diff --git "a/flash_attn/impls/flash_attention.html" "b/flash_attn/impls/flash_attention.html"
--- "a/flash_attn/impls/flash_attention.html"
+++ "b/flash_attn/impls/flash_attention.html"
@@ -4,6 +4,11 @@
     <meta charset="UTF-8">
     <meta name="viewport" content="width=device-width, initial-scale=1.0">
     <title>flash_attention</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 | 0.26s
+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>
@@ -3896,7 +4123,7 @@ Cell: nv | 0.26s
 </div>
 </div>
 <div id="output-nv" class="cell-output">
-<div class="cell-stdout"><pre class="stdout-text">Wed Oct 29 15:50:02 2025       
+<div class="cell-stdout"><pre class="stdout-text">Thu Oct 30 15:52:36 2025       
 +-----------------------------------------------------------------------------------------+
 | NVIDIA-SMI 570.195.03             Driver Version: 570.195.03     CUDA Version: 12.8     |
 |-----------------------------------------+------------------------+----------------------+
@@ -3905,7 +4132,7 @@ Cell: nv | 0.26s
 |                                         |                        |               MIG M. |
 |=========================================+========================+======================|
 |   0  NVIDIA L40S                    On  |   00000000:4D:00.0 Off |                    0 |
-| N/A   29C    P0            165W /  350W |       0MiB /  46068MiB |     61%      Default |
+| N/A   30C    P0             75W /  350W |       0MiB /  46068MiB |     11%      Default |
 |                                         |                        |                  N/A |
 +-----------------------------------------+------------------------+----------------------+
 
@@ -3927,9 +4154,9 @@ Cell: nv | 0.26s
 <span class="collapse-indicators">
 <span onclick="toggleCode('benchmark')" style="cursor: pointer;">▼ code</span> 
 <span onclick="toggleOutput('benchmark')" style="cursor: pointer;">▼ output</span>
- <span id="uv-indicator-benchmark" style="cursor: default; opacity: 0.3;">▶ uv-logs</span>
+ <span id="uv-indicator-benchmark" onclick="toggleUvLogsFromHeader('benchmark')" style="cursor: pointer;">▶ uv-logs</span>
 </span> | 
-Cell: benchmark | 3.82s
+Cell: benchmark | 7.50s
  | <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>
@@ -3980,29 +4207,29 @@ PROFILE TRACE: torch_flash_ma | cuda_attn_L128_bfloat16
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                         torch_flash_ma         0.00%       0.000us         0.00%       0.000us       0.000us       3.562ms       101.45%       3.562ms       3.562ms             1  
-                                         torch_flash_ma         6.38%     328.580us        45.84%       2.360ms       2.360ms       0.000us         0.00%       3.551ms       3.551ms             1  
-                     aten::scaled_dot_product_attention         0.79%      40.571us         4.12%     212.315us      70.772us       0.000us         0.00%       2.798ms     932.779us             3  
-              aten::_scaled_dot_product_flash_attention         0.52%      26.642us         3.34%     171.744us      57.248us       0.000us         0.00%       2.798ms     932.779us             3  
-                         aten::_flash_attention_forward         0.74%      37.939us         2.40%     123.383us      41.128us       2.798ms        79.71%       2.798ms     932.779us             3  
-void pytorch_flash::flash_fwd_kernel&lt;Flash_fwd_kerne...         0.00%       0.000us         0.00%       0.000us       0.000us       2.798ms        79.71%       2.798ms     932.779us             3  
-                                       aten::contiguous         0.27%      13.720us        34.12%       1.757ms     146.409us       0.000us         0.00%     752.288us      62.691us            12  
-                                            aten::clone         0.73%      37.449us        33.85%       1.743ms     145.266us       0.000us         0.00%     752.288us      62.691us            12  
-                                            aten::copy_         1.68%      86.484us        31.57%       1.625ms     135.456us     712.095us        20.29%     752.288us      62.691us            12  
-void at::native::elementwise_kernel&lt;128, 4, at::nati...         0.00%       0.000us         0.00%       0.000us       0.000us     712.095us        20.29%     712.095us      59.341us            12  
-                                Activity Buffer Request        28.00%       1.442ms        28.00%       1.442ms       1.442ms      40.193us         1.14%      40.193us      40.193us             1  
-                                        aten::transpose         1.22%      62.637us         1.64%      84.218us       3.509us       0.000us         0.00%       0.000us       0.000us            24  
-                                       aten::as_strided         0.42%      21.581us         0.42%      21.581us       0.899us       0.000us         0.00%       0.000us       0.000us            24  
-                                       aten::empty_like         0.48%      24.619us         1.97%     101.523us       6.768us       0.000us         0.00%       0.000us       0.000us            15  
-                                            aten::empty         1.76%      90.465us         1.76%      90.465us       3.769us       0.000us         0.00%       0.000us       0.000us            24  
-                                       cudaLaunchKernel         2.36%     121.521us         2.36%     121.521us       8.101us       0.000us         0.00%       0.000us       0.000us            15  
-                                    aten::empty_strided         0.31%      15.721us         0.31%      15.721us       5.240us       0.000us         0.00%       0.000us       0.000us             3  
-                                 cudaDeviceGetAttribute         0.04%       2.280us         0.04%       2.280us       0.380us       0.000us         0.00%       0.000us       0.000us             6  
-                                   cudaFuncSetAttribute         0.16%       8.181us         0.16%       8.181us       2.727us       0.000us         0.00%       0.000us       0.000us             3  
-                                  cudaDeviceSynchronize        54.16%       2.789ms        54.16%       2.789ms       2.789ms       0.000us         0.00%       0.000us       0.000us             1  
+                                         torch_flash_ma         0.00%       0.000us         0.00%       0.000us       0.000us       3.587ms       102.23%       3.587ms       3.587ms             1  
+                                         torch_flash_ma         7.11%     370.236us        47.42%       2.468ms       2.468ms       0.000us         0.00%       3.549ms       3.549ms             1  
+                     aten::scaled_dot_product_attention         0.85%      44.391us         4.44%     231.334us      77.111us       0.000us         0.00%       2.791ms     930.498us             3  
+              aten::_scaled_dot_product_flash_attention         0.51%      26.381us         3.59%     186.943us      62.314us       0.000us         0.00%       2.791ms     930.498us             3  
+                         aten::_flash_attention_forward         0.76%      39.658us         2.57%     134.002us      44.667us       2.791ms        79.55%       2.791ms     930.498us             3  
+void pytorch_flash::flash_fwd_kernel&lt;Flash_fwd_kerne...         0.00%       0.000us         0.00%       0.000us       0.000us       2.791ms        79.55%       2.791ms     930.498us             3  
+                                       aten::contiguous         0.30%      15.641us        34.37%       1.789ms     149.098us       0.000us         0.00%     757.697us      63.141us            12  
+                                            aten::clone         0.74%      38.596us        34.07%       1.774ms     147.794us       0.000us         0.00%     757.697us      63.141us            12  
+                                            aten::copy_         1.78%      92.553us        31.63%       1.647ms     137.218us     717.505us        20.45%     757.697us      63.141us            12  
+void at::native::elementwise_kernel&lt;128, 4, at::nati...         0.00%       0.000us         0.00%       0.000us       0.000us     717.505us        20.45%     717.505us      59.792us            12  
+                                Activity Buffer Request        27.90%       1.452ms        27.90%       1.452ms       1.452ms      40.192us         1.15%      40.192us      40.192us             1  
+                                        aten::transpose         1.49%      77.390us         2.00%     104.302us       4.346us       0.000us         0.00%       0.000us       0.000us            24  
+                                       aten::as_strided         0.52%      26.912us         0.52%      26.912us       1.121us       0.000us         0.00%       0.000us       0.000us            24  
+                                       aten::empty_like         0.55%      28.453us         2.13%     110.953us       7.397us       0.000us         0.00%       0.000us       0.000us            15  
+                                            aten::empty         1.93%     100.211us         1.93%     100.211us       4.175us       0.000us         0.00%       0.000us       0.000us            24  
+                                       cudaLaunchKernel         2.45%     127.363us         2.45%     127.363us       8.491us       0.000us         0.00%       0.000us       0.000us            15  
+                                    aten::empty_strided         0.32%      16.580us         0.32%      16.580us       5.527us       0.000us         0.00%       0.000us       0.000us             3  
+                                 cudaDeviceGetAttribute         0.05%       2.441us         0.05%       2.441us       0.407us       0.000us         0.00%       0.000us       0.000us             6  
+                                   cudaFuncSetAttribute         0.18%       9.241us         0.18%       9.241us       3.080us       0.000us         0.00%       0.000us       0.000us             3  
+                                  cudaDeviceSynchronize        52.58%       2.737ms        52.58%       2.737ms       2.737ms       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 5.149ms
-Self CUDA time total: 3.510ms
+Self CPU time total: 5.205ms
+Self CUDA time total: 3.509ms
 
 
 
@@ -4012,29 +4239,29 @@ PROFILE TRACE: torch_flash_ma | cuda_attn_L256_bfloat16
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                         torch_flash_ma         4.71%     257.538us        44.52%       2.436ms       2.436ms       0.000us         0.00%       3.763ms       3.763ms             1  
-                                         torch_flash_ma         0.00%       0.000us         0.00%       0.000us       0.000us       3.719ms       100.29%       3.719ms       3.719ms             1  
-                     aten::scaled_dot_product_attention         0.45%      24.440us         3.30%     180.683us      60.228us       0.000us         0.00%       2.948ms     982.525us             3  
-              aten::_scaled_dot_product_flash_attention         0.35%      18.890us         2.86%     156.243us      52.081us       0.000us         0.00%       2.948ms     982.525us             3  
-                         aten::_flash_attention_forward         0.68%      37.218us         2.07%     113.133us      37.711us       2.948ms        79.49%       2.948ms     982.525us             3  
-void pytorch_flash::flash_fwd_kernel&lt;Flash_fwd_kerne...         0.00%       0.000us         0.00%       0.000us       0.000us       2.948ms        79.49%       2.948ms     982.525us             3  
-                                       aten::contiguous         0.16%       8.651us        35.72%       1.955ms     162.890us       0.000us         0.00%     815.678us      67.973us            12  
-                                            aten::clone         0.48%      26.452us        35.56%       1.946ms     162.169us       0.000us         0.00%     815.678us      67.973us            12  
-                                            aten::copy_         1.81%      99.279us        33.97%       1.859ms     154.885us     760.479us        20.51%     815.678us      67.973us            12  
-void at::native::elementwise_kernel&lt;128, 4, at::nati...         0.00%       0.000us         0.00%       0.000us       0.000us     760.479us        20.51%     760.479us      63.373us            12  
-                                Activity Buffer Request        30.60%       1.674ms        30.60%       1.674ms       1.674ms      55.199us         1.49%      55.199us      55.199us             1  
-                                        aten::transpose         0.92%      50.270us         1.23%      67.460us       2.811us       0.000us         0.00%       0.000us       0.000us            24  
-                                       aten::as_strided         0.31%      17.190us         0.31%      17.190us       0.716us       0.000us         0.00%       0.000us       0.000us            24  
-                                       aten::empty_like         0.34%      18.723us         1.45%      79.503us       5.300us       0.000us         0.00%       0.000us       0.000us            15  
-                                            aten::empty         1.39%      75.933us         1.39%      75.933us       3.164us       0.000us         0.00%       0.000us       0.000us            24  
-                                       cudaLaunchKernel         1.98%     108.143us         1.98%     108.143us       7.210us       0.000us         0.00%       0.000us       0.000us            15  
-                                    aten::empty_strided         0.25%      13.599us         0.25%      13.599us       4.533us       0.000us         0.00%       0.000us       0.000us             3  
-                                 cudaDeviceGetAttribute         0.03%       1.831us         0.03%       1.831us       0.305us       0.000us         0.00%       0.000us       0.000us             6  
-                                   cudaFuncSetAttribute         0.07%       3.690us         0.07%       3.690us       1.230us       0.000us         0.00%       0.000us       0.000us             3  
-                                  cudaDeviceSynchronize        55.48%       3.036ms        55.48%       3.036ms       3.036ms       0.000us         0.00%       0.000us       0.000us             1  
+                                         torch_flash_ma         4.72%     248.136us        41.78%       2.196ms       2.196ms       0.000us         0.00%       3.803ms       3.803ms             1  
+                                         torch_flash_ma         0.00%       0.000us         0.00%       0.000us       0.000us       3.759ms       100.28%       3.759ms       3.759ms             1  
+                     aten::scaled_dot_product_attention         0.51%      26.852us         3.40%     178.734us      59.578us       0.000us         0.00%       2.990ms     996.607us             3  
+              aten::_scaled_dot_product_flash_attention         0.35%      18.418us         2.89%     151.882us      50.627us       0.000us         0.00%       2.990ms     996.607us             3  
+                         aten::_flash_attention_forward         0.65%      34.063us         2.10%     110.562us      36.854us       2.990ms        79.76%       2.990ms     996.607us             3  
+void pytorch_flash::flash_fwd_kernel&lt;Flash_fwd_kerne...         0.00%       0.000us         0.00%       0.000us       0.000us       2.990ms        79.76%       2.990ms     996.607us             3  
+                                       aten::contiguous         0.19%      10.079us        32.75%       1.721ms     143.446us       0.000us         0.00%     813.629us      67.802us            12  
+                                            aten::clone         0.54%      28.151us        32.56%       1.711ms     142.606us       0.000us         0.00%     813.629us      67.802us            12  
+                                            aten::copy_         1.97%     103.281us        30.84%       1.621ms     135.084us     758.782us        20.24%     813.629us      67.802us            12  
+void at::native::elementwise_kernel&lt;128, 4, at::nati...         0.00%       0.000us         0.00%       0.000us       0.000us     758.782us        20.24%     758.782us      63.232us            12  
+                                Activity Buffer Request        27.29%       1.434ms        27.29%       1.434ms       1.434ms      54.847us         1.46%      54.847us      54.847us             1  
+                                        aten::transpose         0.98%      51.741us         1.34%      70.423us       2.934us       0.000us         0.00%       0.000us       0.000us            24  
+                                       aten::as_strided         0.36%      18.682us         0.36%      18.682us       0.778us       0.000us         0.00%       0.000us       0.000us            24  
+                                       aten::empty_like         0.38%      19.848us         1.54%      80.939us       5.396us       0.000us         0.00%       0.000us       0.000us            15  
+                                            aten::empty         1.45%      76.001us         1.45%      76.001us       3.167us       0.000us         0.00%       0.000us       0.000us            24  
+                                       cudaLaunchKernel         2.04%     106.952us         2.04%     106.952us       7.130us       0.000us         0.00%       0.000us       0.000us            15  
+                                    aten::empty_strided         0.26%      13.850us         0.26%      13.850us       4.617us       0.000us         0.00%       0.000us       0.000us             3  
+                                 cudaDeviceGetAttribute         0.04%       1.860us         0.04%       1.860us       0.310us       0.000us         0.00%       0.000us       0.000us             6  
+                                   cudaFuncSetAttribute         0.07%       3.760us         0.07%       3.760us       1.253us       0.000us         0.00%       0.000us       0.000us             3  
+                                  cudaDeviceSynchronize        58.22%       3.060ms        58.22%       3.060ms       3.060ms       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 5.472ms
-Self CUDA time total: 3.708ms
+Self CPU time total: 5.255ms
+Self CUDA time total: 3.749ms
 
 
 
@@ -4044,29 +4271,29 @@ PROFILE TRACE: torch_flash_ma | cuda_attn_L320_bfloat16
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                         torch_flash_ma         4.65%     248.558us        40.70%       2.176ms       2.176ms       0.000us         0.00%       3.868ms       3.868ms             1  
-                                         torch_flash_ma         0.00%       0.000us         0.00%       0.000us       0.000us       3.819ms       100.29%       3.819ms       3.819ms             1  
-                     aten::scaled_dot_product_attention         0.45%      24.181us         3.36%     179.834us      59.945us       0.000us         0.00%       3.027ms       1.009ms             3  
-              aten::_scaled_dot_product_flash_attention         0.34%      18.100us         2.91%     155.653us      51.884us       0.000us         0.00%       3.027ms       1.009ms             3  
-                         aten::_flash_attention_forward         0.73%      38.760us         2.16%     115.412us      38.471us       3.027ms        79.48%       3.027ms       1.009ms             3  
-void pytorch_flash::flash_fwd_kernel&lt;Flash_fwd_kerne...         0.00%       0.000us         0.00%       0.000us       0.000us       3.027ms        79.48%       3.027ms       1.009ms             3  
-                                       aten::contiguous         0.16%       8.609us        31.88%       1.704ms     142.018us       0.000us         0.00%     841.280us      70.107us            12  
-                                            aten::clone         0.50%      26.820us        31.72%       1.696ms     141.301us       0.000us         0.00%     841.280us      70.107us            12  
-                                            aten::copy_         1.47%      78.703us        30.10%       1.609ms     134.076us     781.631us        20.52%     841.280us      70.107us            12  
-void at::native::elementwise_kernel&lt;128, 4, at::nati...         0.00%       0.000us         0.00%       0.000us       0.000us     781.631us        20.52%     781.631us      65.136us            12  
-                                Activity Buffer Request        27.11%       1.449ms        27.11%       1.449ms       1.449ms      59.649us         1.57%      59.649us      59.649us             1  
-                                        aten::transpose         0.90%      48.151us         1.22%      65.102us       2.713us       0.000us         0.00%       0.000us       0.000us            24  
-                                       aten::as_strided         0.32%      16.951us         0.32%      16.951us       0.706us       0.000us         0.00%       0.000us       0.000us            24  
-                                       aten::empty_like         0.35%      18.789us         1.49%      79.862us       5.324us       0.000us         0.00%       0.000us       0.000us            15  
-                                            aten::empty         1.38%      73.892us         1.38%      73.892us       3.079us       0.000us         0.00%       0.000us       0.000us            24  
-                                       cudaLaunchKernel         1.96%     104.680us         1.96%     104.680us       6.979us       0.000us         0.00%       0.000us       0.000us            15  
-                                    aten::empty_strided         0.28%      15.081us         0.28%      15.081us       5.027us       0.000us         0.00%       0.000us       0.000us             3  
-                                 cudaDeviceGetAttribute         0.03%       1.791us         0.03%       1.791us       0.299us       0.000us         0.00%       0.000us       0.000us             6  
-                                   cudaFuncSetAttribute         0.07%       3.500us         0.07%       3.500us       1.167us       0.000us         0.00%       0.000us       0.000us             3  
-                                  cudaDeviceSynchronize        59.30%       3.169ms        59.30%       3.169ms       3.169ms       0.000us         0.00%       0.000us       0.000us             1  
+                                         torch_flash_ma         4.59%     242.054us        41.69%       2.201ms       2.201ms       0.000us         0.00%       3.795ms       3.795ms             1  
+                                         torch_flash_ma         0.00%       0.000us         0.00%       0.000us       0.000us       3.746ms       100.27%       3.746ms       3.746ms             1  
+                     aten::scaled_dot_product_attention         0.50%      26.150us         3.40%     179.413us      59.804us       0.000us         0.00%       2.957ms     985.581us             3  
+              aten::_scaled_dot_product_flash_attention         0.35%      18.371us         2.90%     153.263us      51.088us       0.000us         0.00%       2.957ms     985.581us             3  
+                         aten::_flash_attention_forward         0.64%      34.041us         2.11%     111.213us      37.071us       2.957ms        79.14%       2.957ms     985.581us             3  
+void pytorch_flash::flash_fwd_kernel&lt;Flash_fwd_kerne...         0.00%       0.000us         0.00%       0.000us       0.000us       2.957ms        79.14%       2.957ms     985.581us             3  
+                                       aten::contiguous         0.19%       9.991us        32.85%       1.734ms     144.489us       0.000us         0.00%     838.147us      69.846us            12  
+                                            aten::clone         0.52%      27.541us        32.66%       1.724ms     143.657us       0.000us         0.00%     838.147us      69.846us            12  
+                                            aten::copy_         1.47%      77.641us        30.91%       1.632ms     135.987us     779.363us        20.86%     838.147us      69.846us            12  
+void at::native::elementwise_kernel&lt;128, 4, at::nati...         0.00%       0.000us         0.00%       0.000us       0.000us     779.363us        20.86%     779.363us      64.947us            12  
+                                Activity Buffer Request        27.89%       1.472ms        27.89%       1.472ms       1.472ms      58.784us         1.57%      58.784us      58.784us             1  
+                                        aten::transpose         0.96%      50.819us         1.31%      69.110us       2.880us       0.000us         0.00%       0.000us       0.000us            24  
+                                       aten::as_strided         0.35%      18.291us         0.35%      18.291us       0.762us       0.000us         0.00%       0.000us       0.000us            24  
+                                       aten::empty_like         0.38%      20.141us         1.58%      83.392us       5.559us       0.000us         0.00%       0.000us       0.000us            15  
+                                            aten::empty         1.49%      78.782us         1.49%      78.782us       3.283us       0.000us         0.00%       0.000us       0.000us            24  
+                                       cudaLaunchKernel         1.99%     104.800us         1.99%     104.800us       6.987us       0.000us         0.00%       0.000us       0.000us            15  
+                                    aten::empty_strided         0.27%      14.320us         0.27%      14.320us       4.773us       0.000us         0.00%       0.000us       0.000us             3  
+                                 cudaDeviceGetAttribute         0.04%       1.870us         0.04%       1.870us       0.312us       0.000us         0.00%       0.000us       0.000us             6  
+                                   cudaFuncSetAttribute         0.07%       3.720us         0.07%       3.720us       1.240us       0.000us         0.00%       0.000us       0.000us             3  
+                                  cudaDeviceSynchronize        58.31%       3.078ms        58.31%       3.078ms       3.078ms       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 5.345ms
-Self CUDA time total: 3.808ms
+Self CPU time total: 5.279ms
+Self CUDA time total: 3.736ms
 
 
 
@@ -4076,29 +4303,29 @@ PROFILE TRACE: torch_flash_ma | cuda_attn_L384_bfloat16
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                         torch_flash_ma         4.50%     255.237us        42.25%       2.398ms       2.398ms       0.000us         0.00%       3.984ms       3.984ms             1  
-                                         torch_flash_ma         0.00%       0.000us         0.00%       0.000us       0.000us       3.936ms       100.28%       3.936ms       3.936ms             1  
-                     aten::scaled_dot_product_attention         0.42%      23.840us         3.17%     179.904us      59.968us       0.000us         0.00%       3.135ms       1.045ms             3  
-              aten::_scaled_dot_product_flash_attention         0.36%      20.442us         2.75%     156.064us      52.021us       0.000us         0.00%       3.135ms       1.045ms             3  
-                         aten::_flash_attention_forward         0.68%      38.721us         1.99%     113.183us      37.728us       3.135ms        79.87%       3.135ms       1.045ms             3  
-void pytorch_flash::flash_fwd_kernel&lt;Flash_fwd_kerne...         0.00%       0.000us         0.00%       0.000us       0.000us       3.135ms        79.87%       3.135ms       1.045ms             3  
-                                       aten::contiguous         0.17%       9.382us        33.81%       1.919ms     159.915us       0.000us         0.00%     848.416us      70.701us            12  
-                                            aten::clone         0.52%      29.639us        33.64%       1.910ms     159.133us       0.000us         0.00%     848.416us      70.701us            12  
-                                            aten::copy_         1.40%      79.644us        32.03%       1.818ms     151.492us     790.048us        20.13%     848.416us      70.701us            12  
-void at::native::elementwise_kernel&lt;128, 4, at::nati...         0.00%       0.000us         0.00%       0.000us       0.000us     790.048us        20.13%     790.048us      65.837us            12  
-                                Activity Buffer Request        25.14%       1.427ms        25.14%       1.427ms       1.427ms      58.368us         1.49%      58.368us      58.368us             1  
-                                        aten::transpose         0.87%      49.289us         1.17%      66.169us       2.757us       0.000us         0.00%       0.000us       0.000us            24  
-                                       aten::as_strided         0.30%      16.880us         0.30%      16.880us       0.703us       0.000us         0.00%       0.000us       0.000us            24  
-                                       aten::empty_like         0.35%      19.852us         1.42%      80.662us       5.377us       0.000us         0.00%       0.000us       0.000us            15  
-                                            aten::empty         1.32%      74.981us         1.32%      74.981us       3.124us       0.000us         0.00%       0.000us       0.000us            24  
-                                       cudaLaunchKernel         5.89%     334.125us         5.89%     334.125us      22.275us       0.000us         0.00%       0.000us       0.000us            15  
-                                    aten::empty_strided         0.24%      13.720us         0.24%      13.720us       4.573us       0.000us         0.00%       0.000us       0.000us             3  
-                                 cudaDeviceGetAttribute         0.03%       1.760us         0.03%       1.760us       0.293us       0.000us         0.00%       0.000us       0.000us             6  
-                                   cudaFuncSetAttribute         0.06%       3.570us         0.06%       3.570us       1.190us       0.000us         0.00%       0.000us       0.000us             3  
-                                  cudaDeviceSynchronize        57.75%       3.278ms        57.75%       3.278ms       3.278ms       0.000us         0.00%       0.000us       0.000us             1  
+                                         torch_flash_ma         4.47%     246.252us        42.66%       2.352ms       2.352ms       0.000us         0.00%       3.878ms       3.878ms             1  
+                                         torch_flash_ma         0.00%       0.000us         0.00%       0.000us       0.000us       3.831ms       100.28%       3.831ms       3.831ms             1  
+                     aten::scaled_dot_product_attention         0.47%      26.180us         3.22%     177.714us      59.238us       0.000us         0.00%       3.035ms       1.012ms             3  
+              aten::_scaled_dot_product_flash_attention         0.34%      18.934us         2.75%     151.534us      50.511us       0.000us         0.00%       3.035ms       1.012ms             3  
+                         aten::_flash_attention_forward         0.60%      33.169us         1.99%     109.931us      36.644us       3.035ms        79.45%       3.035ms       1.012ms             3  
+void pytorch_flash::flash_fwd_kernel&lt;Flash_fwd_kerne...         0.00%       0.000us         0.00%       0.000us       0.000us       3.035ms        79.45%       3.035ms       1.012ms             3  
+                                       aten::contiguous         0.19%      10.269us        34.14%       1.882ms     156.829us       0.000us         0.00%     843.264us      70.272us            12  
+                                            aten::clone         0.51%      27.861us        33.95%       1.872ms     155.974us       0.000us         0.00%     843.264us      70.272us            12  
+                                            aten::copy_         1.39%      76.612us        32.27%       1.779ms     148.225us     785.216us        20.55%     843.264us      70.272us            12  
+void at::native::elementwise_kernel&lt;128, 4, at::nati...         0.00%       0.000us         0.00%       0.000us       0.000us     785.216us        20.55%     785.216us      65.435us            12  
+                                Activity Buffer Request        26.00%       1.433ms        26.00%       1.433ms       1.433ms      58.048us         1.52%      58.048us      58.048us             1  
+                                        aten::transpose         0.90%      49.620us         1.24%      68.282us       2.845us       0.000us         0.00%       0.000us       0.000us            24  
+                                       aten::as_strided         0.34%      18.662us         0.34%      18.662us       0.778us       0.000us         0.00%       0.000us       0.000us            24  
+                                       aten::empty_like         0.37%      20.139us         1.52%      83.911us       5.594us       0.000us         0.00%       0.000us       0.000us            15  
+                                            aten::empty         1.44%      79.524us         1.44%      79.524us       3.313us       0.000us         0.00%       0.000us       0.000us            24  
+                                       cudaLaunchKernel         5.29%     291.664us         5.29%     291.664us      19.444us       0.000us         0.00%       0.000us       0.000us            15  
+                                    aten::empty_strided         0.25%      13.850us         0.25%      13.850us       4.617us       0.000us         0.00%       0.000us       0.000us             3  
+                                 cudaDeviceGetAttribute         0.03%       1.810us         0.03%       1.810us       0.302us       0.000us         0.00%       0.000us       0.000us             6  
+                                   cudaFuncSetAttribute         0.07%       3.620us         0.07%       3.620us       1.207us       0.000us         0.00%       0.000us       0.000us             3  
+                                  cudaDeviceSynchronize        57.34%       3.161ms        57.34%       3.161ms       3.161ms       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 5.676ms
-Self CUDA time total: 3.925ms
+Self CPU time total: 5.512ms
+Self CUDA time total: 3.820ms
 
 
 
@@ -4108,29 +4335,29 @@ PROFILE TRACE: torch_flash_ma | cuda_attn_L448_bfloat16
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                         torch_flash_ma         5.07%     311.056us        40.82%       2.505ms       2.505ms       0.000us         0.00%       4.409ms       4.409ms             1  
-                                         torch_flash_ma         0.00%       0.000us         0.00%       0.000us       0.000us       4.359ms       100.26%       4.359ms       4.359ms             1  
-                     aten::scaled_dot_product_attention         0.41%      24.931us         3.07%     188.265us      62.755us       0.000us         0.00%       3.539ms       1.180ms             3  
-              aten::_scaled_dot_product_flash_attention         0.33%      20.199us         2.66%     163.334us      54.445us       0.000us         0.00%       3.539ms       1.180ms             3  
-                         aten::_flash_attention_forward         0.67%      41.371us         1.94%     118.823us      39.608us       3.539ms        81.38%       3.539ms       1.180ms             3  
-void pytorch_flash::flash_fwd_kernel&lt;Flash_fwd_kerne...         0.00%       0.000us         0.00%       0.000us       0.000us       3.539ms        81.38%       3.539ms       1.180ms             3  
-                                       aten::contiguous         0.16%       9.771us        31.97%       1.962ms     163.526us       0.000us         0.00%     870.819us      72.568us            12  
-                                            aten::clone         0.47%      28.779us        31.82%       1.953ms     162.712us       0.000us         0.00%     870.819us      72.568us            12  
-                                            aten::copy_         1.27%      77.896us        30.33%       1.862ms     155.132us     809.571us        18.62%     870.819us      72.568us            12  
-void at::native::elementwise_kernel&lt;128, 4, at::nati...         0.00%       0.000us         0.00%       0.000us       0.000us     809.571us        18.62%     809.571us      67.464us            12  
-                                Activity Buffer Request        24.14%       1.481ms        24.14%       1.481ms       1.481ms      61.248us         1.41%      61.248us      61.248us             1  
-                                        aten::transpose         0.82%      50.583us         1.11%      68.092us       2.837us       0.000us         0.00%       0.000us       0.000us            24  
-                                       aten::as_strided         0.29%      17.509us         0.29%      17.509us       0.730us       0.000us         0.00%       0.000us       0.000us            24  
-                                       aten::empty_like         0.32%      19.913us         1.33%      81.883us       5.459us       0.000us         0.00%       0.000us       0.000us            15  
-                                            aten::empty         1.23%      75.660us         1.23%      75.660us       3.153us       0.000us         0.00%       0.000us       0.000us            24  
-                                       cudaLaunchKernel         5.31%     325.825us         5.31%     325.825us      21.722us       0.000us         0.00%       0.000us       0.000us            15  
-                                    aten::empty_strided         0.24%      14.770us         0.24%      14.770us       4.923us       0.000us         0.00%       0.000us       0.000us             3  
+                                         torch_flash_ma         4.69%     283.303us        42.14%       2.547ms       2.547ms       0.000us         0.00%       4.304ms       4.304ms             1  
+                                         torch_flash_ma         0.00%       0.000us         0.00%       0.000us       0.000us       4.254ms       100.24%       4.254ms       4.254ms             1  
+                     aten::scaled_dot_product_attention         0.82%      49.722us         3.53%     213.285us      71.095us       0.000us         0.00%       3.439ms       1.146ms             3  
+              aten::_scaled_dot_product_flash_attention         0.34%      20.582us         2.71%     163.563us      54.521us       0.000us         0.00%       3.439ms       1.146ms             3  
+                         aten::_flash_attention_forward         0.62%      37.231us         1.93%     116.771us      38.924us       3.439ms        81.02%       3.439ms       1.146ms             3  
+void pytorch_flash::flash_fwd_kernel&lt;Flash_fwd_kerne...         0.00%       0.000us         0.00%       0.000us       0.000us       3.439ms        81.02%       3.439ms       1.146ms             3  
+                                       aten::contiguous         0.18%      10.912us        32.97%       1.993ms     166.068us       0.000us         0.00%     865.695us      72.141us            12  
+                                            aten::clone         0.50%      30.059us        32.79%       1.982ms     165.158us       0.000us         0.00%     865.695us      72.141us            12  
+                                            aten::copy_         1.39%      83.902us        31.17%       1.884ms     157.000us     805.439us        18.98%     865.695us      72.141us            12  
+void at::native::elementwise_kernel&lt;128, 4, at::nati...         0.00%       0.000us         0.00%       0.000us       0.000us     805.439us        18.98%     805.439us      67.120us            12  
+                                Activity Buffer Request        24.08%       1.456ms        24.08%       1.456ms       1.456ms      60.256us         1.42%      60.256us      60.256us             1  
+                                        aten::transpose         1.06%      63.793us         1.39%      84.162us       3.507us       0.000us         0.00%       0.000us       0.000us            24  
+                                       aten::as_strided         0.34%      20.369us         0.34%      20.369us       0.849us       0.000us         0.00%       0.000us       0.000us            24  
+                                       aten::empty_like         0.36%      21.791us         1.46%      88.331us       5.889us       0.000us         0.00%       0.000us       0.000us            15  
+                                            aten::empty         1.33%      80.570us         1.33%      80.570us       3.357us       0.000us         0.00%       0.000us       0.000us            24  
+                                       cudaLaunchKernel         6.09%     368.355us         6.09%     368.355us      24.557us       0.000us         0.00%       0.000us       0.000us            15  
+                                    aten::empty_strided         0.25%      15.000us         0.25%      15.000us       5.000us       0.000us         0.00%       0.000us       0.000us             3  
                                  cudaDeviceGetAttribute         0.03%       1.990us         0.03%       1.990us       0.332us       0.000us         0.00%       0.000us       0.000us             6  
-                                   cudaFuncSetAttribute         0.06%       3.670us         0.06%       3.670us       1.223us       0.000us         0.00%       0.000us       0.000us             3  
-                                  cudaDeviceSynchronize        59.18%       3.632ms        59.18%       3.632ms       3.632ms       0.000us         0.00%       0.000us       0.000us             1  
+                                   cudaFuncSetAttribute         0.07%       4.160us         0.07%       4.160us       1.387us       0.000us         0.00%       0.000us       0.000us             3  
+                                  cudaDeviceSynchronize        57.86%       3.497ms        57.86%       3.497ms       3.497ms       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 6.137ms
-Self CUDA time total: 4.348ms
+Self CPU time total: 6.045ms
+Self CUDA time total: 4.244ms
 
 
 
@@ -4140,39 +4367,45 @@ PROFILE TRACE: torch_flash_ma | cuda_attn_L512_bfloat16
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                    Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-                                         torch_flash_ma         4.13%     252.675us        38.98%       2.384ms       2.384ms       0.000us         0.00%       4.451ms       4.451ms             1  
-                                         torch_flash_ma         0.00%       0.000us         0.00%       0.000us       0.000us       4.400ms       100.24%       4.400ms       4.400ms             1  
-                     aten::scaled_dot_product_attention         0.50%      30.480us         3.11%     190.334us      63.445us       0.000us         0.00%       3.566ms       1.189ms             3  
-              aten::_scaled_dot_product_flash_attention         0.31%      19.082us         2.61%     159.854us      53.285us       0.000us         0.00%       3.566ms       1.189ms             3  
-                         aten::_flash_attention_forward         0.62%      38.112us         1.93%     118.053us      39.351us       3.566ms        81.24%       3.566ms       1.189ms             3  
-void pytorch_flash::flash_fwd_kernel&lt;Flash_fwd_kerne...         0.00%       0.000us         0.00%       0.000us       0.000us       3.566ms        81.24%       3.566ms       1.189ms             3  
-                                       aten::contiguous         0.16%       9.891us        31.02%       1.897ms     158.059us       0.000us         0.00%     884.831us      73.736us            12  
-                                            aten::clone         0.50%      30.290us        30.85%       1.887ms     157.234us       0.000us         0.00%     884.831us      73.736us            12  
-                                            aten::copy_         1.28%      78.520us        29.35%       1.795ms     149.550us     823.711us        18.76%     884.831us      73.736us            12  
-void at::native::elementwise_kernel&lt;128, 4, at::nati...         0.00%       0.000us         0.00%       0.000us       0.000us     823.711us        18.76%     823.711us      68.643us            12  
-                                Activity Buffer Request        23.29%       1.424ms        23.29%       1.424ms       1.424ms      61.120us         1.39%      61.120us      61.120us             1  
-                                        aten::transpose         0.81%      49.593us         1.09%      66.721us       2.780us       0.000us         0.00%       0.000us       0.000us            24  
-                                       aten::as_strided         0.28%      17.128us         0.28%      17.128us       0.714us       0.000us         0.00%       0.000us       0.000us            24  
-                                       aten::empty_like         0.33%      20.381us         1.35%      82.362us       5.491us       0.000us         0.00%       0.000us       0.000us            15  
-                                            aten::empty         1.23%      74.920us         1.23%      74.920us       3.122us       0.000us         0.00%       0.000us       0.000us            24  
-                                       cudaLaunchKernel         5.19%     317.558us         5.19%     317.558us      21.171us       0.000us         0.00%       0.000us       0.000us            15  
-                                    aten::empty_strided         0.25%      15.161us         0.25%      15.161us       5.054us       0.000us         0.00%       0.000us       0.000us             3  
-                                 cudaDeviceGetAttribute         0.03%       1.791us         0.03%       1.791us       0.299us       0.000us         0.00%       0.000us       0.000us             6  
-                                   cudaFuncSetAttribute         0.06%       3.670us         0.06%       3.670us       1.223us       0.000us         0.00%       0.000us       0.000us             3  
-                                  cudaDeviceSynchronize        61.02%       3.732ms        61.02%       3.732ms       3.732ms       0.000us         0.00%       0.000us       0.000us             1  
+                                         torch_flash_ma         4.04%     248.485us        39.71%       2.440ms       2.440ms       0.000us         0.00%       4.431ms       4.431ms             1  
+                                         torch_flash_ma         0.00%       0.000us         0.00%       0.000us       0.000us       4.380ms       100.24%       4.380ms       4.380ms             1  
+                     aten::scaled_dot_product_attention         0.42%      25.679us         2.90%     178.082us      59.361us       0.000us         0.00%       3.552ms       1.184ms             3  
+              aten::_scaled_dot_product_flash_attention         0.29%      17.912us         2.48%     152.403us      50.801us       0.000us         0.00%       3.552ms       1.184ms             3  
+                         aten::_flash_attention_forward         0.56%      34.360us         1.81%     111.452us      37.151us       3.552ms        81.28%       3.552ms       1.184ms             3  
+void pytorch_flash::flash_fwd_kernel&lt;Flash_fwd_kerne...         0.00%       0.000us         0.00%       0.000us       0.000us       3.552ms        81.28%       3.552ms       1.184ms             3  
+                                       aten::contiguous         0.17%      10.359us        32.01%       1.967ms     163.915us       0.000us         0.00%     879.392us      73.283us            12  
+                                            aten::clone         0.45%      27.371us        31.84%       1.957ms     163.052us       0.000us         0.00%     879.392us      73.283us            12  
+                                            aten::copy_         1.33%      81.681us        30.34%       1.864ms     155.367us     818.048us        18.72%     879.392us      73.283us            12  
+void at::native::elementwise_kernel&lt;128, 4, at::nati...         0.00%       0.000us         0.00%       0.000us       0.000us     818.048us        18.72%     818.048us      68.171us            12  
+                                Activity Buffer Request        23.48%       1.443ms        23.48%       1.443ms       1.443ms      61.344us         1.40%      61.344us      61.344us             1  
+                                        aten::transpose         0.84%      51.433us         1.14%      69.901us       2.913us       0.000us         0.00%       0.000us       0.000us            24  
+                                       aten::as_strided         0.30%      18.468us         0.30%      18.468us       0.769us       0.000us         0.00%       0.000us       0.000us            24  
+                                       aten::empty_like         0.32%      19.754us         1.37%      83.993us       5.600us       0.000us         0.00%       0.000us       0.000us            15  
+                                            aten::empty         1.26%      77.740us         1.26%      77.740us       3.239us       0.000us         0.00%       0.000us       0.000us            24  
+                                       cudaLaunchKernel         5.92%     364.005us         5.92%     364.005us      24.267us       0.000us         0.00%       0.000us       0.000us            15  
+                                    aten::empty_strided         0.23%      14.381us         0.23%      14.381us       4.794us       0.000us         0.00%       0.000us       0.000us             3  
+                                 cudaDeviceGetAttribute         0.03%       1.840us         0.03%       1.840us       0.307us       0.000us         0.00%       0.000us       0.000us             6  
+                                   cudaFuncSetAttribute         0.07%       4.180us         0.07%       4.180us       1.393us       0.000us         0.00%       0.000us       0.000us             3  
+                                  cudaDeviceSynchronize        60.29%       3.705ms        60.29%       3.705ms       3.705ms       0.000us         0.00%       0.000us       0.000us             1  
 -------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
-Self CPU time total: 6.115ms
-Self CUDA time total: 4.390ms
+Self CPU time total: 6.146ms
+Self CUDA time total: 4.370ms
 
 
 impl                     wl                  p50(ms)  ok
-torch_flash_ma           cuda_attn_L128_bfloat16     1.21  True
+torch_flash_ma           cuda_attn_L128_bfloat16     1.22  True
 torch_flash_ma           cuda_attn_L256_bfloat16     1.27  True
-torch_flash_ma           cuda_attn_L320_bfloat16     1.30  True
-torch_flash_ma           cuda_attn_L384_bfloat16     1.32  True
-torch_flash_ma           cuda_attn_L448_bfloat16     1.48  True
+torch_flash_ma           cuda_attn_L320_bfloat16     1.29  True
+torch_flash_ma           cuda_attn_L384_bfloat16     1.30  True
+torch_flash_ma           cuda_attn_L448_bfloat16     1.45  True
 torch_flash_ma           cuda_attn_L512_bfloat16     1.49  True
 </pre></div>
+<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 37 packages in 225ms
+</div>
+</div>
 <div class="cell-artifacts">
 <h4>Artifacts:</h4>
 <a href="artifacts/benchmark/attention.jsonl" class="artifact" target="_blank">attention.jsonl</a>