cudaLaunchKernel         2.51%      52.760us         2.51%      52.760us       8.793us       0.000us         0.00%       0.000us       0.000us             6  
+                                  cudaDeviceSynchronize         0.33%       6.840us         0.33%       6.840us       6.840us       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 2.073ms
+Self CPU time total: 2.099ms
 Self CUDA time total: 23.168us
 
 
@@ -4024,23 +4251,23 @@ PROFILE TRACE: hf_kernels_rotary | cuda_B1_S128_H8_D128_R64
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     341.917us      1420.81%     341.917us     341.917us             1  
-                                      hf_kernels_rotary         9.02%     171.465us        99.73%       1.896ms       1.896ms       0.000us         0.00%      25.377us      25.377us             1  
-                          _rotary_dba7d1e::apply_rotary         2.21%      42.031us         4.55%      86.422us      14.404us      16.192us        67.28%      16.192us       2.699us             6  
-void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      16.192us        67.28%      16.192us       2.699us             6  
-                                            aten::clone         1.19%      22.618us        84.01%       1.597ms     266.223us       0.000us         0.00%       9.185us       1.531us             6  
-                                            aten::copy_         2.09%      39.723us        81.08%       1.542ms     256.918us       7.873us        32.72%       9.185us       1.531us             6  
-                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us       7.873us        32.72%       7.873us       1.312us             6  
-                                Activity Buffer Request        76.21%       1.449ms        76.21%       1.449ms       1.449ms       1.312us         5.45%       1.312us       1.312us             1  
-                                    aten::empty_strided         1.75%      33.211us         1.75%      33.211us       5.535us       0.000us         0.00%       0.000us       0.000us             6  
-                                        cudaMemcpyAsync         2.78%      52.791us         2.78%      52.791us       8.798us       0.000us         0.00%       0.000us       0.000us             6  
-                                            aten::slice         1.69%      32.169us         2.16%      40.981us       3.415us       0.000us         0.00%       0.000us       0.000us            12  
-                                       aten::as_strided         0.46%       8.812us         0.46%       8.812us       0.734us       0.000us         0.00%       0.000us       0.000us            12  
-                                       cudaLaunchKernel         2.33%      44.391us         2.33%      44.391us       7.399us       0.000us         0.00%       0.000us       0.000us             6  
-                                  cudaDeviceSynchronize         0.27%       5.060us         0.27%       5.060us       5.060us       0.000us         0.00%       0.000us       0.000us             1  
+                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     347.903us      1449.48%     347.903us     347.903us             1  
+                                      hf_kernels_rotary         8.54%     161.773us        99.74%       1.890ms       1.890ms       0.000us         0.00%      25.314us      25.314us             1  
+                          _rotary_dba7d1e::apply_rotary         2.18%      41.260us         4.61%      87.431us      14.572us      16.194us        67.47%      16.194us       2.699us             6  
+void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      16.194us        67.47%      16.194us       2.699us             6  
+                                            aten::clone         1.21%      22.941us        84.30%       1.597ms     266.206us       0.000us         0.00%       9.120us       1.520us             6  
+                                            aten::copy_         2.05%      38.809us        81.33%       1.541ms     256.844us       7.808us        32.53%       9.120us       1.520us             6  
+                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us       7.808us        32.53%       7.808us       1.301us             6  
+                                Activity Buffer Request        76.43%       1.448ms        76.43%       1.448ms       1.448ms       1.312us         5.47%       1.312us       1.312us             1  
+                                    aten::empty_strided         1.75%      33.230us         1.75%      33.230us       5.538us       0.000us         0.00%       0.000us       0.000us             6  
+                                        cudaMemcpyAsync         2.85%      54.092us         2.85%      54.092us       9.015us       0.000us         0.00%       0.000us       0.000us             6  
+                                            aten::slice         1.79%      33.972us         2.29%      43.382us       3.615us       0.000us         0.00%       0.000us       0.000us            12  
+                                       aten::as_strided         0.50%       9.410us         0.50%       9.410us       0.784us       0.000us         0.00%       0.000us       0.000us            12  
+                                       cudaLaunchKernel         2.44%      46.171us         2.44%      46.171us       7.695us       0.000us         0.00%       0.000us       0.000us             6  
+                                  cudaDeviceSynchronize         0.26%       4.990us         0.26%       4.990us       4.990us       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 1.901ms
-Self CUDA time total: 24.065us
+Self CPU time total: 1.895ms
+Self CUDA time total: 24.002us
 
 
 
@@ -4050,22 +4277,22 @@ PROFILE TRACE: hf_kernels_rotary | cuda_B1_S128_H32_D64_R32
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     339.260us      1398.72%     339.260us     339.260us             1  
-                                      hf_kernels_rotary         9.18%     174.993us        99.74%       1.901ms       1.901ms       0.000us         0.00%      25.567us      25.567us             1  
-                          _rotary_dba7d1e::apply_rotary         2.30%      43.881us         4.51%      86.021us      14.337us      16.479us        67.94%      16.479us       2.746us             6  
-void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      16.479us        67.94%      16.479us       2.746us             6  
-                                            aten::clone         1.43%      27.180us        83.89%       1.599ms     266.516us       0.000us         0.00%       9.088us       1.515us             6  
-                                            aten::copy_         2.04%      38.899us        80.70%       1.538ms     256.369us       7.776us        32.06%       9.088us       1.515us             6  
+                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     344.799us      1421.56%     344.799us     344.799us             1  
+                                      hf_kernels_rotary         8.36%     157.652us        99.72%       1.880ms       1.880ms       0.000us         0.00%      25.535us      25.535us             1  
+                          _rotary_dba7d1e::apply_rotary         2.20%      41.393us         4.58%      86.433us      14.405us      16.479us        67.94%      16.479us       2.747us             6  
+void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      16.479us        67.94%      16.479us       2.747us             6  
+                                            aten::clone         1.19%      22.449us        84.54%       1.594ms     265.688us       0.000us         0.00%       9.056us       1.509us             6  
+                                            aten::copy_         1.98%      37.391us        81.51%       1.537ms     256.168us       7.776us        32.06%       9.056us       1.509us             6  
                          Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us       7.776us        32.06%       7.776us       1.296us             6  
-                                Activity Buffer Request        75.82%       1.445ms        75.82%       1.445ms       1.445ms       1.312us         5.41%       1.312us       1.312us             1  
-                                    aten::empty_strided         1.77%      33.702us         1.77%      33.702us       5.617us       0.000us         0.00%       0.000us       0.000us             6  
-                                        cudaMemcpyAsync         2.83%      54.013us         2.83%      54.013us       9.002us       0.000us         0.00%       0.000us       0.000us             6  
-                                            aten::slice         1.70%      32.344us         2.15%      41.003us       3.417us       0.000us         0.00%       0.000us       0.000us            12  
-                                       aten::as_strided         0.45%       8.659us         0.45%       8.659us       0.722us       0.000us         0.00%       0.000us       0.000us            12  
-                                       cudaLaunchKernel         2.21%      42.140us         2.21%      42.140us       7.023us       0.000us         0.00%       0.000us       0.000us             6  
-                                  cudaDeviceSynchronize         0.26%       5.011us         0.26%       5.011us       5.011us       0.000us         0.00%       0.000us       0.000us             1  
+                                Activity Buffer Request        76.55%       1.443ms        76.55%       1.443ms       1.443ms       1.280us         5.28%       1.280us       1.280us             1  
+                                    aten::empty_strided         1.84%      34.673us         1.84%      34.673us       5.779us       0.000us         0.00%       0.000us       0.000us             6  
+                                        cudaMemcpyAsync         2.98%      56.200us         2.98%      56.200us       9.367us       0.000us         0.00%       0.000us       0.000us             6  
+                                            aten::slice         1.75%      32.991us         2.23%      42.120us       3.510us       0.000us         0.00%       0.000us       0.000us            12  
+                                       aten::as_strided         0.48%       9.129us         0.48%       9.129us       0.761us       0.000us         0.00%       0.000us       0.000us            12  
+                                       cudaLaunchKernel         2.39%      45.040us         2.39%      45.040us       7.507us       0.000us         0.00%       0.000us       0.000us             6  
+                                  cudaDeviceSynchronize         0.28%       5.250us         0.28%       5.250us       5.250us       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 1.906ms
+Self CPU time total: 1.886ms
 Self CUDA time total: 24.255us
 
 
@@ -4076,23 +4303,23 @@ PROFILE TRACE: hf_kernels_rotary | cuda_B1_S128_H32_D128_R64
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     337.722us      1204.77%     337.722us     337.722us             1  
-                                      hf_kernels_rotary         8.26%     171.103us        99.73%       2.067ms       2.067ms       0.000us         0.00%      29.792us      29.792us             1  
-                          _rotary_dba7d1e::apply_rotary         1.99%      41.331us         4.00%      82.932us      13.822us      17.632us        62.90%      17.632us       2.939us             6  
-void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      17.632us        62.90%      17.632us       2.939us             6  
-                                            aten::clone         1.32%      27.454us        85.43%       1.770ms     295.062us       0.000us         0.00%      12.160us       2.027us             6  
-                                            aten::copy_         1.75%      36.211us        82.55%       1.711ms     285.086us      10.400us        37.10%      12.160us       2.027us             6  
-                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      10.400us        37.10%      10.400us       1.733us             6  
-                                Activity Buffer Request        68.85%       1.427ms        68.85%       1.427ms       1.427ms       1.760us         6.28%       1.760us       1.760us             1  
-                                    aten::empty_strided         1.56%      32.399us         1.56%      32.399us       5.400us       0.000us         0.00%       0.000us       0.000us             6  
-                                        cudaMemcpyAsync        11.95%     247.595us        11.95%     247.595us      41.266us       0.000us         0.00%       0.000us       0.000us             6  
-                                            aten::slice         1.60%      33.121us         2.04%      42.171us       3.514us       0.000us         0.00%       0.000us       0.000us            12  
-                                       aten::as_strided         0.44%       9.050us         0.44%       9.050us       0.754us       0.000us         0.00%       0.000us       0.000us            12  
-                                       cudaLaunchKernel         2.01%      41.601us         2.01%      41.601us       6.933us       0.000us         0.00%       0.000us       0.000us             6  
-                                  cudaDeviceSynchronize         0.27%       5.640us         0.27%       5.640us       5.640us       0.000us         0.00%       0.000us       0.000us             1  
+                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     344.221us      1225.16%     344.221us     344.221us             1  
+                                      hf_kernels_rotary         7.87%     162.633us        99.75%       2.060ms       2.060ms       0.000us         0.00%      29.824us      29.824us             1  
+                          _rotary_dba7d1e::apply_rotary         1.96%      40.432us         4.15%      85.752us      14.292us      17.728us        63.10%      17.728us       2.955us             6  
+void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      17.728us        63.10%      17.728us       2.955us             6  
+                                            aten::clone         1.05%      21.772us        85.59%       1.768ms     294.674us       0.000us         0.00%      12.096us       2.016us             6  
+                                            aten::copy_         1.75%      36.131us        82.94%       1.713ms     285.533us      10.368us        36.90%      12.096us       2.016us             6  
+                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      10.368us        36.90%      10.368us       1.728us             6  
+                                Activity Buffer Request        69.12%       1.428ms        69.12%       1.428ms       1.428ms       1.728us         6.15%       1.728us       1.728us             1  
+                                    aten::empty_strided         1.60%      33.071us         1.60%      33.071us       5.512us       0.000us         0.00%       0.000us       0.000us             6  
+                                        cudaMemcpyAsync        12.07%     249.233us        12.07%     249.233us      41.539us       0.000us         0.00%       0.000us       0.000us             6  
+                                            aten::slice         1.63%      33.600us         2.13%      43.960us       3.663us       0.000us         0.00%       0.000us       0.000us            12  
+                                       aten::as_strided         0.50%      10.360us         0.50%      10.360us       0.863us       0.000us         0.00%       0.000us       0.000us            12  
+                                       cudaLaunchKernel         2.19%      45.320us         2.19%      45.320us       7.553us       0.000us         0.00%       0.000us       0.000us             6  
+                                  cudaDeviceSynchronize         0.25%       5.220us         0.25%       5.220us       5.220us       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 2.072ms
-Self CUDA time total: 28.032us
+Self CPU time total: 2.066ms
+Self CUDA time total: 28.096us
 
 
 
@@ -4102,23 +4329,23 @@ PROFILE TRACE: hf_kernels_rotary | cuda_B1_S512_H8_D64_R32
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     338.845us      1402.50%     338.845us     338.845us             1  
-                                      hf_kernels_rotary         8.30%     171.257us        99.77%       2.058ms       2.058ms       0.000us         0.00%      25.440us      25.440us             1  
-                          _rotary_dba7d1e::apply_rotary         2.01%      41.399us         4.04%      83.350us      13.892us      16.448us        68.08%      16.448us       2.741us             6  
-void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      16.448us        68.08%      16.448us       2.741us             6  
-                                            aten::clone         1.39%      28.751us        85.42%       1.762ms     293.702us       0.000us         0.00%       8.992us       1.499us             6  
-                                            aten::copy_         1.85%      38.232us        82.52%       1.702ms     283.730us       7.712us        31.92%       8.992us       1.499us             6  
-                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us       7.712us        31.92%       7.712us       1.285us             6  
-                                Activity Buffer Request        69.46%       1.433ms        69.46%       1.433ms       1.433ms       1.280us         5.30%       1.280us       1.280us             1  
-                                    aten::empty_strided         1.51%      31.081us         1.51%      31.081us       5.180us       0.000us         0.00%       0.000us       0.000us             6  
-                                        cudaMemcpyAsync        11.21%     231.173us        11.21%     231.173us      38.529us       0.000us         0.00%       0.000us       0.000us             6  
-                                            aten::slice         1.58%      32.650us         2.01%      41.461us       3.455us       0.000us         0.00%       0.000us       0.000us            12  
-                                       aten::as_strided         0.43%       8.811us         0.43%       8.811us       0.734us       0.000us         0.00%       0.000us       0.000us            12  
-                                       cudaLaunchKernel         2.03%      41.951us         2.03%      41.951us       6.992us       0.000us         0.00%       0.000us       0.000us             6  
-                                  cudaDeviceSynchronize         0.23%       4.720us         0.23%       4.720us       4.720us       0.000us         0.00%       0.000us       0.000us             1  
+                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     345.758us      1419.83%     345.758us     345.758us             1  
+                                      hf_kernels_rotary         7.72%     159.843us        99.76%       2.064ms       2.064ms       0.000us         0.00%      25.664us      25.664us             1  
+                          _rotary_dba7d1e::apply_rotary         1.98%      40.892us         4.09%      84.633us      14.106us      16.544us        67.94%      16.544us       2.757us             6  
+void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      16.544us        67.94%      16.544us       2.757us             6  
+                                            aten::clone         1.14%      23.531us        85.80%       1.775ms     295.882us       0.000us         0.00%       9.120us       1.520us             6  
+                                            aten::copy_         1.76%      36.431us        83.03%       1.718ms     286.337us       7.808us        32.06%       9.120us       1.520us             6  
+                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us       7.808us        32.06%       7.808us       1.301us             6  
+                                Activity Buffer Request        69.77%       1.444ms        69.77%       1.444ms       1.444ms       1.312us         5.39%       1.312us       1.312us             1  
+                                    aten::empty_strided         1.63%      33.740us         1.63%      33.740us       5.623us       0.000us         0.00%       0.000us       0.000us             6  
+                                        cudaMemcpyAsync        11.50%     237.923us        11.50%     237.923us      39.654us       0.000us         0.00%       0.000us       0.000us             6  
+                                            aten::slice         1.68%      34.750us         2.15%      44.540us       3.712us       0.000us         0.00%       0.000us       0.000us            12  
+                                       aten::as_strided         0.47%       9.790us         0.47%       9.790us       0.816us       0.000us         0.00%       0.000us       0.000us            12  
+                                       cudaLaunchKernel         2.11%      43.741us         2.11%      43.741us       7.290us       0.000us         0.00%       0.000us       0.000us             6  
+                                  cudaDeviceSynchronize         0.24%       4.890us         0.24%       4.890us       4.890us       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 2.063ms
-Self CUDA time total: 24.160us
+Self CPU time total: 2.069ms
+Self CUDA time total: 24.352us
 
 
 
@@ -4128,23 +4355,23 @@ PROFILE TRACE: hf_kernels_rotary | cuda_B1_S512_H8_D128_R64
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     335.067us      1195.26%     335.067us     335.067us             1  
-                                      hf_kernels_rotary        20.13%     167.343us        99.42%     826.269us     826.269us       0.000us         0.00%      29.857us      29.857us             1  
-                          _rotary_dba7d1e::apply_rotary         5.16%      42.850us        10.17%      84.521us      14.087us      17.537us        62.56%      17.537us       2.923us             6  
-void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      17.537us        62.56%      17.537us       2.923us             6  
-                                            aten::clone         2.67%      22.181us        64.36%     534.923us      89.154us       0.000us         0.00%      12.320us       2.053us             6  
-                                            aten::copy_         4.47%      37.140us        57.76%     480.051us      80.008us      10.496us        37.44%      12.320us       2.053us             6  
-                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      10.496us        37.44%      10.496us       1.749us             6  
-                                Activity Buffer Request        26.04%     216.435us        26.04%     216.435us     216.435us       1.824us         6.51%       1.824us       1.824us             1  
-                                    aten::empty_strided         3.93%      32.691us         3.93%      32.691us       5.448us       0.000us         0.00%       0.000us       0.000us             6  
-                                        cudaMemcpyAsync        27.25%     226.476us        27.25%     226.476us      37.746us       0.000us         0.00%       0.000us       0.000us             6  
-                                            aten::slice         3.75%      31.143us         4.75%      39.482us       3.290us       0.000us         0.00%       0.000us       0.000us            12  
-                                       aten::as_strided         1.00%       8.339us         1.00%       8.339us       0.695us       0.000us         0.00%       0.000us       0.000us            12  
-                                       cudaLaunchKernel         5.01%      41.671us         5.01%      41.671us       6.945us       0.000us         0.00%       0.000us       0.000us             6  
-                                  cudaDeviceSynchronize         0.58%       4.851us         0.58%       4.851us       4.851us       0.000us         0.00%       0.000us       0.000us             1  
+                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     375.259us      1340.31%     375.259us     375.259us             1  
+                                      hf_kernels_rotary         7.92%     165.422us        99.76%       2.085ms       2.085ms       0.000us         0.00%      29.790us      29.790us             1  
+                          _rotary_dba7d1e::apply_rotary         2.01%      42.019us         4.24%      88.630us      14.772us      17.566us        62.74%      17.566us       2.928us             6  
+void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      17.566us        62.74%      17.566us       2.928us             6  
+                                            aten::clone         1.13%      23.560us        85.51%       1.787ms     297.810us       0.000us         0.00%      12.224us       2.037us             6  
+                                            aten::copy_         1.86%      38.872us        82.84%       1.731ms     288.508us      10.432us        37.26%      12.224us       2.037us             6  
+                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      10.432us        37.26%      10.432us       1.739us             6  
+                                Activity Buffer Request        68.75%       1.437ms        68.75%       1.437ms       1.437ms       1.792us         6.40%       1.792us       1.792us             1  
+                                    aten::empty_strided         1.54%      32.252us         1.54%      32.252us       5.375us       0.000us         0.00%       0.000us       0.000us             6  
+                                        cudaMemcpyAsync        12.23%     255.474us        12.23%     255.474us      42.579us       0.000us         0.00%       0.000us       0.000us             6  
+                                            aten::slice         1.66%      34.672us         2.10%      43.902us       3.658us       0.000us         0.00%       0.000us       0.000us            12  
+                                       aten::as_strided         0.44%       9.230us         0.44%       9.230us       0.769us       0.000us         0.00%       0.000us       0.000us            12  
+                                       cudaLaunchKernel         2.23%      46.611us         2.23%      46.611us       7.769us       0.000us         0.00%       0.000us       0.000us             6  
+                                  cudaDeviceSynchronize         0.24%       4.930us         0.24%       4.930us       4.930us       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 831.120us
-Self CUDA time total: 28.033us
+Self CPU time total: 2.090ms
+Self CUDA time total: 27.998us
 
 
 
@@ -4154,23 +4381,23 @@ PROFILE TRACE: hf_kernels_rotary | cuda_B1_S512_H32_D64_R32
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     332.760us       825.36%     332.760us     332.760us             1  
-                                      hf_kernels_rotary        19.35%     167.193us        99.38%     858.880us     858.880us       0.000us         0.00%      43.165us      43.165us             1  
-                          _rotary_dba7d1e::apply_rotary         4.67%      40.341us         9.39%      81.181us      13.530us      23.229us        57.62%      23.229us       3.871us             6  
-void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      23.229us        57.62%      23.229us       3.871us             6  
-                                            aten::clone         2.64%      22.801us        65.85%     569.083us      94.847us       0.000us         0.00%      19.936us       3.323us             6  
-                                            aten::copy_         4.30%      37.172us        59.60%     515.092us      85.849us      17.088us        42.38%      19.936us       3.323us             6  
-                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      17.088us        42.38%      17.088us       2.848us             6  
-                                Activity Buffer Request        29.73%     256.965us        29.73%     256.965us     256.965us       2.848us         7.06%       2.848us       2.848us             1  
-                                    aten::empty_strided         3.61%      31.190us         3.61%      31.190us       5.198us       0.000us         0.00%       0.000us       0.000us             6  
-                                        cudaMemcpyAsync        25.57%     220.955us        25.57%     220.955us      36.826us       0.000us         0.00%       0.000us       0.000us             6  
-                                            aten::slice         3.76%      32.492us         4.79%      41.423us       3.452us       0.000us         0.00%       0.000us       0.000us            12  
-                                       aten::as_strided         1.03%       8.931us         1.03%       8.931us       0.744us       0.000us         0.00%       0.000us       0.000us            12  
-                                       cudaLaunchKernel         4.73%      40.840us         4.73%      40.840us       6.807us       0.000us         0.00%       0.000us       0.000us             6  
-                                  cudaDeviceSynchronize         0.62%       5.380us         0.62%       5.380us       5.380us       0.000us         0.00%       0.000us       0.000us             1  
+                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     346.557us       858.83%     346.557us     346.557us             1  
+                                      hf_kernels_rotary         7.80%     160.642us        99.76%       2.055ms       2.055ms       0.000us         0.00%      43.200us      43.200us             1  
+                          _rotary_dba7d1e::apply_rotary         2.00%      41.122us         4.23%      87.123us      14.521us      23.424us        58.05%      23.424us       3.904us             6  
+void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      23.424us        58.05%      23.424us       3.904us             6  
+                                            aten::clone         1.11%      22.900us        85.69%       1.765ms     294.130us       0.000us         0.00%      19.776us       3.296us             6  
+                                            aten::copy_         1.80%      37.091us        82.95%       1.708ms     284.737us      16.928us        41.95%      19.776us       3.296us             6  
+                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      16.928us        41.95%      16.928us       2.821us             6  
+                                Activity Buffer Request        70.02%       1.442ms        70.02%       1.442ms       1.442ms       2.848us         7.06%       2.848us       2.848us             1  
+                                    aten::empty_strided         1.62%      33.460us         1.62%      33.460us       5.577us       0.000us         0.00%       0.000us       0.000us             6  
+                                        cudaMemcpyAsync        11.13%     229.194us        11.13%     229.194us      38.199us       0.000us         0.00%       0.000us       0.000us             6  
+                                            aten::slice         1.60%      33.049us         2.04%      42.051us       3.504us       0.000us         0.00%       0.000us       0.000us            12  
+                                       aten::as_strided         0.44%       9.002us         0.44%       9.002us       0.750us       0.000us         0.00%       0.000us       0.000us            12  
+                                       cudaLaunchKernel         2.23%      46.001us         2.23%      46.001us       7.667us       0.000us         0.00%       0.000us       0.000us             6  
+                                  cudaDeviceSynchronize         0.24%       4.950us         0.24%       4.950us       4.950us       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 864.260us
-Self CUDA time total: 40.317us
+Self CPU time total: 2.060ms
+Self CUDA time total: 40.352us
 
 
 
@@ -4180,23 +4407,23 @@ PROFILE TRACE: hf_kernels_rotary | cuda_B1_S512_H32_D128_R64
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     353.785us       452.38%     353.785us     353.785us             1  
-                                      hf_kernels_rotary        18.85%     162.244us        99.38%     855.480us     855.480us       0.000us         0.00%      90.716us      90.716us             1  
-                                            aten::clone         2.60%      22.411us        65.93%     567.532us      94.589us       0.000us         0.00%      52.253us       8.709us             6  
-                                            aten::copy_         4.73%      40.709us        58.92%     507.190us      84.532us      39.742us        50.82%      52.253us       8.709us             6  
-                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      39.742us        50.82%      39.742us       6.624us             6  
-                          _rotary_dba7d1e::apply_rotary         4.83%      41.551us         9.72%      83.643us      13.941us      38.463us        49.18%      38.463us       6.410us             6  
-void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      38.463us        49.18%      38.463us       6.410us             6  
-                                Activity Buffer Request        28.85%     248.356us        28.85%     248.356us     248.356us      12.511us        16.00%      12.511us      12.511us             1  
-                                    aten::empty_strided         4.41%      37.931us         4.41%      37.931us       6.322us       0.000us         0.00%       0.000us       0.000us             6  
-                                        cudaMemcpyAsync        25.34%     218.125us        25.34%     218.125us      36.354us       0.000us         0.00%       0.000us       0.000us             6  
-                                            aten::slice         3.86%      33.191us         4.89%      42.061us       3.505us       0.000us         0.00%       0.000us       0.000us            12  
-                                       aten::as_strided         1.03%       8.870us         1.03%       8.870us       0.739us       0.000us         0.00%       0.000us       0.000us            12  
-                                       cudaLaunchKernel         4.89%      42.092us         4.89%      42.092us       7.015us       0.000us         0.00%       0.000us       0.000us             6  
-                                  cudaDeviceSynchronize         0.62%       5.350us         0.62%       5.350us       5.350us       0.000us         0.00%       0.000us       0.000us             1  
+                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     349.374us       446.91%     349.374us     349.374us             1  
+                                      hf_kernels_rotary         8.00%     163.391us        99.76%       2.039ms       2.039ms       0.000us         0.00%      90.720us      90.720us             1  
+                                            aten::clone         1.09%      22.181us        85.39%       1.745ms     290.833us       0.000us         0.00%      52.224us       8.704us             6  
+                                            aten::copy_         1.85%      37.761us        82.69%       1.690ms     281.650us      39.680us        50.76%      52.224us       8.704us             6  
+                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      39.680us        50.76%      39.680us       6.613us             6  
+                          _rotary_dba7d1e::apply_rotary         2.10%      42.834us         4.25%      86.883us      14.481us      38.496us        49.24%      38.496us       6.416us             6  
+void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      38.496us        49.24%      38.496us       6.416us             6  
+                                Activity Buffer Request        69.78%       1.426ms        69.78%       1.426ms       1.426ms      12.544us        16.05%      12.544us      12.544us             1  
+                                    aten::empty_strided         1.61%      32.920us         1.61%      32.920us       5.487us       0.000us         0.00%       0.000us       0.000us             6  
+                                        cudaMemcpyAsync        11.06%     226.094us        11.06%     226.094us      37.682us       0.000us         0.00%       0.000us       0.000us             6  
+                                            aten::slice         1.62%      33.171us         2.12%      43.331us       3.611us       0.000us         0.00%       0.000us       0.000us            12  
+                                       aten::as_strided         0.50%      10.160us         0.50%      10.160us       0.847us       0.000us         0.00%       0.000us       0.000us            12  
+                                       cudaLaunchKernel         2.16%      44.049us         2.16%      44.049us       7.341us       0.000us         0.00%       0.000us       0.000us             6  
+                                  cudaDeviceSynchronize         0.24%       4.951us         0.24%       4.951us       4.951us       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 860.830us
-Self CUDA time total: 78.205us
+Self CPU time total: 2.044ms
+Self CUDA time total: 78.176us
 
 
 
@@ -4206,23 +4433,23 @@ PROFILE TRACE: hf_kernels_rotary | cuda_B1_S2048_H8_D64_R32
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     337.342us       833.35%     337.342us     337.342us             1  
-                                      hf_kernels_rotary         8.49%     173.955us        99.77%       2.043ms       2.043ms       0.000us         0.00%      43.328us      43.328us             1  
-                          _rotary_dba7d1e::apply_rotary         2.03%      41.590us         4.02%      82.231us      13.705us      23.487us        58.02%      23.487us       3.915us             6  
-void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      23.487us        58.02%      23.487us       3.915us             6  
-                                            aten::clone         1.34%      27.379us        85.23%       1.745ms     290.890us       0.000us         0.00%      19.841us       3.307us             6  
-                                            aten::copy_         1.78%      36.424us        82.41%       1.688ms     281.287us      16.993us        41.98%      19.841us       3.307us             6  
-                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      16.993us        41.98%      16.993us       2.832us             6  
-                                Activity Buffer Request        70.11%       1.436ms        70.11%       1.436ms       1.436ms       2.848us         7.04%       2.848us       2.848us             1  
-                                    aten::empty_strided         1.48%      30.241us         1.48%      30.241us       5.040us       0.000us         0.00%       0.000us       0.000us             6  
-                                        cudaMemcpyAsync        10.52%     215.434us        10.52%     215.434us      35.906us       0.000us         0.00%       0.000us       0.000us             6  
-                                            aten::slice         1.62%      33.159us         2.03%      41.651us       3.471us       0.000us         0.00%       0.000us       0.000us            12  
-                                       aten::as_strided         0.41%       8.492us         0.41%       8.492us       0.708us       0.000us         0.00%       0.000us       0.000us            12  
-                                       cudaLaunchKernel         1.98%      40.641us         1.98%      40.641us       6.773us       0.000us         0.00%       0.000us       0.000us             6  
-                                  cudaDeviceSynchronize         0.23%       4.700us         0.23%       4.700us       4.700us       0.000us         0.00%       0.000us       0.000us             1  
+                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     333.879us       824.19%     333.879us     333.879us             1  
+                                      hf_kernels_rotary        18.73%     154.483us        99.41%     820.134us     820.134us       0.000us         0.00%      43.327us      43.327us             1  
+                          _rotary_dba7d1e::apply_rotary         4.89%      40.361us        10.02%      82.702us      13.784us      23.422us        57.82%      23.422us       3.904us             6  
+void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      23.422us        57.82%      23.422us       3.904us             6  
+                                            aten::clone         2.46%      20.259us        65.56%     540.868us      90.145us       0.000us         0.00%      19.905us       3.317us             6  
+                                            aten::copy_         4.70%      38.811us        59.16%     488.099us      81.350us      17.088us        42.18%      19.905us       3.317us             6  
+                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      17.088us        42.18%      17.088us       2.848us             6  
+                                Activity Buffer Request        27.39%     225.944us        27.39%     225.944us     225.944us       2.817us         6.95%       2.817us       2.817us             1  
+                                    aten::empty_strided         3.94%      32.510us         3.94%      32.510us       5.418us       0.000us         0.00%       0.000us       0.000us             6  
+                                        cudaMemcpyAsync        27.07%     223.344us        27.07%     223.344us      37.224us       0.000us         0.00%       0.000us       0.000us             6  
+                                            aten::slice         3.93%      32.394us         5.10%      42.081us       3.507us       0.000us         0.00%       0.000us       0.000us            12  
+                                       aten::as_strided         1.17%       9.687us         1.17%       9.687us       0.807us       0.000us         0.00%       0.000us       0.000us            12  
+                                       cudaLaunchKernel         5.13%      42.341us         5.13%      42.341us       7.057us       0.000us         0.00%       0.000us       0.000us             6  
+                                  cudaDeviceSynchronize         0.59%       4.860us         0.59%       4.860us       4.860us       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 2.048ms
-Self CUDA time total: 40.480us
+Self CPU time total: 824.994us
+Self CUDA time total: 40.510us
 
 
 
@@ -4232,23 +4459,23 @@ PROFILE TRACE: hf_kernels_rotary | cuda_B1_S2048_H8_D128_R64
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     367.387us       482.60%     367.387us     367.387us             1  
-                                      hf_kernels_rotary         8.43%     173.690us        99.77%       2.056ms       2.056ms       0.000us         0.00%      86.687us      86.687us             1  
-                                            aten::clone         1.25%      25.689us        84.25%       1.736ms     289.338us       0.000us         0.00%      47.648us       7.941us             6  
-                                            aten::copy_         1.77%      36.381us        81.42%       1.678ms     279.615us      37.088us        48.72%      47.648us       7.941us             6  
-                          _rotary_dba7d1e::apply_rotary         2.83%      58.403us         5.00%     103.123us      17.187us      39.039us        51.28%      39.039us       6.506us             6  
-void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      39.039us        51.28%      39.039us       6.506us             6  
-                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      37.088us        48.72%      37.088us       6.181us             6  
-                                Activity Buffer Request        69.27%       1.427ms        69.27%       1.427ms       1.427ms      10.560us        13.87%      10.560us      10.560us             1  
-                                    aten::empty_strided         1.58%      32.653us         1.58%      32.653us       5.442us       0.000us         0.00%       0.000us       0.000us             6  
-                                        cudaMemcpyAsync        10.38%     213.985us        10.38%     213.985us      35.664us       0.000us         0.00%       0.000us       0.000us             6  
-                                            aten::slice         1.64%      33.864us         2.09%      42.973us       3.581us       0.000us         0.00%       0.000us       0.000us            12  
-                                       aten::as_strided         0.44%       9.109us         0.44%       9.109us       0.759us       0.000us         0.00%       0.000us       0.000us            12  
-                                       cudaLaunchKernel         2.17%      44.720us         2.17%      44.720us       7.453us       0.000us         0.00%       0.000us       0.000us             6  
-                                  cudaDeviceSynchronize         0.23%       4.800us         0.23%       4.800us       4.800us       0.000us         0.00%       0.000us       0.000us             1  
+                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     338.778us       450.33%     338.778us     338.778us             1  
+                                      hf_kernels_rotary        18.40%     151.937us        99.39%     820.824us     820.824us       0.000us         0.00%      85.723us      85.723us             1  
+                                            aten::clone         2.47%      20.430us        65.45%     540.538us      90.090us       0.000us         0.00%      47.293us       7.882us             6  
+                                            aten::copy_         4.41%      36.400us        59.08%     487.928us      81.321us      36.798us        48.92%      47.293us       7.882us             6  
+                          _rotary_dba7d1e::apply_rotary         4.89%      40.390us        10.51%      86.760us      14.460us      38.430us        51.08%      38.430us       6.405us             6  
+void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      38.430us        51.08%      38.430us       6.405us             6  
+                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      36.798us        48.92%      36.798us       6.133us             6  
+                                Activity Buffer Request        27.74%     229.134us        27.74%     229.134us     229.134us      10.495us        13.95%      10.495us      10.495us             1  
+                                    aten::empty_strided         3.90%      32.180us         3.90%      32.180us       5.363us       0.000us         0.00%       0.000us       0.000us             6  
+                                        cudaMemcpyAsync        26.93%     222.394us        26.93%     222.394us      37.066us       0.000us         0.00%       0.000us       0.000us             6  
+                                            aten::slice         3.90%      32.180us         5.04%      41.589us       3.466us       0.000us         0.00%       0.000us       0.000us            12  
+                                       aten::as_strided         1.14%       9.409us         1.14%       9.409us       0.784us       0.000us         0.00%       0.000us       0.000us            12  
+                                       cudaLaunchKernel         5.61%      46.370us         5.61%      46.370us       7.728us       0.000us         0.00%       0.000us       0.000us             6  
+                                  cudaDeviceSynchronize         0.61%       5.040us         0.61%       5.040us       5.040us       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 2.061ms
-Self CUDA time total: 76.127us
+Self CPU time total: 825.864us
+Self CUDA time total: 75.228us
 
 
 
@@ -4258,23 +4485,23 @@ PROFILE TRACE: hf_kernels_rotary | cuda_B1_S2048_H32_D64_R32
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     350.717us       252.54%     350.717us     350.717us             1  
-                                      hf_kernels_rotary         8.20%     168.514us        99.75%       2.049ms       2.049ms       0.000us         0.00%     162.494us     162.494us             1  
-                                            aten::clone         1.38%      28.280us        85.37%       1.754ms     292.317us       0.000us         0.00%     102.494us      17.082us             6  
-                                            aten::copy_         1.89%      38.810us        82.42%       1.693ms     282.225us      78.878us        56.80%     102.494us      17.082us             6  
-                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      78.878us        56.80%      78.878us      13.146us             6  
-                          _rotary_dba7d1e::apply_rotary         2.03%      41.642us         4.17%      85.643us      14.274us      60.000us        43.20%      60.000us      10.000us             6  
-void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      60.000us        43.20%      60.000us      10.000us             6  
-                                Activity Buffer Request        70.32%       1.445ms        70.32%       1.445ms       1.445ms      23.616us        17.00%      23.616us      23.616us             1  
-                                    aten::empty_strided         1.57%      32.271us         1.57%      32.271us       5.379us       0.000us         0.00%       0.000us       0.000us             6  
-                                        cudaMemcpyAsync        10.22%     209.905us        10.22%     209.905us      34.984us       0.000us         0.00%       0.000us       0.000us             6  
-                                            aten::slice         1.59%      32.591us         2.01%      41.291us       3.441us       0.000us         0.00%       0.000us       0.000us            12  
-                                       aten::as_strided         0.42%       8.700us         0.42%       8.700us       0.725us       0.000us         0.00%       0.000us       0.000us            12  
-                                       cudaLaunchKernel         2.14%      44.001us         2.14%      44.001us       7.333us       0.000us         0.00%       0.000us       0.000us             6  
-                                  cudaDeviceSynchronize         0.25%       5.120us         0.25%       5.120us       5.120us       0.000us         0.00%       0.000us       0.000us             1  
+                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     338.815us       244.98%     338.815us     338.815us             1  
+                                      hf_kernels_rotary        17.96%     152.299us        99.45%     843.474us     843.474us       0.000us         0.00%     161.823us     161.823us             1  
+                                            aten::clone         2.40%      20.339us        66.32%     562.460us      93.743us       0.000us         0.00%     102.176us      17.029us             6  
+                                            aten::copy_         4.27%      36.251us        60.21%     510.629us      85.105us      78.656us        56.87%     102.176us      17.029us             6  
+                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      78.656us        56.87%      78.656us      13.109us             6  
+                          _rotary_dba7d1e::apply_rotary         4.86%      41.202us        10.23%      86.763us      14.460us      59.647us        43.13%      59.647us       9.941us             6  
+void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      59.647us        43.13%      59.647us       9.941us             6  
+                                Activity Buffer Request        30.37%     257.584us        30.37%     257.584us     257.584us      23.520us        17.01%      23.520us      23.520us             1  
+                                    aten::empty_strided         3.71%      31.492us         3.71%      31.492us       5.249us       0.000us         0.00%       0.000us       0.000us             6  
+                                        cudaMemcpyAsync        25.56%     216.794us        25.56%     216.794us      36.132us       0.000us         0.00%       0.000us       0.000us             6  
+                                            aten::slice         3.89%      32.951us         4.95%      41.952us       3.496us       0.000us         0.00%       0.000us       0.000us            12  
+                                       aten::as_strided         1.06%       9.001us         1.06%       9.001us       0.750us       0.000us         0.00%       0.000us       0.000us            12  
+                                       cudaLaunchKernel         5.37%      45.561us         5.37%      45.561us       7.594us       0.000us         0.00%       0.000us       0.000us             6  
+                                  cudaDeviceSynchronize         0.55%       4.640us         0.55%       4.640us       4.640us       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 2.054ms
-Self CUDA time total: 138.878us
+Self CPU time total: 848.114us
+Self CUDA time total: 138.303us
 
 
 
@@ -4284,23 +4511,23 @@ PROFILE TRACE: hf_kernels_rotary | cuda_B1_S2048_H32_D128_R64
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                      hf_kernels_rotary         7.33%     173.322us        87.05%       2.058ms       2.058ms       0.000us         0.00%     773.117us     773.117us             1  
-                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     713.886us       101.15%     713.886us     713.886us             1  
-                                            aten::clone         1.12%      26.510us        74.03%       1.750ms     291.637us       0.000us         0.00%     574.975us      95.829us             6  
-                                            aten::copy_         1.66%      39.271us        70.83%       1.674ms     279.060us     507.647us        71.93%     574.975us      95.829us             6  
-                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us     507.647us        71.93%     507.647us      84.608us             6  
-                          _rotary_dba7d1e::apply_rotary         1.93%      45.683us         3.90%      92.264us      15.377us     198.142us        28.07%     198.142us      33.024us             6  
-void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us     198.142us        28.07%     198.142us      33.024us             6  
-                                Activity Buffer Request        60.04%       1.419ms        60.04%       1.419ms       1.419ms      67.328us         9.54%      67.328us      67.328us             1  
-                                    aten::empty_strided         2.07%      48.953us         2.07%      48.953us       8.159us       0.000us         0.00%       0.000us       0.000us             6  
-                                        cudaMemcpyAsync         9.13%     215.804us         9.13%     215.804us      35.967us       0.000us         0.00%       0.000us       0.000us             6  
-                                            aten::slice         1.42%      33.620us         1.79%      42.281us       3.523us       0.000us         0.00%       0.000us       0.000us            12  
-                                       aten::as_strided         0.37%       8.661us         0.37%       8.661us       0.722us       0.000us         0.00%       0.000us       0.000us            12  
-                                       cudaLaunchKernel         1.97%      46.581us         1.97%      46.581us       7.764us       0.000us         0.00%       0.000us       0.000us             6  
-                                  cudaDeviceSynchronize        12.95%     306.087us        12.95%     306.087us     306.087us       0.000us         0.00%       0.000us       0.000us             1  
+                                      hf_kernels_rotary        12.84%     152.812us        71.89%     855.575us     855.575us       0.000us         0.00%     769.625us     769.625us             1  
+                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     710.234us       101.16%     710.234us     710.234us             1  
+                                            aten::clone         1.76%      21.001us        48.07%     572.021us      95.337us       0.000us         0.00%     572.987us      95.498us             6  
+                                            aten::copy_         3.15%      37.471us        43.65%     519.450us      86.575us     505.436us        71.99%     572.987us      95.498us             6  
+                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us     505.436us        71.99%     505.436us      84.239us             6  
+                          _rotary_dba7d1e::apply_rotary         3.42%      40.722us         7.33%      87.262us      14.544us     196.638us        28.01%     196.638us      32.773us             6  
+void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us     196.638us        28.01%     196.638us      32.773us             6  
+                                Activity Buffer Request        21.90%     260.665us        21.90%     260.665us     260.665us      67.551us         9.62%      67.551us      67.551us             1  
+                                    aten::empty_strided         2.65%      31.570us         2.65%      31.570us       5.262us       0.000us         0.00%       0.000us       0.000us             6  
+                                        cudaMemcpyAsync        18.60%     221.314us        18.60%     221.314us      36.886us       0.000us         0.00%       0.000us       0.000us             6  
+                                            aten::slice         2.82%      33.601us         3.65%      43.480us       3.623us       0.000us         0.00%       0.000us       0.000us            12  
+                                       aten::as_strided         0.83%       9.879us         0.83%       9.879us       0.823us       0.000us         0.00%       0.000us       0.000us            12  
+                                       cudaLaunchKernel         3.91%      46.540us         3.91%      46.540us       7.757us       0.000us         0.00%       0.000us       0.000us             6  
+                                  cudaDeviceSynchronize        28.11%     334.485us        28.11%     334.485us     334.485us       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 2.364ms
-Self CUDA time total: 705.789us
+Self CPU time total: 1.190ms
+Self CUDA time total: 702.074us
 
 
 
@@ -4310,23 +4537,23 @@ PROFILE TRACE: hf_kernels_rotary | cuda_B2_S128_H8_D64_R32
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     345.724us      1295.48%     345.724us     345.724us             1  
-                                      hf_kernels_rotary         8.56%     176.117us        99.77%       2.053ms       2.053ms       0.000us         0.00%      27.999us      27.999us             1  
-                          _rotary_dba7d1e::apply_rotary         2.01%      41.279us         4.04%      83.070us      13.845us      18.817us        70.51%      18.817us       3.136us             6  
-void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      18.817us        70.51%      18.817us       3.136us             6  
-                                            aten::clone         1.59%      32.672us        85.14%       1.752ms     292.043us       0.000us         0.00%       9.182us       1.530us             6  
-                                            aten::copy_         1.83%      37.700us        81.90%       1.685ms     280.910us       7.870us        29.49%       9.182us       1.530us             6  
-                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us       7.870us        29.49%       7.870us       1.312us             6  
-                                Activity Buffer Request        70.01%       1.441ms        70.01%       1.441ms       1.441ms       1.312us         4.92%       1.312us       1.312us             1  
-                                    aten::empty_strided         1.66%      34.130us         1.66%      34.130us       5.688us       0.000us         0.00%       0.000us       0.000us             6  
-                                        cudaMemcpyAsync        10.06%     206.965us        10.06%     206.965us      34.494us       0.000us         0.00%       0.000us       0.000us             6  
-                                            aten::slice         1.59%      32.681us         2.04%      41.911us       3.493us       0.000us         0.00%       0.000us       0.000us            12  
-                                       aten::as_strided         0.45%       9.230us         0.45%       9.230us       0.769us       0.000us         0.00%       0.000us       0.000us            12  
-                                       cudaLaunchKernel         2.03%      41.791us         2.03%      41.791us       6.965us       0.000us         0.00%       0.000us       0.000us             6  
-                                  cudaDeviceSynchronize         0.23%       4.640us         0.23%       4.640us       4.640us       0.000us         0.00%       0.000us       0.000us             1  
+                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     340.957us      1280.69%     340.957us     340.957us             1  
+                                      hf_kernels_rotary        17.85%     154.192us        99.45%     858.915us     858.915us       0.000us         0.00%      27.935us      27.935us             1  
+                          _rotary_dba7d1e::apply_rotary         4.82%      41.593us        10.09%      87.173us      14.529us      18.719us        70.31%      18.719us       3.120us             6  
+void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      18.719us        70.31%      18.719us       3.120us             6  
+                                            aten::clone         2.51%      21.701us        66.67%     575.779us      95.963us       0.000us         0.00%       9.216us       1.536us             6  
+                                            aten::copy_         4.05%      34.978us        60.54%     522.828us      87.138us       7.904us        29.69%       9.216us       1.536us             6  
+                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us       7.904us        29.69%       7.904us       1.317us             6  
+                                Activity Buffer Request        30.68%     265.004us        30.68%     265.004us     265.004us       1.312us         4.93%       1.312us       1.312us             1  
+                                    aten::empty_strided         3.62%      31.250us         3.62%      31.250us       5.208us       0.000us         0.00%       0.000us       0.000us             6  
+                                        cudaMemcpyAsync        25.80%     222.846us        25.80%     222.846us      37.141us       0.000us         0.00%       0.000us       0.000us             6  
+                                            aten::slice         3.77%      32.522us         4.84%      41.771us       3.481us       0.000us         0.00%       0.000us       0.000us            12  
+                                       aten::as_strided         1.07%       9.249us         1.07%       9.249us       0.771us       0.000us         0.00%       0.000us       0.000us            12  
+                                       cudaLaunchKernel         5.28%      45.580us         5.28%      45.580us       7.597us       0.000us         0.00%       0.000us       0.000us             6  
+                                  cudaDeviceSynchronize         0.55%       4.760us         0.55%       4.760us       4.760us       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 2.058ms
-Self CUDA time total: 26.687us
+Self CPU time total: 863.675us
+Self CUDA time total: 26.623us
 
 
 
@@ -4336,23 +4563,23 @@ PROFILE TRACE: hf_kernels_rotary | cuda_B2_S128_H8_D128_R64
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     334.455us      1253.30%     334.455us     334.455us             1  
-                                      hf_kernels_rotary        18.38%     152.071us        99.44%     822.719us     822.719us       0.000us         0.00%      27.966us      27.966us             1  
-                          _rotary_dba7d1e::apply_rotary         5.29%      43.741us        10.38%      85.902us      14.317us      18.975us        71.10%      18.975us       3.163us             6  
-void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      18.975us        71.10%      18.975us       3.163us             6  
-                                            aten::clone         2.47%      20.399us        65.75%     544.023us      90.670us       0.000us         0.00%       8.991us       1.498us             6  
-                                            aten::copy_         4.79%      39.600us        59.70%     493.952us      82.325us       7.711us        28.90%       8.991us       1.498us             6  
-                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us       7.711us        28.90%       7.711us       1.285us             6  
-                                Activity Buffer Request        30.27%     250.456us        30.27%     250.456us     250.456us       1.280us         4.80%       1.280us       1.280us             1  
-                                    aten::empty_strided         3.59%      29.672us         3.59%      29.672us       4.945us       0.000us         0.00%       0.000us       0.000us             6  
-                                        cudaMemcpyAsync        24.64%     203.896us        24.64%     203.896us      33.983us       0.000us         0.00%       0.000us       0.000us             6  
-                                            aten::slice         3.84%      31.802us         4.92%      40.723us       3.394us       0.000us         0.00%       0.000us       0.000us            12  
-                                       aten::as_strided         1.08%       8.921us         1.08%       8.921us       0.743us       0.000us         0.00%       0.000us       0.000us            12  
-                                       cudaLaunchKernel         5.10%      42.161us         5.10%      42.161us       7.027us       0.000us         0.00%       0.000us       0.000us             6  
-                                  cudaDeviceSynchronize         0.56%       4.640us         0.56%       4.640us       4.640us       0.000us         0.00%       0.000us       0.000us             1  
+                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     331.838us      1247.93%     331.838us     331.838us             1  
+                                      hf_kernels_rotary        18.40%     149.763us        99.33%     808.424us     808.424us       0.000us         0.00%      27.871us      27.871us             1  
+                          _rotary_dba7d1e::apply_rotary         5.12%      41.640us        10.68%      86.941us      14.490us      18.879us        71.00%      18.879us       3.147us             6  
+void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      18.879us        71.00%      18.879us       3.147us             6  
+                                            aten::clone         2.56%      20.830us        65.24%     531.000us      88.500us       0.000us         0.00%       8.992us       1.499us             6  
+                                            aten::copy_         4.49%      36.550us        58.98%     480.009us      80.001us       7.712us        29.00%       8.992us       1.499us             6  
+                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us       7.712us        29.00%       7.712us       1.285us             6  
+                                Activity Buffer Request        28.18%     229.375us        28.18%     229.375us     229.375us       1.280us         4.81%       1.280us       1.280us             1  
+                                    aten::empty_strided         3.71%      30.161us         3.71%      30.161us       5.027us       0.000us         0.00%       0.000us       0.000us             6  
+                                        cudaMemcpyAsync        26.30%     214.084us        26.30%     214.084us      35.681us       0.000us         0.00%       0.000us       0.000us             6  
+                                            aten::slice         3.92%      31.890us         5.00%      40.720us       3.393us       0.000us         0.00%       0.000us       0.000us            12  
+                                       aten::as_strided         1.08%       8.830us         1.08%       8.830us       0.736us       0.000us         0.00%       0.000us       0.000us            12  
+                                       cudaLaunchKernel         5.57%      45.301us         5.57%      45.301us       7.550us       0.000us         0.00%       0.000us       0.000us             6  
+                                  cudaDeviceSynchronize         0.67%       5.440us         0.67%       5.440us       5.440us       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 827.359us
-Self CUDA time total: 26.686us
+Self CPU time total: 813.864us
+Self CUDA time total: 26.591us
 
 
 
@@ -4362,23 +4589,23 @@ PROFILE TRACE: hf_kernels_rotary | cuda_B2_S128_H32_D64_R32
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     335.033us      1090.71%     335.033us     335.033us             1  
-                                      hf_kernels_rotary        18.88%     152.276us        99.36%     801.289us     801.289us       0.000us         0.00%      32.445us      32.445us             1  
-                          _rotary_dba7d1e::apply_rotary         5.08%      40.990us        10.38%      83.672us      13.945us      20.127us        65.52%      20.127us       3.354us             6  
-void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      20.127us        65.52%      20.127us       3.354us             6  
-                                            aten::clone         2.64%      21.299us        65.14%     525.331us      87.555us       0.000us         0.00%      12.318us       2.053us             6  
-                                            aten::copy_         5.22%      42.109us        58.69%     473.291us      78.882us      10.590us        34.48%      12.318us       2.053us             6  
-                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      10.590us        34.48%      10.590us       1.765us             6  
-                                Activity Buffer Request        26.39%     212.815us        26.39%     212.815us     212.815us       1.728us         5.63%       1.728us       1.728us             1  
-                                    aten::empty_strided         3.81%      30.741us         3.81%      30.741us       5.123us       0.000us         0.00%       0.000us       0.000us             6  
-                                        cudaMemcpyAsync        27.08%     218.367us        27.08%     218.367us      36.394us       0.000us         0.00%       0.000us       0.000us             6  
-                                            aten::slice         3.88%      31.271us         4.96%      40.010us       3.334us       0.000us         0.00%       0.000us       0.000us            12  
-                                       aten::as_strided         1.08%       8.739us         1.08%       8.739us       0.728us       0.000us         0.00%       0.000us       0.000us            12  
-                                       cudaLaunchKernel         5.29%      42.682us         5.29%      42.682us       7.114us       0.000us         0.00%       0.000us       0.000us             6  
-                                  cudaDeviceSynchronize         0.64%       5.180us         0.64%       5.180us       5.180us       0.000us         0.00%       0.000us       0.000us             1  
+                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     353.852us      1157.89%     353.852us     353.852us             1  
+                                      hf_kernels_rotary         7.66%     156.034us        99.77%       2.033ms       2.033ms       0.000us         0.00%      32.320us      32.320us             1  
+                          _rotary_dba7d1e::apply_rotary         2.04%      41.512us         4.26%      86.762us      14.460us      20.159us        65.97%      20.159us       3.360us             6  
+void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      20.159us        65.97%      20.159us       3.360us             6  
+                                            aten::clone         1.10%      22.431us        85.66%       1.746ms     290.955us       0.000us         0.00%      12.161us       2.027us             6  
+                                            aten::copy_         2.23%      45.431us        82.85%       1.688ms     281.408us      10.401us        34.03%      12.161us       2.027us             6  
+                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      10.401us        34.03%      10.401us       1.734us             6  
+                                Activity Buffer Request        70.07%       1.428ms        70.07%       1.428ms       1.428ms       1.760us         5.76%       1.760us       1.760us             1  
+                                    aten::empty_strided         1.71%      34.849us         1.71%      34.849us       5.808us       0.000us         0.00%       0.000us       0.000us             6  
+                                        cudaMemcpyAsync        10.54%     214.913us        10.54%     214.913us      35.819us       0.000us         0.00%       0.000us       0.000us             6  
+                                            aten::slice         1.68%      34.241us         2.20%      44.770us       3.731us       0.000us         0.00%       0.000us       0.000us            12  
+                                       aten::as_strided         0.52%      10.529us         0.52%      10.529us       0.877us       0.000us         0.00%       0.000us       0.000us            12  
+                                       cudaLaunchKernel         2.22%      45.250us         2.22%      45.250us       7.542us       0.000us         0.00%       0.000us       0.000us             6  
+                                  cudaDeviceSynchronize         0.23%       4.770us         0.23%       4.770us       4.770us       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 806.469us
-Self CUDA time total: 30.717us
+Self CPU time total: 2.038ms
+Self CUDA time total: 30.560us
 
 
 
@@ -4388,23 +4615,23 @@ PROFILE TRACE: hf_kernels_rotary | cuda_B2_S128_H32_D128_R64
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     336.501us       787.14%     336.501us     336.501us             1  
-                                      hf_kernels_rotary        18.58%     152.224us        99.36%     814.019us     814.019us       0.000us         0.00%      45.662us      45.662us             1  
-                          _rotary_dba7d1e::apply_rotary         5.02%      41.151us        10.13%      82.992us      13.832us      25.695us        60.11%      25.695us       4.283us             6  
-void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      25.695us        60.11%      25.695us       4.283us             6  
-                                            aten::clone         2.59%      21.259us        65.61%     537.562us      89.594us       0.000us         0.00%      19.967us       3.328us             6  
-                                            aten::copy_         4.69%      38.391us        59.23%     485.282us      80.880us      17.055us        39.89%      19.967us       3.328us             6  
-                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      17.055us        39.89%      17.055us       2.842us             6  
-                                Activity Buffer Request        29.84%     244.476us        29.84%     244.476us     244.476us       2.912us         6.81%       2.912us       2.912us             1  
-                                    aten::empty_strided         3.79%      31.021us         3.79%      31.021us       5.170us       0.000us         0.00%       0.000us       0.000us             6  
-                                        cudaMemcpyAsync        24.71%     202.415us        24.71%     202.415us      33.736us       0.000us         0.00%       0.000us       0.000us             6  
-                                            aten::slice         3.95%      32.360us         5.03%      41.241us       3.437us       0.000us         0.00%       0.000us       0.000us            12  
-                                       aten::as_strided         1.08%       8.881us         1.08%       8.881us       0.740us       0.000us         0.00%       0.000us       0.000us            12  
-                                       cudaLaunchKernel         5.11%      41.841us         5.11%      41.841us       6.973us       0.000us         0.00%       0.000us       0.000us             6  
-                                  cudaDeviceSynchronize         0.64%       5.261us         0.64%       5.261us       5.261us       0.000us         0.00%       0.000us       0.000us             1  
+                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     367.612us       860.51%     367.612us     367.612us             1  
+                                      hf_kernels_rotary         7.69%     158.003us        99.76%       2.050ms       2.050ms       0.000us         0.00%      45.568us      45.568us             1  
+                          _rotary_dba7d1e::apply_rotary         2.04%      41.961us         4.25%      87.391us      14.565us      25.759us        60.30%      25.759us       4.293us             6  
+void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      25.759us        60.30%      25.759us       4.293us             6  
+                                            aten::clone         1.11%      22.799us        84.82%       1.743ms     290.528us       0.000us         0.00%      19.809us       3.301us             6  
+                                            aten::copy_         1.88%      38.712us        82.12%       1.688ms     281.267us      16.961us        39.70%      19.809us       3.301us             6  
+                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      16.961us        39.70%      16.961us       2.827us             6  
+                                Activity Buffer Request        69.69%       1.432ms        69.69%       1.432ms       1.432ms       2.848us         6.67%       2.848us       2.848us             1  
+                                    aten::empty_strided         1.59%      32.771us         1.59%      32.771us       5.462us       0.000us         0.00%       0.000us       0.000us             6  
+                                        cudaMemcpyAsync        10.54%     216.613us        10.54%     216.613us      36.102us       0.000us         0.00%       0.000us       0.000us             6  
+                                            aten::slice         2.51%      51.572us         3.00%      61.672us       5.139us       0.000us         0.00%       0.000us       0.000us            12  
+                                       aten::as_strided         0.49%      10.100us         0.49%      10.100us       0.842us       0.000us         0.00%       0.000us       0.000us            12  
+                                       cudaLaunchKernel         2.21%      45.430us         2.21%      45.430us       7.572us       0.000us         0.00%       0.000us       0.000us             6  
+                                  cudaDeviceSynchronize         0.24%       4.849us         0.24%       4.849us       4.849us       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 819.280us
-Self CUDA time total: 42.750us
+Self CPU time total: 2.055ms
+Self CUDA time total: 42.720us
 
 
 
@@ -4414,23 +4641,23 @@ PROFILE TRACE: hf_kernels_rotary | cuda_B2_S512_H8_D64_R32
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     330.306us      1080.84%     330.306us     330.306us             1  
-                                      hf_kernels_rotary        18.42%     149.681us        99.42%     807.979us     807.979us       0.000us         0.00%      32.321us      32.321us             1  
-                          _rotary_dba7d1e::apply_rotary         5.10%      41.443us        10.19%      82.853us      13.809us      20.128us        65.86%      20.128us       3.355us             6  
-void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      20.128us        65.86%      20.128us       3.355us             6  
-                                            aten::clone         2.61%      21.203us        65.84%     535.084us      89.181us       0.000us         0.00%      12.193us       2.032us             6  
-                                            aten::copy_         4.51%      36.639us        59.42%     482.940us      80.490us      10.432us        34.14%      12.193us       2.032us             6  
-                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      10.432us        34.14%      10.432us       1.739us             6  
-                                Activity Buffer Request        29.60%     240.586us        29.60%     240.586us     240.586us       1.761us         5.76%       1.761us       1.761us             1  
-                                    aten::empty_strided         3.81%      30.941us         3.81%      30.941us       5.157us       0.000us         0.00%       0.000us       0.000us             6  
-                                        cudaMemcpyAsync        25.31%     205.715us        25.31%     205.715us      34.286us       0.000us         0.00%       0.000us       0.000us             6  
-                                            aten::slice         3.91%      31.761us         4.97%      40.361us       3.363us       0.000us         0.00%       0.000us       0.000us            12  
-                                       aten::as_strided         1.06%       8.600us         1.06%       8.600us       0.717us       0.000us         0.00%       0.000us       0.000us            12  
-                                       cudaLaunchKernel         5.10%      41.410us         5.10%      41.410us       6.902us       0.000us         0.00%       0.000us       0.000us             6  
-                                  cudaDeviceSynchronize         0.58%       4.711us         0.58%       4.711us       4.711us       0.000us         0.00%       0.000us       0.000us             1  
+                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     347.614us      1135.14%     347.614us     347.614us             1  
+                                      hf_kernels_rotary         7.64%     156.781us        99.77%       2.046ms       2.046ms       0.000us         0.00%      32.383us      32.383us             1  
+                          _rotary_dba7d1e::apply_rotary         2.01%      41.122us         4.16%      85.392us      14.232us      20.223us        66.04%      20.223us       3.370us             6  
+void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      20.223us        66.04%      20.223us       3.370us             6  
+                                            aten::clone         1.11%      22.841us        85.79%       1.759ms     293.223us       0.000us         0.00%      12.160us       2.027us             6  
+                                            aten::copy_         1.81%      37.030us        83.06%       1.703ms     283.910us      10.400us        33.96%      12.160us       2.027us             6  
+                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      10.400us        33.96%      10.400us       1.733us             6  
+                                Activity Buffer Request        70.68%       1.449ms        70.68%       1.449ms       1.449ms       1.760us         5.75%       1.760us       1.760us             1  
+                                    aten::empty_strided         1.61%      33.040us         1.61%      33.040us       5.507us       0.000us         0.00%       0.000us       0.000us             6  
+                                        cudaMemcpyAsync        10.58%     216.984us        10.58%     216.984us      36.164us       0.000us         0.00%       0.000us       0.000us             6  
+                                            aten::slice         1.70%      34.784us         2.17%      44.532us       3.711us       0.000us         0.00%       0.000us       0.000us            12  
+                                       aten::as_strided         0.48%       9.748us         0.48%       9.748us       0.812us       0.000us         0.00%       0.000us       0.000us            12  
+                                       cudaLaunchKernel         2.16%      44.270us         2.16%      44.270us       7.378us       0.000us         0.00%       0.000us       0.000us             6  
+                                  cudaDeviceSynchronize         0.23%       4.760us         0.23%       4.760us       4.760us       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 812.690us
-Self CUDA time total: 30.560us
+Self CPU time total: 2.051ms
+Self CUDA time total: 30.623us
 
 
 
@@ -4440,23 +4667,23 @@ PROFILE TRACE: hf_kernels_rotary | cuda_B2_S512_H8_D128_R64
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     339.800us       797.80%     339.800us     339.800us             1  
-                                      hf_kernels_rotary        14.79%     151.874us        99.55%       1.022ms       1.022ms       0.000us         0.00%      45.440us      45.440us             1  
-                          _rotary_dba7d1e::apply_rotary         4.15%      42.610us         8.29%      85.131us      14.188us      25.536us        59.95%      25.536us       4.256us             6  
-void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      25.536us        59.95%      25.536us       4.256us             6  
-                                            aten::clone         2.08%      21.390us        72.46%     743.968us     123.995us       0.000us         0.00%      19.904us       3.317us             6  
-                                            aten::copy_         3.85%      39.501us        67.34%     691.417us     115.236us      17.056us        40.05%      19.904us       3.317us             6  
-                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      17.056us        40.05%      17.056us       2.843us             6  
-                                Activity Buffer Request        43.74%     449.121us        43.74%     449.121us     449.121us       2.848us         6.69%       2.848us       2.848us             1  
-                                    aten::empty_strided         3.03%      31.161us         3.03%      31.161us       5.193us       0.000us         0.00%       0.000us       0.000us             6  
-                                        cudaMemcpyAsync        19.75%     202.795us        19.75%     202.795us      33.799us       0.000us         0.00%       0.000us       0.000us             6  
-                                            aten::slice         3.15%      32.321us         4.01%      41.121us       3.427us       0.000us         0.00%       0.000us       0.000us            12  
-                                       aten::as_strided         0.86%       8.800us         0.86%       8.800us       0.733us       0.000us         0.00%       0.000us       0.000us            12  
-                                       cudaLaunchKernel         4.14%      42.521us         4.14%      42.521us       7.087us       0.000us         0.00%       0.000us       0.000us             6  
-                                  cudaDeviceSynchronize         0.45%       4.640us         0.45%       4.640us       4.640us       0.000us         0.00%       0.000us       0.000us             1  
+                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     328.444us       771.23%     328.444us     328.444us             1  
+                                      hf_kernels_rotary        18.84%     150.934us        99.38%     796.084us     796.084us       0.000us         0.00%      45.403us      45.403us             1  
+                          _rotary_dba7d1e::apply_rotary         5.06%      40.529us        10.59%      84.811us      14.135us      25.693us        60.33%      25.693us       4.282us             6  
+void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      25.693us        60.33%      25.693us       4.282us             6  
+                                            aten::clone         2.49%      19.929us        64.90%     519.868us      86.645us       0.000us         0.00%      19.710us       3.285us             6  
+                                            aten::copy_         4.41%      35.321us        58.57%     469.148us      78.191us      16.894us        39.67%      19.710us       3.285us             6  
+                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      16.894us        39.67%      16.894us       2.816us             6  
+                                Activity Buffer Request        27.59%     221.013us        27.59%     221.013us     221.013us       2.816us         6.61%       2.816us       2.816us             1  
+                                    aten::empty_strided         3.84%      30.791us         3.84%      30.791us       5.132us       0.000us         0.00%       0.000us       0.000us             6  
+                                        cudaMemcpyAsync        26.57%     212.814us        26.57%     212.814us      35.469us       0.000us         0.00%       0.000us       0.000us             6  
+                                            aten::slice         3.92%      31.361us         5.05%      40.471us       3.373us       0.000us         0.00%       0.000us       0.000us            12  
+                                       aten::as_strided         1.14%       9.110us         1.14%       9.110us       0.759us       0.000us         0.00%       0.000us       0.000us            12  
+                                       cudaLaunchKernel         5.53%      44.282us         5.53%      44.282us       7.380us       0.000us         0.00%       0.000us       0.000us             6  
+                                  cudaDeviceSynchronize         0.62%       4.951us         0.62%       4.951us       4.951us       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 1.027ms
-Self CUDA time total: 42.592us
+Self CPU time total: 801.035us
+Self CUDA time total: 42.587us
 
 
 
@@ -4466,23 +4693,23 @@ PROFILE TRACE: hf_kernels_rotary | cuda_B2_S512_H32_D64_R32
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     339.355us       384.24%     339.355us     339.355us             1  
-                                      hf_kernels_rotary        17.67%     151.144us        99.35%     849.740us     849.740us       0.000us         0.00%     103.359us     103.359us             1  
-                                            aten::clone         2.51%      21.431us        67.11%     573.953us      95.659us       0.000us         0.00%      62.527us      10.421us             6  
-                                            aten::copy_         4.55%      38.879us        60.94%     521.222us      86.870us      47.487us        53.77%      62.527us      10.421us             6  
-                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      47.487us        53.77%      47.487us       7.915us             6  
-                          _rotary_dba7d1e::apply_rotary         5.04%      43.111us         9.90%      84.683us      14.114us      40.832us        46.23%      40.832us       6.805us             6  
-void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      40.832us        46.23%      40.832us       6.805us             6  
-                                Activity Buffer Request        33.08%     282.937us        33.08%     282.937us     282.937us      15.040us        17.03%      15.040us      15.040us             1  
-                                    aten::empty_strided         3.66%      31.300us         3.66%      31.300us       5.217us       0.000us         0.00%       0.000us       0.000us             6  
-                                        cudaMemcpyAsync        23.31%     199.406us        23.31%     199.406us      33.234us       0.000us         0.00%       0.000us       0.000us             6  
-                                            aten::slice         3.68%      31.469us         4.67%      39.960us       3.330us       0.000us         0.00%       0.000us       0.000us            12  
-                                       aten::as_strided         0.99%       8.491us         0.99%       8.491us       0.708us       0.000us         0.00%       0.000us       0.000us            12  
-                                       cudaLaunchKernel         4.86%      41.572us         4.86%      41.572us       6.929us       0.000us         0.00%       0.000us       0.000us             6  
-                                  cudaDeviceSynchronize         0.65%       5.540us         0.65%       5.540us       5.540us       0.000us         0.00%       0.000us       0.000us             1  
+                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     338.910us       380.70%     338.910us     338.910us             1  
+                                      hf_kernels_rotary        14.14%     150.935us        99.54%       1.062ms       1.062ms       0.000us         0.00%     104.734us     104.734us             1  
+                                            aten::clone         2.00%      21.371us        73.53%     784.703us     130.784us       0.000us         0.00%      63.775us      10.629us             6  
+                                            aten::copy_         3.58%      38.219us        68.59%     731.952us     121.992us      48.063us        53.99%      63.775us      10.629us             6  
+                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      48.063us        53.99%      48.063us       8.010us             6  
+                          _rotary_dba7d1e::apply_rotary         3.85%      41.059us         8.05%      85.950us      14.325us      40.959us        46.01%      40.959us       6.826us             6  
+void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      40.959us        46.01%      40.959us       6.826us             6  
+                                Activity Buffer Request        44.86%     478.699us        44.86%     478.699us     478.699us      15.712us        17.65%      15.712us      15.712us             1  
+                                    aten::empty_strided         2.94%      31.380us         2.94%      31.380us       5.230us       0.000us         0.00%       0.000us       0.000us             6  
+                                        cudaMemcpyAsync        20.15%     215.034us        20.15%     215.034us      35.839us       0.000us         0.00%       0.000us       0.000us             6  
+                                            aten::slice         2.96%      31.591us         3.81%      40.690us       3.391us       0.000us         0.00%       0.000us       0.000us            12  
+                                       aten::as_strided         0.85%       9.099us         0.85%       9.099us       0.758us       0.000us         0.00%       0.000us       0.000us            12  
+                                       cudaLaunchKernel         4.21%      44.891us         4.21%      44.891us       7.482us       0.000us         0.00%       0.000us       0.000us             6  
+                                  cudaDeviceSynchronize         0.46%       4.900us         0.46%       4.900us       4.900us       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 855.280us
-Self CUDA time total: 88.319us
+Self CPU time total: 1.067ms
+Self CUDA time total: 89.022us
 
 
 
@@ -4492,23 +4719,23 @@ PROFILE TRACE: hf_kernels_rotary | cuda_B2_S512_H32_D128_R64
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     340.473us       232.72%     340.473us     340.473us             1  
-                                      hf_kernels_rotary        20.27%     170.703us        99.41%     837.069us     837.069us       0.000us         0.00%     170.267us     170.267us             1  
-                                            aten::clone         2.43%      20.451us        64.32%     541.612us      90.269us       0.000us         0.00%     106.876us      17.813us             6  
-                                            aten::copy_         4.52%      38.100us        58.32%     491.079us      81.846us      82.909us        56.67%     106.876us      17.813us             6  
-                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      82.909us        56.67%      82.909us      13.818us             6  
-                          _rotary_dba7d1e::apply_rotary         5.01%      42.193us        10.04%      84.563us      14.094us      63.391us        43.33%      63.391us      10.565us             6  
-void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      63.391us        43.33%      63.391us      10.565us             6  
-                                Activity Buffer Request        30.43%     256.245us        30.43%     256.245us     256.245us      23.967us        16.38%      23.967us      23.967us             1  
-                                    aten::empty_strided         3.57%      30.082us         3.57%      30.082us       5.014us       0.000us         0.00%       0.000us       0.000us             6  
-                                        cudaMemcpyAsync        23.36%     196.734us        23.36%     196.734us      32.789us       0.000us         0.00%       0.000us       0.000us             6  
-                                            aten::slice         3.79%      31.880us         4.77%      40.191us       3.349us       0.000us         0.00%       0.000us       0.000us            12  
-                                       aten::as_strided         0.99%       8.311us         0.99%       8.311us       0.693us       0.000us         0.00%       0.000us       0.000us            12  
-                                       cudaLaunchKernel         5.03%      42.370us         5.03%      42.370us       7.062us       0.000us         0.00%       0.000us       0.000us             6  
-                                  cudaDeviceSynchronize         0.59%       4.950us         0.59%       4.950us       4.950us       0.000us         0.00%       0.000us       0.000us             1  
+                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     336.057us       230.21%     336.057us     336.057us             1  
+                                      hf_kernels_rotary        18.72%     149.775us        99.40%     795.224us     795.224us       0.000us         0.00%     169.949us     169.949us             1  
+                                            aten::clone         2.52%      20.180us        65.04%     520.348us      86.725us       0.000us         0.00%     106.527us      17.755us             6  
+                                            aten::copy_         4.49%      35.890us        58.61%     468.868us      78.145us      82.559us        56.55%     106.527us      17.755us             6  
+                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      82.559us        56.55%      82.559us      13.760us             6  
+                          _rotary_dba7d1e::apply_rotary         5.12%      40.981us        10.49%      83.942us      13.990us      63.422us        43.45%      63.422us      10.570us             6  
+void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      63.422us        43.45%      63.422us      10.570us             6  
+                                Activity Buffer Request        27.82%     222.544us        27.82%     222.544us     222.544us      23.968us        16.42%      23.968us      23.968us             1  
+                                    aten::empty_strided         3.91%      31.300us         3.91%      31.300us       5.217us       0.000us         0.00%       0.000us       0.000us             6  
+                                        cudaMemcpyAsync        26.30%     210.434us        26.30%     210.434us      35.072us       0.000us         0.00%       0.000us       0.000us             6  
+                                            aten::slice         3.94%      31.518us         5.14%      41.159us       3.430us       0.000us         0.00%       0.000us       0.000us            12  
+                                       aten::as_strided         1.21%       9.641us         1.21%       9.641us       0.803us       0.000us         0.00%       0.000us       0.000us            12  
+                                       cudaLaunchKernel         5.37%      42.961us         5.37%      42.961us       7.160us       0.000us         0.00%       0.000us       0.000us             6  
+                                  cudaDeviceSynchronize         0.60%       4.790us         0.60%       4.790us       4.790us       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 842.019us
-Self CUDA time total: 146.300us
+Self CPU time total: 800.014us
+Self CUDA time total: 145.981us
 
 
 
@@ -4518,23 +4745,23 @@ PROFILE TRACE: hf_kernels_rotary | cuda_B2_S2048_H8_D64_R32
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     372.221us       493.09%     372.221us     372.221us             1  
-                                      hf_kernels_rotary         8.48%     174.314us        99.77%       2.050ms       2.050ms       0.000us         0.00%      82.399us      82.399us             1  
-                          _rotary_dba7d1e::apply_rotary         1.96%      40.332us         4.07%      83.672us      13.945us      41.887us        55.49%      41.887us       6.981us             6  
-void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      41.887us        55.49%      41.887us       6.981us             6  
-                                            aten::clone         1.40%      28.690us        85.15%       1.749ms     291.567us       0.000us         0.00%      40.512us       6.752us             6  
-                                            aten::copy_         2.00%      41.101us        81.21%       1.669ms     278.085us      33.600us        44.51%      40.512us       6.752us             6  
-                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      33.600us        44.51%      33.600us       5.600us             6  
-                                Activity Buffer Request        69.57%       1.429ms        69.57%       1.429ms       1.429ms       6.912us         9.16%       6.912us       6.912us             1  
-                                    aten::empty_strided         2.54%      52.203us         2.54%      52.203us       8.701us       0.000us         0.00%       0.000us       0.000us             6  
-                                        cudaMemcpyAsync         9.64%     198.094us         9.64%     198.094us      33.016us       0.000us         0.00%       0.000us       0.000us             6  
-                                            aten::slice         1.63%      33.409us         2.06%      42.400us       3.533us       0.000us         0.00%       0.000us       0.000us            12  
-                                       aten::as_strided         0.44%       8.991us         0.44%       8.991us       0.749us       0.000us         0.00%       0.000us       0.000us            12  
-                                       cudaLaunchKernel         2.11%      43.340us         2.11%      43.340us       7.223us       0.000us         0.00%       0.000us       0.000us             6  
-                                  cudaDeviceSynchronize         0.23%       4.681us         0.23%       4.681us       4.681us       0.000us         0.00%       0.000us       0.000us             1  
+                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     339.836us       451.90%     339.836us     339.836us             1  
+                                      hf_kernels_rotary        18.57%     150.269us        99.38%     804.154us     804.154us       0.000us         0.00%      81.986us      81.986us             1  
+                          _rotary_dba7d1e::apply_rotary         4.99%      40.401us        10.49%      84.862us      14.144us      41.601us        55.32%      41.601us       6.933us             6  
+void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      41.601us        55.32%      41.601us       6.933us             6  
+                                            aten::clone         2.54%      20.532us        64.81%     524.439us      87.406us       0.000us         0.00%      40.385us       6.731us             6  
+                                            aten::copy_         4.41%      35.708us        58.24%     471.217us      78.536us      33.601us        44.68%      40.385us       6.731us             6  
+                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      33.601us        44.68%      33.601us       5.600us             6  
+                                Activity Buffer Request        27.71%     224.174us        27.71%     224.174us     224.174us       6.784us         9.02%       6.784us       6.784us             1  
+                                    aten::empty_strided         4.04%      32.690us         4.04%      32.690us       5.448us       0.000us         0.00%       0.000us       0.000us             6  
+                                        cudaMemcpyAsync        26.12%     211.335us        26.12%     211.335us      35.223us       0.000us         0.00%       0.000us       0.000us             6  
+                                            aten::slice         4.32%      34.924us         5.51%      44.584us       3.715us       0.000us         0.00%       0.000us       0.000us            12  
+                                       aten::as_strided         1.19%       9.660us         1.19%       9.660us       0.805us       0.000us         0.00%       0.000us       0.000us            12  
+                                       cudaLaunchKernel         5.49%      44.461us         5.49%      44.461us       7.410us       0.000us         0.00%       0.000us       0.000us             6  
+                                  cudaDeviceSynchronize         0.62%       4.981us         0.62%       4.981us       4.981us       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 2.054ms
-Self CUDA time total: 75.487us
+Self CPU time total: 809.135us
+Self CUDA time total: 75.202us
 
 
 
@@ -4544,23 +4771,23 @@ PROFILE TRACE: hf_kernels_rotary | cuda_B2_S2048_H8_D128_R64
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     352.925us       241.86%     352.925us     352.925us             1  
-                                      hf_kernels_rotary         8.45%     172.475us        99.74%       2.036ms       2.036ms       0.000us         0.00%     169.664us     169.664us             1  
-                                            aten::clone         1.31%      26.731us        85.08%       1.737ms     289.500us       0.000us         0.00%     105.664us      17.611us             6  
-                                            aten::copy_         1.93%      39.410us        82.15%       1.677ms     279.535us      81.920us        56.14%     105.664us      17.611us             6  
-                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      81.920us        56.14%      81.920us      13.653us             6  
-                          _rotary_dba7d1e::apply_rotary         2.10%      42.891us         4.20%      85.681us      14.280us      64.000us        43.86%      64.000us      10.667us             6  
-void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      64.000us        43.86%      64.000us      10.667us             6  
-                                Activity Buffer Request        70.60%       1.441ms        70.60%       1.441ms       1.441ms      23.744us        16.27%      23.744us      23.744us             1  
-                                    aten::empty_strided         1.62%      33.061us         1.62%      33.061us       5.510us       0.000us         0.00%       0.000us       0.000us             6  
-                                        cudaMemcpyAsync         9.62%     196.364us         9.62%     196.364us      32.727us       0.000us         0.00%       0.000us       0.000us             6  
-                                            aten::slice         1.58%      32.223us         2.02%      41.242us       3.437us       0.000us         0.00%       0.000us       0.000us            12  
-                                       aten::as_strided         0.44%       9.019us         0.44%       9.019us       0.752us       0.000us         0.00%       0.000us       0.000us            12  
-                                       cudaLaunchKernel         2.10%      42.790us         2.10%      42.790us       7.132us       0.000us         0.00%       0.000us       0.000us             6  
-                                  cudaDeviceSynchronize         0.26%       5.250us         0.26%       5.250us       5.250us       0.000us         0.00%       0.000us       0.000us             1  
+                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     372.859us       256.14%     372.859us     372.859us             1  
+                                      hf_kernels_rotary        18.64%     161.451us        99.43%     861.125us     861.125us       0.000us         0.00%     169.279us     169.279us             1  
+                                            aten::clone         2.47%      21.401us        63.58%     550.631us      91.772us       0.000us         0.00%     105.373us      17.562us             6  
+                                            aten::copy_         4.30%      37.239us        57.31%     496.359us      82.727us      81.662us        56.10%     105.373us      17.562us             6  
+                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us      81.662us        56.10%      81.662us      13.610us             6  
+                          _rotary_dba7d1e::apply_rotary         5.12%      44.341us        12.24%     106.023us      17.671us      63.906us        43.90%      63.906us      10.651us             6  
+void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us      63.906us        43.90%      63.906us      10.651us             6  
+                                Activity Buffer Request        28.62%     247.854us        28.62%     247.854us     247.854us      23.711us        16.29%      23.711us      23.711us             1  
+                                    aten::empty_strided         3.80%      32.871us         3.80%      32.871us       5.479us       0.000us         0.00%       0.000us       0.000us             6  
+                                        cudaMemcpyAsync        24.39%     211.266us        24.39%     211.266us      35.211us       0.000us         0.00%       0.000us       0.000us             6  
+                                            aten::slice         3.88%      33.609us         4.97%      43.020us       3.585us       0.000us         0.00%       0.000us       0.000us            12  
+                                       aten::as_strided         1.09%       9.411us         1.09%       9.411us       0.784us       0.000us         0.00%       0.000us       0.000us            12  
+                                       cudaLaunchKernel         7.12%      61.682us         7.12%      61.682us      10.280us       0.000us         0.00%       0.000us       0.000us             6  
+                                  cudaDeviceSynchronize         0.57%       4.969us         0.57%       4.969us       4.969us       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 2.042ms
-Self CUDA time total: 145.920us
+Self CPU time total: 866.094us
+Self CUDA time total: 145.568us
 
 
 
@@ -4570,23 +4797,23 @@ PROFILE TRACE: hf_kernels_rotary | cuda_B2_S2048_H32_D64_R32
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                      hf_kernels_rotary        19.65%     213.486us        76.93%     835.739us     835.739us       0.000us         0.00%     746.680us     746.680us             1  
-                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     687.960us       101.16%     687.960us     687.960us             1  
-                                            aten::clone         1.98%      21.549us        45.41%     493.341us      82.224us       0.000us         0.00%     557.242us      92.874us             6  
-                                            aten::copy_         3.46%      37.632us        40.54%     440.431us      73.405us     490.619us        72.14%     557.242us      92.874us             6  
-                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us     490.619us        72.14%     490.619us      81.770us             6  
-                          _rotary_dba7d1e::apply_rotary         4.15%      45.032us         8.07%      87.672us      14.612us     189.438us        27.86%     189.438us      31.573us             6  
-void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us     189.438us        27.86%     189.438us      31.573us             6  
-                                Activity Buffer Request        19.06%     207.035us        19.06%     207.035us     207.035us      66.623us         9.80%      66.623us      66.623us             1  
-                                    aten::empty_strided         2.89%      31.361us         2.89%      31.361us       5.227us       0.000us         0.00%       0.000us       0.000us             6  
-                                        cudaMemcpyAsync        18.02%     195.764us        18.02%     195.764us      32.627us       0.000us         0.00%       0.000us       0.000us             6  
-                                            aten::slice         2.96%      32.192us         3.80%      41.240us       3.437us       0.000us         0.00%       0.000us       0.000us            12  
-                                       aten::as_strided         0.83%       9.048us         0.83%       9.048us       0.754us       0.000us         0.00%       0.000us       0.000us            12  
-                                       cudaLaunchKernel         3.93%      42.640us         3.93%      42.640us       7.107us       0.000us         0.00%       0.000us       0.000us             6  
-                                  cudaDeviceSynchronize        23.07%     250.625us        23.07%     250.625us     250.625us       0.000us         0.00%       0.000us       0.000us             1  
+                                      hf_kernels_rotary        13.02%     148.583us        72.32%     825.404us     825.404us       0.000us         0.00%     745.510us     745.510us             1  
+                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us     687.015us       101.19%     687.015us     687.015us             1  
+                                            aten::clone         1.76%      20.130us        47.96%     547.368us      91.228us       0.000us         0.00%     556.292us      92.715us             6  
+                                            aten::copy_         3.18%      36.280us        43.26%     493.818us      82.303us     489.699us        72.13%     556.292us      92.715us             6  
+                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us     489.699us        72.13%     489.699us      81.617us             6  
+                          _rotary_dba7d1e::apply_rotary         3.57%      40.732us         7.58%      86.552us      14.425us     189.218us        27.87%     189.218us      31.536us             6  
+void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us     189.218us        27.87%     189.218us      31.536us             6  
+                                Activity Buffer Request        21.89%     249.905us        21.89%     249.905us     249.905us      66.593us         9.81%      66.593us      66.593us             1  
+                                    aten::empty_strided         2.93%      33.420us         2.93%      33.420us       5.570us       0.000us         0.00%       0.000us       0.000us             6  
+                                        cudaMemcpyAsync        18.19%     207.633us        18.19%     207.633us      34.606us       0.000us         0.00%       0.000us       0.000us             6  
+                                            aten::slice         2.92%      33.351us         3.76%      42.901us       3.575us       0.000us         0.00%       0.000us       0.000us            12  
+                                       aten::as_strided         0.84%       9.550us         0.84%       9.550us       0.796us       0.000us         0.00%       0.000us       0.000us            12  
+                                       cudaLaunchKernel         4.01%      45.820us         4.01%      45.820us       7.637us       0.000us         0.00%       0.000us       0.000us             6  
+                                  cudaDeviceSynchronize        27.68%     315.986us        27.68%     315.986us     315.986us       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 1.086ms
-Self CUDA time total: 680.057us
+Self CPU time total: 1.141ms
+Self CUDA time total: 678.917us
 
 
 
@@ -4596,30 +4823,30 @@ PROFILE TRACE: hf_kernels_rotary | cuda_B2_S2048_H32_D128_R64
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                      hf_kernels_rotary         5.18%     149.334us        27.72%     799.139us     799.139us       0.000us         0.00%       2.625ms       2.625ms             1  
-                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us       2.454ms       100.33%       2.454ms       2.454ms             1  
-                                            aten::clone         0.74%      21.311us        18.12%     522.352us      87.059us       0.000us         0.00%       1.393ms     232.143us             6  
-                                            aten::copy_         1.34%      38.611us        16.30%     469.821us      78.303us       1.214ms        49.62%       1.393ms     232.143us             6  
-                          _rotary_dba7d1e::apply_rotary         1.48%      42.661us         3.03%      87.271us      14.545us       1.232ms        50.38%       1.232ms     205.327us             6  
-void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us       1.232ms        50.38%       1.232ms     205.327us             6  
-                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us       1.214ms        49.62%       1.214ms     202.255us             6  
-                                Activity Buffer Request         8.25%     237.786us         8.25%     237.786us     237.786us     179.327us         7.33%     179.327us     179.327us             1  
-                                    aten::empty_strided         1.08%      31.220us         1.08%      31.220us       5.203us       0.000us         0.00%       0.000us       0.000us             6  
-                                        cudaMemcpyAsync         6.71%     193.424us         6.71%     193.424us      32.237us       0.000us         0.00%       0.000us       0.000us             6  
-                                            aten::slice         1.11%      31.861us         1.39%      40.182us       3.349us       0.000us         0.00%       0.000us       0.000us            12  
-                                       aten::as_strided         0.29%       8.321us         0.29%       8.321us       0.693us       0.000us         0.00%       0.000us       0.000us            12  
-                                       cudaLaunchKernel         1.55%      44.610us         1.55%      44.610us       7.435us       0.000us         0.00%       0.000us       0.000us             6  
-                                  cudaDeviceSynchronize        72.28%       2.083ms        72.28%       2.083ms       2.083ms       0.000us         0.00%       0.000us       0.000us             1  
+                                      hf_kernels_rotary         5.26%     153.062us        28.60%     832.074us     832.074us       0.000us         0.00%       2.627ms       2.627ms             1  
+                                      hf_kernels_rotary         0.00%       0.000us         0.00%       0.000us       0.000us       2.451ms       100.32%       2.451ms       2.451ms             1  
+                                            aten::clone         0.71%      20.751us        18.92%     550.432us      91.739us       0.000us         0.00%       1.403ms     233.752us             6  
+                                            aten::copy_         1.33%      38.628us        17.10%     497.389us      82.898us       1.219ms        49.87%       1.403ms     233.752us             6  
+                          _rotary_dba7d1e::apply_rotary         1.42%      41.449us         2.89%      84.050us      14.008us       1.225ms        50.13%       1.225ms     204.141us             6  
+void at::native::(anonymous namespace)::unrolled_ele...         0.00%       0.000us         0.00%       0.000us       0.000us       1.225ms        50.13%       1.225ms     204.141us             6  
+                         Memcpy DtoD (Device -&gt; Device)         0.00%       0.000us         0.00%       0.000us       0.000us       1.219ms        49.87%       1.219ms     203.112us             6  
+                                Activity Buffer Request         8.62%     250.725us         8.62%     250.725us     250.725us     183.838us         7.52%     183.838us     183.838us             1  
+                                    aten::empty_strided         1.11%      32.292us         1.11%      32.292us       5.382us       0.000us         0.00%       0.000us       0.000us             6  
+                                        cudaMemcpyAsync         7.15%     208.036us         7.15%     208.036us      34.673us       0.000us         0.00%       0.000us       0.000us             6  
+                                            aten::slice         1.18%      34.219us         1.53%      44.530us       3.711us       0.000us         0.00%       0.000us       0.000us            12  
+                                       aten::as_strided         0.35%      10.311us         0.35%      10.311us       0.859us       0.000us         0.00%       0.000us       0.000us            12  
+                                       cudaLaunchKernel         1.46%      42.601us         1.46%      42.601us       7.100us       0.000us         0.00%       0.000us       0.000us             6  
+                                  cudaDeviceSynchronize        71.40%       2.077ms        71.40%       2.077ms       2.077ms       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 2.882ms
-Self CUDA time total: 2.445ms
+Self CPU time total: 2.909ms
+Self CUDA time total: 2.444ms
 
 
 impl                     wl                  p50(ms)  ok
 hf_kernels_rotary        cuda_B1_S128_H32_D128_R64     0.09  True
 hf_kernels_rotary        cuda_B1_S128_H32_D64_R32     0.09  True
 hf_kernels_rotary        cuda_B1_S128_H8_D128_R64     0.09  True
-hf_kernels_rotary        cuda_B1_S128_H8_D64_R32     0.07  True
+hf_kernels_rotary        cuda_B1_S128_H8_D64_R32     0.08  True
 hf_kernels_rotary        cuda_B1_S2048_H32_D128_R64     0.26  True
 hf_kernels_rotary        cuda_B1_S2048_H32_D64_R32     0.09  True
 hf_kernels_rotary        cuda_B1_S2048_H8_D128_R64     0.09  True
@@ -4632,7 +4859,7 @@ hf_kernels_rotary        cuda_B2_S128_H32_D128_R64     0.09  True
 hf_kernels_rotary        cuda_B2_S128_H32_D64_R32     0.09  True
 hf_kernels_rotary        cuda_B2_S128_H8_D128_R64     0.09  True
 hf_kernels_rotary        cuda_B2_S128_H8_D64_R32     0.09  True
-hf_kernels_rotary        cuda_B2_S2048_H32_D128_R64     0.84  True
+hf_kernels_rotary        cuda_B2_S2048_H32_D128_R64     0.85  True
 hf_kernels_rotary        cuda_B2_S2048_H32_D64_R32     0.26  True
 hf_kernels_rotary        cuda_B2_S2048_H8_D128_R64     0.09  True
 hf_kernels_rotary        cuda_B2_S2048_H8_D64_R32     0.09  True
@@ -4644,12 +4871,14 @@ hf_kernels_rotary        cuda_B2_S512_H8_D64_R32     0.09  True
 <div class="uv-install-logs" id="uv-logs-benchmark">
 <div class="uv-logs-header" onclick="toggleUvLogs(this)">▶ UV Install Logs</div>
 <div class="uv-logs-content" style="display: none;">
-Installed 15 packages in 15ms
+Downloading hf-xet (3.2MiB)
+ Downloading hf-xet
+Installed 52 packages in 244ms
 </div>
 </div>
 <div class="cell-stderr">Fetching 5 files:   0%|          | 0/5 [00:00&lt;?, ?it/s]
-Fetching 5 files: 100%|██████████| 5/5 [00:00&lt;00:00, 15.01it/s]
-Fetching 5 files: 100%|██████████| 5/5 [00:00&lt;00:00, 15.00it/s]</div>
+Fetching 5 files: 100%|██████████| 5/5 [00:00&lt;00:00, 15.33it/s]
+Fetching 5 files: 100%|██████████| 5/5 [00:00&lt;00:00, 15.31it/s]</div>
 <div class="cell-artifacts">
 <h4>Artifacts:</h4>
 <a href="artifacts/benchmark/rotary.jsonl" class="artifact" target="_blank">rotary.jsonl</a>