vibe coding gguf

Vibe coding design baby
Some Linux Laptop/Desktop detection and goodbye penguin
2026-03-10 18:32:48 +00:00 · 2026-03-10 17:21:28 +00:00 · 2026-03-10 17:03:56 +00:00 · 2026-03-10 16:55:28 +00:00 · 2026-03-10 16:08:03 +00:00 · 2026-03-10 16:08:03 +00:00
29 changed files with 1890 additions and 42 deletions
@@ -0,0 +1,30 @@
+#!/bin/bash
+set -e
+
+echo "=== Starting overnight bench runs at $(date) ==="
+
+echo "--- [1/8] Qwen3.5-27B-GPTQ-Int4 ---"
+uv run bench/exo_bench.py --model "mlx-community/Qwen3.5-27B-GPTQ-Int4" --pp 700 --tg 32067,33085 --repeat 1
+
+echo "--- [2/8] NVIDIA-Nemotron-3-Nano-30B-A3B (1120,1330,23100) ---"
+uv run bench/exo_bench.py --model "mlx-community/NVIDIA-Nemotron-3-Nano-30B-A3B-MLX-BF16" --pp 700 --tg 1120,1330,23100 --repeat 1
+
+echo "--- [3/8] Qwen3.5-35B-A3B-bf16 ---"
+uv run bench/exo_bench.py --model "mlx-community/Qwen3.5-35B-A3B-bf16" --pp 700 --tg 6200,6450,25600,38000 --repeat 1
+
+echo "--- [4/8] Qwen3.5-122B-A10B-GPTQ-Int4 ---"
+uv run bench/exo_bench.py --model "Qwen/Qwen3.5-122B-A10B-GPTQ-Int4" --pp 700 --tg 36000 --repeat 1
+
+echo "--- [5/8] Qwen3.5-27B-FP8 ---"
+uv run bench/exo_bench.py --model "Qwen/Qwen3.5-27B-FP8" --pp 700 --tg 33736,35133 --repeat 1
+
+echo "--- [6/8] GLM-4.7-Flash-bf16 ---"
+uv run bench/exo_bench.py --model "mlx-community/GLM-4.7-Flash-bf16" --pp 700 --tg 29000 --repeat 1
+
+echo "--- [7/8] NVIDIA-Nemotron-3-Nano-30B-A3B (23000,1200) ---"
+uv run bench/exo_bench.py --model "mlx-community/NVIDIA-Nemotron-3-Nano-30B-A3B-MLX-BF16" --pp 700 --tg 23000,1200 --repeat 1
+
+echo "--- [8/8] Qwen3.5-27B-bf16 ---"
+uv run bench/exo_bench.py --model "mlx-community/Qwen3.5-27B-bf16" --pp 700 --tg 32000,35400 --repeat 1
+
+echo "=== All bench runs complete at $(date) ==="
@@ -9,7 +9,7 @@
   */

  interface Props {
-    /** "macbook pro" | "mac studio" | "mac mini" etc. */
+    /** "macbook pro" | "mac studio" | "mac mini" | "dgx spark" | "linux" etc. */
    deviceType: string;
    /** Center X coordinate in SVG space */
    cx: number;
@@ -38,10 +38,43 @@
  const LOGO_NATIVE_WIDTH = 814;
  const LOGO_NATIVE_HEIGHT = 1000;

+  // NVIDIA logo SVG path
+  const NVIDIA_LOGO_PATH =
+    "M0.81 0.429V0.299c0.013 -0.001 0.026 -0.002 0.038 -0.002 0.355 -0.011 0.588 0.306 0.588 0.306S1.186 0.952 0.916 0.952c-0.036 0 -0.071 -0.006 -0.105 -0.017V0.542c0.138 0.017 0.166 0.078 0.249 0.216l0.185 -0.155s-0.135 -0.177 -0.362 -0.177c-0.024 -0.001 -0.048 0.001 -0.072 0.003m0 -0.429v0.194l0.038 -0.002c0.494 -0.017 0.816 0.405 0.816 0.405s-0.37 0.45 -0.754 0.45c-0.034 0 -0.066 -0.003 -0.099 -0.009v0.12c0.027 0.003 0.055 0.006 0.082 0.006 0.358 0 0.618 -0.183 0.869 -0.399 0.042 0.034 0.212 0.114 0.247 0.15 -0.238 0.2 -0.794 0.361 -1.11 0.361 -0.03 0 -0.059 -0.002 -0.088 -0.005v0.169h1.362V0zm0 0.935v0.102c-0.331 -0.059 -0.423 -0.404 -0.423 -0.404s0.159 -0.176 0.423 -0.205v0.112h-0.001C0.671 0.524 0.562 0.654 0.562 0.654s0.062 0.218 0.248 0.282m-0.588 -0.316s0.196 -0.29 0.589 -0.32V0.194C0.376 0.229 0 0.597 0 0.597s0.213 0.616 0.81 0.672v-0.112c-0.438 -0.054 -0.588 -0.538 -0.588 -0.538";
+
  const wireColor = "rgba(179,179,179,0.8)";
  const strokeWidth = 1.5;

  const modelLower = $derived(deviceType.toLowerCase());
+  const isSpark = $derived(
+    modelLower.includes("dgx") || modelLower.includes("gx10"),
+  );
+  const isLinux = $derived(!isSpark && modelLower.startsWith("linux"));
+  const isLinuxLaptop = $derived(isLinux && modelLower.includes("laptop"));
+
+  // ── DGX Spark dimensions ──
+  const dgxW = $derived(size * 1.55);
+  const dgxH = $derived(size * 0.58);
+  const dgxX = $derived(cx - dgxW / 2);
+  const dgxY = $derived(cy - dgxH / 2);
+  const dgxChassisX = $derived(dgxX - dgxW * 0.03);
+  const dgxChassisW = $derived(dgxW * 1.05);
+  const dgxHandleW = $derived(dgxW * 0.27);
+  const dgxHandleGap = $derived(dgxH * 0.05);
+  const dgxHandleH = $derived(dgxH - dgxHandleGap * 2);
+  const dgxHandleY = $derived(dgxY + dgxHandleGap);
+  const dgxInnerHandleW = $derived(dgxW * 0.12);
+  const dgxInnerHandleH = $derived(dgxHandleH - dgxH * 0.06);
+  const dgxLeftHandleX = $derived(dgxX + 4);
+  const dgxRightHandleX = $derived(dgxX + dgxW - dgxHandleW - 4);
+  const dgxClipId = $derived(`di-dgx-${uid}`);
+  const dgxTextureId = $derived(`di-dgx-tex-${uid}`);
+
+  // ── Linux Desktop dimensions (reuses Mac Studio proportions) ──
+  const linuxDesktopClipId = $derived(`di-linux-desktop-${uid}`);
+
+  // ── Linux Laptop dimensions (reuses MacBook proportions) ──
+  const linuxScreenClipId = $derived(`di-linux-screen-${uid}`);

  // ── Mac Studio dimensions (same ratios as TopologyGraph) ──
  const studioW = $derived(size * 1.25);
@@ -114,7 +147,264 @@
  const studioClipId = $derived(`di-studio-${uid}`);
 </script>

-{#if modelLower === "mac studio" || modelLower === "mac mini"}
+{#if isSpark}
+  <!-- DGX Spark -->
+  <defs>
+    <clipPath id={dgxClipId}>
+      <rect x={dgxX} y={dgxY} width={dgxW} height={dgxH} rx="3" />
+    </clipPath>
+    <pattern
+      id={dgxTextureId}
+      patternUnits="userSpaceOnUse"
+      width="8"
+      height="8"
+    >
+      <rect width="8" height="8" fill="#6f6248" />
+      <circle cx="2" cy="2" r="1" fill="#5a4f3b" opacity="0.5" />
+      <circle cx="6" cy="6" r="1" fill="#4a4232" opacity="0.45" />
+    </pattern>
+  </defs>
+
+  <!-- Main body -->
+  <rect
+    x={dgxChassisX}
+    y={dgxY}
+    width={dgxChassisW}
+    height={dgxH}
+    rx="3"
+    fill="url(#{dgxTextureId})"
+    stroke={wireColor}
+    stroke-width={strokeWidth}
+  />
+
+  <!-- Side border accents -->
+  <rect
+    x={dgxChassisX}
+    y={dgxY}
+    width={dgxW * 0.02}
+    height={dgxH}
+    fill="#8a7a56"
+  />
+  <rect
+    x={dgxChassisX + dgxChassisW - dgxW * 0.02}
+    y={dgxY}
+    width={dgxW * 0.02}
+    height={dgxH}
+    fill="#8a7a56"
+  />
+
+  <!-- Memory fill -->
+  {#if ramPercent > 0}
+    <rect
+      x={dgxX}
+      y={dgxY + dgxH - (ramPercent / 100) * dgxH}
+      width={dgxW}
+      height={(ramPercent / 100) * dgxH}
+      fill="rgba(255,215,0,0.45)"
+      clip-path="url(#{dgxClipId})"
+    />
+  {/if}
+
+  <!-- Left handle -->
+  <rect
+    x={dgxLeftHandleX}
+    y={dgxHandleY}
+    width={dgxHandleW}
+    height={dgxHandleH}
+    rx="2.4"
+    fill="#b3a170"
+    stroke="#403723"
+    stroke-width="0.7"
+  />
+  <rect
+    x={dgxLeftHandleX + dgxHandleW * 0.06}
+    y={dgxHandleY + dgxH * 0.03}
+    width={dgxInnerHandleW}
+    height={dgxInnerHandleH}
+    rx="1.6"
+    fill="#8a7a56"
+  />
+
+  <!-- Right handle -->
+  <rect
+    x={dgxRightHandleX}
+    y={dgxHandleY}
+    width={dgxHandleW}
+    height={dgxHandleH}
+    rx="2.4"
+    fill="#b3a170"
+    stroke="#403723"
+    stroke-width="0.7"
+  />
+  <rect
+    x={dgxRightHandleX + dgxHandleW - dgxInnerHandleW - dgxHandleW * 0.08}
+    y={dgxHandleY + dgxH * 0.03}
+    width={dgxInnerHandleW}
+    height={dgxInnerHandleH}
+    rx="1.6"
+    fill="#8a7a56"
+  />
+
+  <!-- NVIDIA logo (rotated 90deg on left handle) -->
+  {@const badgeW = dgxW * 0.09}
+  {@const badgeH = dgxHandleH * 0.5}
+  {@const badgeX = dgxLeftHandleX + dgxHandleW - badgeW - dgxHandleW * 0.06}
+  {@const badgeYPos = dgxHandleY + (dgxHandleH - badgeH) / 2}
+  {@const textSz = badgeW * 0.58}
+  {@const logoW = textSz * 1.2}
+  {@const logoH = logoW * (1.438 / 2.174)}
+  {@const ctrX = badgeX + badgeW / 2 - badgeW * 0.03}
+  {@const ctrY = badgeYPos + badgeH / 2}
+  {@const labelGap = badgeW * 0.15}
+  {@const totalW = logoW + labelGap + textSz * 3.6}
+  <g transform="rotate(90 {ctrX} {ctrY})">
+    <svg
+      x={ctrX - totalW / 2}
+      y={ctrY - logoH / 2}
+      width={logoW}
+      height={logoH}
+      viewBox="0 0 2.174 1.438"
+    >
+      <path d={NVIDIA_LOGO_PATH} fill="#76b900" />
+    </svg>
+    <text
+      x={ctrX - totalW / 2 + logoW + labelGap}
+      y={ctrY}
+      text-anchor="start"
+      dominant-baseline="middle"
+      fill="#8a7a56"
+      font-size={textSz}
+      font-family="monospace"
+      font-weight="700">NVIDIA</text
+    >
+  </g>
+{:else if isLinuxLaptop}
+  <!-- Linux Laptop — MacBook shape with Tux logo -->
+  <defs>
+    <clipPath id={linuxScreenClipId}>
+      <rect
+        x={mbScreenX + mbBezel}
+        y={mbY + mbBezel}
+        width={mbScreenW - mbBezel * 2}
+        height={mbScreenH - mbBezel * 2}
+        rx="2"
+      />
+    </clipPath>
+  </defs>
+
+  <rect
+    x={mbScreenX}
+    y={mbY}
+    width={mbScreenW}
+    height={mbScreenH}
+    rx="3"
+    fill="#1a1a1a"
+    stroke={wireColor}
+    stroke-width={strokeWidth}
+  />
+  <rect
+    x={mbScreenX + mbBezel}
+    y={mbY + mbBezel}
+    width={mbScreenW - mbBezel * 2}
+    height={mbScreenH - mbBezel * 2}
+    rx="2"
+    fill="#0a0a12"
+  />
+  {#if ramPercent > 0}
+    <rect
+      x={mbScreenX + mbBezel}
+      y={mbY + mbBezel + (mbMemTotalH - mbMemH)}
+      width={mbScreenW - mbBezel * 2}
+      height={mbMemH}
+      fill="rgba(255,215,0,0.85)"
+      clip-path="url(#{linuxScreenClipId})"
+    />
+  {/if}
+
+  <!-- Terminal prompt on screen -->
+  <text
+    x={cx}
+    y={mbY + mbScreenH / 2}
+    text-anchor="middle"
+    dominant-baseline="middle"
+    fill="#FFFFFF"
+    opacity="0.9"
+    font-size={mbScreenH * 0.25}
+    font-family="SF Mono, Monaco, monospace"
+    font-weight="700">{">_"}</text
+  >
+
+  <path
+    d="M {mbBaseTopX} {mbBaseY} L {mbBaseTopX +
+      mbBaseTopW} {mbBaseY} L {mbBaseBottomX + mbBaseBottomW} {mbBaseY +
+      mbBaseH} L {mbBaseBottomX} {mbBaseY + mbBaseH} Z"
+    fill="#2c2c2c"
+    stroke={wireColor}
+    stroke-width="1"
+  />
+  <rect
+    x={mbKbX}
+    y={mbKbY}
+    width={mbKbW}
+    height={mbKbH}
+    fill="rgba(0,0,0,0.2)"
+    rx="2"
+  />
+  <rect
+    x={mbTpX}
+    y={mbTpY}
+    width={mbTpW}
+    height={mbTpH}
+    fill="rgba(255,255,255,0.08)"
+    rx="2"
+  />
+{:else if isLinux}
+  <!-- Linux Desktop — Mac Studio shape with Tux logo -->
+  <defs>
+    <clipPath id={linuxDesktopClipId}>
+      <rect
+        x={studioX}
+        y={studioY + studioTopH}
+        width={studioW}
+        height={studioH - studioTopH}
+        rx={studioCorner - 1}
+      />
+    </clipPath>
+  </defs>
+
+  <rect
+    x={studioX}
+    y={studioY}
+    width={studioW}
+    height={studioH}
+    rx={studioCorner}
+    fill="#1a1a1a"
+    stroke={wireColor}
+    stroke-width={strokeWidth}
+  />
+  {#if ramPercent > 0}
+    <rect
+      x={studioX}
+      y={studioY + studioTopH + (studioMemTotalH - studioMemH)}
+      width={studioW}
+      height={studioMemH}
+      fill="rgba(255,215,0,0.75)"
+      clip-path="url(#{linuxDesktopClipId})"
+    />
+  {/if}
+
+  <!-- Terminal prompt on front face -->
+  <text
+    x={cx}
+    y={studioY + studioTopH + (studioH - studioTopH) / 2}
+    text-anchor="middle"
+    dominant-baseline="middle"
+    fill="rgba(255,255,255,0.5)"
+    font-size={(studioH - studioTopH) * 0.4}
+    font-family="SF Mono, Monaco, monospace"
+    font-weight="700">{">_"}</text
+  >
+{:else if modelLower === "mac studio" || modelLower === "mac mini"}
  <!-- Mac Studio / Mac Mini -->
  <defs>
    <clipPath id={studioClipId}>
@@ -169,8 +169,10 @@

  function getDeviceType(
    name: string,
-  ): "macbook" | "studio" | "mini" | "unknown" {
+  ): "macbook" | "studio" | "mini" | "dgx" | "linux" | "unknown" {
    const lower = name.toLowerCase();
+    if (lower.includes("dgx") || lower.includes("gx10")) return "dgx";
+    if (lower.includes("linux")) return "linux";
    if (lower.includes("macbook")) return "macbook";
    if (lower.includes("studio")) return "studio";
    if (lower.includes("mini")) return "mini";
@@ -278,7 +280,7 @@
    let placementNodes: Array<{
      id: string;
      deviceName: string;
-      deviceType: "macbook" | "studio" | "mini" | "unknown";
+      deviceType: "macbook" | "studio" | "mini" | "dgx" | "linux" | "unknown";
      totalGB: number;
      currentUsedGB: number;
      modelUsageGB: number;
@@ -968,6 +970,137 @@
                    />
                  {/if}
                </g>
+              {:else if node.deviceType === "dgx"}
+                <!-- DGX Spark icon -->
+                {@const s = node.iconSize}
+                {@const dgxW = s * 1.4}
+                {@const dgxH = s * 0.52}
+                <g transform="translate({-dgxW / 2}, {-dgxH / 2})">
+                  <!-- Chassis -->
+                  <rect
+                    x="0"
+                    y="0"
+                    width={dgxW}
+                    height={dgxH}
+                    rx="2"
+                    fill="#6f6248"
+                    stroke={node.isUsed ? "#FFD700" : "#4B5563"}
+                    stroke-width="1.5"
+                  />
+                  <!-- Side accents -->
+                  <rect
+                    x="0"
+                    y="0"
+                    width={dgxW * 0.02}
+                    height={dgxH}
+                    fill="#8a7a56"
+                  />
+                  <rect
+                    x={dgxW - dgxW * 0.02}
+                    y="0"
+                    width={dgxW * 0.02}
+                    height={dgxH}
+                    fill="#8a7a56"
+                  />
+                  <!-- Left handle -->
+                  <rect
+                    x={dgxW * 0.04}
+                    y={dgxH * 0.08}
+                    width={dgxW * 0.22}
+                    height={dgxH * 0.84}
+                    rx="2"
+                    fill="#b3a170"
+                    stroke="#403723"
+                    stroke-width="0.5"
+                  />
+                  <!-- Right handle -->
+                  <rect
+                    x={dgxW - dgxW * 0.04 - dgxW * 0.22}
+                    y={dgxH * 0.08}
+                    width={dgxW * 0.22}
+                    height={dgxH * 0.84}
+                    rx="2"
+                    fill="#b3a170"
+                    stroke="#403723"
+                    stroke-width="0.5"
+                  />
+                  <!-- Memory fill -->
+                  <rect
+                    x="2"
+                    y={dgxH - dgxH * (node.currentPercent / 100)}
+                    width={dgxW - 4}
+                    height={dgxH * (node.currentPercent / 100)}
+                    fill="rgba(255,215,0,0.35)"
+                  />
+                  {#if node.modelUsageGB > 0 && node.isUsed}
+                    <rect
+                      x="2"
+                      y={dgxH - dgxH * (node.newPercent / 100)}
+                      width={dgxW - 4}
+                      height={dgxH *
+                        ((node.newPercent - node.currentPercent) / 100)}
+                      fill="#FFD700"
+                      filter="url(#memGlow-{filterId})"
+                      class="animate-pulse-slow"
+                    />
+                  {/if}
+                </g>
+              {:else if node.deviceType === "linux"}
+                <!-- Linux Tux penguin icon -->
+                {@const sz = node.iconSize}
+                {@const sc = sz / 100}
+                <g transform="translate({-sz / 2}, {-sz / 2})">
+                  <!-- Body -->
+                  <path
+                    d="M50 8c-8 0-14 6-14 13 0 4 2 8 5 10-8 4-16 14-16 28v12c0 4 2 7 5 9l-6 4c-2 1-3 3-3 5v3c0 2 2 4 4 4h10l4-4h22l4 4h10c2 0 4-2 4-4v-3c0-2-1-4-3-5l-6-4c3-2 5-5 5-9V59c0-14-8-24-16-28 3-2 5-6 5-10 0-7-6-13-14-13z"
+                    transform="scale({sc})"
+                    fill="#1a1a1a"
+                    stroke={node.isUsed ? "#FFD700" : "#4B5563"}
+                    stroke-width={1.5 / sc}
+                  />
+                  <!-- Belly -->
+                  <path
+                    d="M38 52c0-8 5-15 12-15s12 7 12 15v14c0 4-5 7-12 7s-12-3-12-7V52z"
+                    transform="scale({sc})"
+                    fill="rgba(220,220,220,0.85)"
+                  />
+                  <!-- Eyes -->
+                  <circle cx={44 * sc} cy={16 * sc} r={2.5 * sc} fill="white" />
+                  <circle cx={56 * sc} cy={16 * sc} r={2.5 * sc} fill="white" />
+                  <circle
+                    cx={44 * sc}
+                    cy={16 * sc}
+                    r={1.2 * sc}
+                    fill="#1a1a1a"
+                  />
+                  <circle
+                    cx={56 * sc}
+                    cy={16 * sc}
+                    r={1.2 * sc}
+                    fill="#1a1a1a"
+                  />
+                  <!-- Beak -->
+                  <path
+                    d="M{46 * sc} {22 * sc} L{50 * sc} {27 * sc} L{54 *
+                      sc} {22 * sc} Z"
+                    fill="#E8A317"
+                  />
+                  <!-- Feet -->
+                  <ellipse
+                    cx={42 * sc}
+                    cy={94 * sc}
+                    rx={6 * sc}
+                    ry={2.5 * sc}
+                    fill="#E8A317"
+                  />
+                  <ellipse
+                    cx={58 * sc}
+                    cy={94 * sc}
+                    rx={6 * sc}
+                    ry={2.5 * sc}
+                    fill="#E8A317"
+                  />
+                </g>
              {:else}
                <!-- Unknown device - hexagon -->
                <g
@@ -117,6 +117,10 @@
  const LOGO_NATIVE_WIDTH = 814;
  const LOGO_NATIVE_HEIGHT = 1000;

+  // NVIDIA logo SVG path (from exo-nvidia)
+  const NVIDIA_LOGO_PATH =
+    "M0.81 0.429V0.299c0.013 -0.001 0.026 -0.002 0.038 -0.002 0.355 -0.011 0.588 0.306 0.588 0.306S1.186 0.952 0.916 0.952c-0.036 0 -0.071 -0.006 -0.105 -0.017V0.542c0.138 0.017 0.166 0.078 0.249 0.216l0.185 -0.155s-0.135 -0.177 -0.362 -0.177c-0.024 -0.001 -0.048 0.001 -0.072 0.003m0 -0.429v0.194l0.038 -0.002c0.494 -0.017 0.816 0.405 0.816 0.405s-0.37 0.45 -0.754 0.45c-0.034 0 -0.066 -0.003 -0.099 -0.009v0.12c0.027 0.003 0.055 0.006 0.082 0.006 0.358 0 0.618 -0.183 0.869 -0.399 0.042 0.034 0.212 0.114 0.247 0.15 -0.238 0.2 -0.794 0.361 -1.11 0.361 -0.03 0 -0.059 -0.002 -0.088 -0.005v0.169h1.362V0zm0 0.935v0.102c-0.331 -0.059 -0.423 -0.404 -0.423 -0.404s0.159 -0.176 0.423 -0.205v0.112h-0.001C0.671 0.524 0.562 0.654 0.562 0.654s0.062 0.218 0.248 0.282m-0.588 -0.316s0.196 -0.29 0.589 -0.32V0.194C0.376 0.229 0 0.597 0 0.597s0.213 0.616 0.81 0.672v-0.112c-0.438 -0.054 -0.588 -0.538 -0.588 -0.538";
+
  function formatBytes(bytes: number, decimals = 1): string {
    if (!bytes || bytes === 0) return "0B";
    const k = 1024;
@@ -554,6 +558,13 @@
      const clipPathId = `clip-${nodeInfo.id.replace(/[^a-zA-Z0-9]/g, "-")}`;

      const modelLower = modelId.toLowerCase();
+      const identity = identitiesData[nodeInfo.id];
+      const nameLower = (friendlyName || "").toLowerCase();
+      const isSpark = modelLower.includes("dgx") || modelLower.includes("gx10");
+      const isLinux =
+        !isSpark &&
+        (modelLower.startsWith("linux") || identity?.osVersion === "Linux");
+      const isLinuxLaptop = isLinux && modelLower.includes("laptop");

      // Check node states for styling
      const isHighlighted = highlightedNodes.has(nodeInfo.id);
@@ -623,7 +634,382 @@
          `${friendlyName}\nID: ${nodeInfo.id.slice(-8)}\nMemory: ${formatBytes(ramUsed)}/${formatBytes(ramTotal)}`,
        );

-      if (modelLower === "mac studio") {
+      if (isSpark) {
+        // NVIDIA DGX Spark — gold chassis with textured front, side handles, and NVIDIA badge
+        iconBaseWidth = nodeRadius * 1.55;
+        iconBaseHeight = nodeRadius * 0.58;
+        const x = nodeInfo.x - iconBaseWidth / 2;
+        const y = nodeInfo.y - iconBaseHeight / 2;
+        const chassisX = x - iconBaseWidth * 0.03;
+        const chassisWidth = iconBaseWidth * 1.05;
+        const cornerRadius = 3;
+
+        const dgxClipId = `dgx-clip-${nodeInfo.id.replace(/[^a-zA-Z0-9]/g, "-")}`;
+        defs
+          .append("clipPath")
+          .attr("id", dgxClipId)
+          .append("rect")
+          .attr("x", x)
+          .attr("y", y)
+          .attr("width", iconBaseWidth)
+          .attr("height", iconBaseHeight)
+          .attr("rx", cornerRadius);
+
+        // Chassis texture pattern
+        const textureId = `chassis-texture-${nodeInfo.id.replace(/[^a-zA-Z0-9]/g, "-")}`;
+        defs
+          .append("pattern")
+          .attr("id", textureId)
+          .attr("patternUnits", "userSpaceOnUse")
+          .attr("width", 8)
+          .attr("height", 8);
+        const texturePattern = defs.select(`#${textureId}`);
+        texturePattern
+          .append("rect")
+          .attr("width", 8)
+          .attr("height", 8)
+          .attr("fill", "#6f6248");
+        texturePattern
+          .append("circle")
+          .attr("cx", 2)
+          .attr("cy", 2)
+          .attr("r", 1)
+          .attr("fill", "#5a4f3b")
+          .attr("opacity", 0.5);
+        texturePattern
+          .append("circle")
+          .attr("cx", 6)
+          .attr("cy", 6)
+          .attr("r", 1)
+          .attr("fill", "#4a4232")
+          .attr("opacity", 0.45);
+
+        // Main body
+        nodeG
+          .append("rect")
+          .attr("class", "node-outline")
+          .attr("x", chassisX)
+          .attr("y", y)
+          .attr("width", chassisWidth)
+          .attr("height", iconBaseHeight)
+          .attr("rx", cornerRadius)
+          .attr("fill", `url(#${textureId})`)
+          .attr("stroke", wireColor)
+          .attr("stroke-width", strokeWidth);
+
+        // Side border accents
+        const sideThickness = iconBaseWidth * 0.02;
+        nodeG
+          .append("rect")
+          .attr("x", chassisX)
+          .attr("y", y)
+          .attr("width", sideThickness)
+          .attr("height", iconBaseHeight)
+          .attr("fill", "#8a7a56");
+        nodeG
+          .append("rect")
+          .attr("x", chassisX + chassisWidth - sideThickness)
+          .attr("y", y)
+          .attr("width", sideThickness)
+          .attr("height", iconBaseHeight)
+          .attr("fill", "#8a7a56");
+
+        // Memory fill (bottom up)
+        if (ramUsagePercent > 0) {
+          const memFillHeight = (ramUsagePercent / 100) * iconBaseHeight;
+          nodeG
+            .append("rect")
+            .attr("x", x)
+            .attr("y", y + iconBaseHeight - memFillHeight)
+            .attr("width", iconBaseWidth)
+            .attr("height", memFillHeight)
+            .attr("fill", "rgba(255,215,0,0.45)")
+            .attr("clip-path", `url(#${dgxClipId})`);
+        }
+
+        // Side handles with inner recess
+        const handleWidth = iconBaseWidth * 0.27;
+        const handleGap = iconBaseHeight * 0.05;
+        const handleHeight = iconBaseHeight - handleGap * 2;
+        const handleY = y + handleGap;
+        const innerHandleWidth = iconBaseWidth * 0.12;
+        const innerHandleHeight = handleHeight - iconBaseHeight * 0.06;
+        const leftHandleX = x + 4;
+        const rightHandleX = x + iconBaseWidth - handleWidth - 4;
+
+        // Left handle
+        nodeG
+          .append("rect")
+          .attr("x", leftHandleX)
+          .attr("y", handleY)
+          .attr("width", handleWidth)
+          .attr("height", handleHeight)
+          .attr("rx", 2.4)
+          .attr("fill", "#b3a170")
+          .attr("stroke", "#403723")
+          .attr("stroke-width", 0.7);
+        nodeG
+          .append("rect")
+          .attr("x", leftHandleX + handleWidth * 0.06)
+          .attr("y", handleY + iconBaseHeight * 0.03)
+          .attr("width", innerHandleWidth)
+          .attr("height", innerHandleHeight)
+          .attr("rx", 1.6)
+          .attr("fill", "#8a7a56");
+
+        // Right handle
+        nodeG
+          .append("rect")
+          .attr("x", rightHandleX)
+          .attr("y", handleY)
+          .attr("width", handleWidth)
+          .attr("height", handleHeight)
+          .attr("rx", 2.4)
+          .attr("fill", "#b3a170")
+          .attr("stroke", "#403723")
+          .attr("stroke-width", 0.7);
+        nodeG
+          .append("rect")
+          .attr(
+            "x",
+            rightHandleX + handleWidth - innerHandleWidth - handleWidth * 0.08,
+          )
+          .attr("y", handleY + iconBaseHeight * 0.03)
+          .attr("width", innerHandleWidth)
+          .attr("height", innerHandleHeight)
+          .attr("rx", 1.6)
+          .attr("fill", "#8a7a56");
+
+        // NVIDIA logo + text label (rotated 90 deg on left handle)
+        const badgeWidth = iconBaseWidth * 0.09;
+        const badgeHeight = handleHeight * 0.5;
+        const badgeX =
+          leftHandleX + handleWidth - badgeWidth - handleWidth * 0.06;
+        const badgeY = handleY + (handleHeight - badgeHeight) / 2;
+        const textSize = badgeWidth * 0.58;
+        const logoWidth = textSize * 1.2;
+        const logoHeight = logoWidth * (1.438 / 2.174);
+        const centerX = badgeX + badgeWidth / 2 - badgeWidth * 0.03;
+        const centerY = badgeY + badgeHeight / 2;
+        const gap = badgeWidth * 0.15;
+        const totalWidth = logoWidth + gap + textSize * 3.6;
+
+        const labelGroup = nodeG
+          .append("g")
+          .attr("transform", `rotate(90 ${centerX} ${centerY})`);
+
+        labelGroup
+          .append("svg")
+          .attr("x", centerX - totalWidth / 2)
+          .attr("y", centerY - logoHeight / 2)
+          .attr("width", logoWidth)
+          .attr("height", logoHeight)
+          .attr("viewBox", "0 0 2.174 1.438")
+          .append("path")
+          .attr("d", NVIDIA_LOGO_PATH)
+          .attr("fill", "#76b900");
+
+        labelGroup
+          .append("text")
+          .attr("x", centerX - totalWidth / 2 + logoWidth + gap)
+          .attr("y", centerY)
+          .attr("text-anchor", "start")
+          .attr("dominant-baseline", "middle")
+          .attr("fill", "#8a7a56")
+          .attr("font-size", textSize)
+          .attr("font-family", "monospace")
+          .attr("font-weight", "700")
+          .text("NVIDIA");
+      } else if (isLinuxLaptop) {
+        // Linux Laptop — same shape as MacBook but with Tux logo
+        iconBaseWidth = nodeRadius * 1.6;
+        iconBaseHeight = nodeRadius * 1.15;
+        const x = nodeInfo.x - iconBaseWidth / 2;
+        const y = nodeInfo.y - iconBaseHeight / 2;
+
+        const screenHeight = iconBaseHeight * 0.7;
+        const baseHeight = iconBaseHeight * 0.3;
+        const screenWidth = iconBaseWidth * 0.85;
+        const screenX = nodeInfo.x - screenWidth / 2;
+        const screenBezel = 3;
+
+        const linuxScreenClipId = `linux-screen-${nodeInfo.id.replace(/[^a-zA-Z0-9]/g, "-")}`;
+        defs
+          .append("clipPath")
+          .attr("id", linuxScreenClipId)
+          .append("rect")
+          .attr("x", screenX + screenBezel)
+          .attr("y", y + screenBezel)
+          .attr("width", screenWidth - screenBezel * 2)
+          .attr("height", screenHeight - screenBezel * 2)
+          .attr("rx", 2);
+
+        // Screen outer frame
+        nodeG
+          .append("rect")
+          .attr("class", "node-outline")
+          .attr("x", screenX)
+          .attr("y", y)
+          .attr("width", screenWidth)
+          .attr("height", screenHeight)
+          .attr("rx", 3)
+          .attr("fill", "#1a1a1a")
+          .attr("stroke", wireColor)
+          .attr("stroke-width", strokeWidth);
+
+        // Screen inner
+        nodeG
+          .append("rect")
+          .attr("x", screenX + screenBezel)
+          .attr("y", y + screenBezel)
+          .attr("width", screenWidth - screenBezel * 2)
+          .attr("height", screenHeight - screenBezel * 2)
+          .attr("rx", 2)
+          .attr("fill", "#0a0a12");
+
+        // Memory fill on screen
+        if (ramUsagePercent > 0) {
+          const memFillTotalHeight = screenHeight - screenBezel * 2;
+          const memFillActualHeight =
+            (ramUsagePercent / 100) * memFillTotalHeight;
+          nodeG
+            .append("rect")
+            .attr("x", screenX + screenBezel)
+            .attr(
+              "y",
+              y + screenBezel + (memFillTotalHeight - memFillActualHeight),
+            )
+            .attr("width", screenWidth - screenBezel * 2)
+            .attr("height", memFillActualHeight)
+            .attr("fill", "rgba(255,215,0,0.85)")
+            .attr("clip-path", `url(#${linuxScreenClipId})`);
+        }
+
+        // Terminal prompt on screen
+        nodeG
+          .append("text")
+          .attr("x", nodeInfo.x)
+          .attr("y", y + screenHeight / 2)
+          .attr("text-anchor", "middle")
+          .attr("dominant-baseline", "middle")
+          .attr("fill", "#FFFFFF")
+          .attr("opacity", 0.9)
+          .attr("font-size", screenHeight * 0.25)
+          .attr("font-family", "SF Mono, Monaco, monospace")
+          .attr("font-weight", "700")
+          .text(">_");
+
+        // Keyboard base (trapezoidal)
+        const baseY = y + screenHeight;
+        const baseTopWidth = screenWidth;
+        const baseBottomWidth = iconBaseWidth;
+        const baseTopX = nodeInfo.x - baseTopWidth / 2;
+        const baseBottomX = nodeInfo.x - baseBottomWidth / 2;
+
+        nodeG
+          .append("path")
+          .attr(
+            "d",
+            `M ${baseTopX} ${baseY} L ${baseTopX + baseTopWidth} ${baseY} L ${baseBottomX + baseBottomWidth} ${baseY + baseHeight} L ${baseBottomX} ${baseY + baseHeight} Z`,
+          )
+          .attr("fill", "#2c2c2c")
+          .attr("stroke", wireColor)
+          .attr("stroke-width", 1);
+
+        // Keyboard area
+        const keyboardX = baseTopX + 6;
+        const keyboardY = baseY + 3;
+        const keyboardWidth = baseTopWidth - 12;
+        const keyboardHeight = baseHeight * 0.55;
+        nodeG
+          .append("rect")
+          .attr("x", keyboardX)
+          .attr("y", keyboardY)
+          .attr("width", keyboardWidth)
+          .attr("height", keyboardHeight)
+          .attr("fill", "rgba(0,0,0,0.2)")
+          .attr("rx", 2);
+
+        // Trackpad
+        const trackpadWidth = baseTopWidth * 0.4;
+        const trackpadX = nodeInfo.x - trackpadWidth / 2;
+        const trackpadY = baseY + keyboardHeight + 5;
+        const trackpadHeight = baseHeight * 0.3;
+        nodeG
+          .append("rect")
+          .attr("x", trackpadX)
+          .attr("y", trackpadY)
+          .attr("width", trackpadWidth)
+          .attr("height", trackpadHeight)
+          .attr("fill", "rgba(255,255,255,0.08)")
+          .attr("rx", 2);
+      } else if (isLinux) {
+        // Linux Desktop — same shape as Mac Studio but with Tux logo
+        iconBaseWidth = nodeRadius * 1.25;
+        iconBaseHeight = nodeRadius * 0.85;
+        const x = nodeInfo.x - iconBaseWidth / 2;
+        const y = nodeInfo.y - iconBaseHeight / 2;
+        const cornerRadius = 4;
+        const topSurfaceHeight = iconBaseHeight * 0.15;
+
+        const linuxDesktopClipId = `linux-desktop-${nodeInfo.id.replace(/[^a-zA-Z0-9]/g, "-")}`;
+        defs
+          .append("clipPath")
+          .attr("id", linuxDesktopClipId)
+          .append("rect")
+          .attr("x", x)
+          .attr("y", y + topSurfaceHeight)
+          .attr("width", iconBaseWidth)
+          .attr("height", iconBaseHeight - topSurfaceHeight)
+          .attr("rx", cornerRadius - 1);
+
+        // Main body
+        nodeG
+          .append("rect")
+          .attr("class", "node-outline")
+          .attr("x", x)
+          .attr("y", y)
+          .attr("width", iconBaseWidth)
+          .attr("height", iconBaseHeight)
+          .attr("rx", cornerRadius)
+          .attr("fill", "#1a1a1a")
+          .attr("stroke", wireColor)
+          .attr("stroke-width", strokeWidth);
+
+        // Memory fill
+        if (ramUsagePercent > 0) {
+          const memFillTotalHeight = iconBaseHeight - topSurfaceHeight;
+          const memFillActualHeight =
+            (ramUsagePercent / 100) * memFillTotalHeight;
+          nodeG
+            .append("rect")
+            .attr("x", x)
+            .attr(
+              "y",
+              y + topSurfaceHeight + (memFillTotalHeight - memFillActualHeight),
+            )
+            .attr("width", iconBaseWidth)
+            .attr("height", memFillActualHeight)
+            .attr("fill", "rgba(255,215,0,0.75)")
+            .attr("clip-path", `url(#${linuxDesktopClipId})`);
+        }
+
+        // Terminal prompt on front face
+        nodeG
+          .append("text")
+          .attr("x", nodeInfo.x)
+          .attr(
+            "y",
+            y + topSurfaceHeight + (iconBaseHeight - topSurfaceHeight) / 2,
+          )
+          .attr("text-anchor", "middle")
+          .attr("dominant-baseline", "middle")
+          .attr("fill", "rgba(255,255,255,0.5)")
+          .attr("font-size", (iconBaseHeight - topSurfaceHeight) * 0.4)
+          .attr("font-family", "SF Mono, Monaco, monospace")
+          .attr("font-weight", "700")
+          .text(">_");
+      } else if (modelLower === "mac studio") {
        // Mac Studio - classic cube with memory fill
        iconBaseWidth = nodeRadius * 1.25;
        iconBaseHeight = nodeRadius * 0.85;
@@ -1182,8 +1568,12 @@
          debugLabelY += debugLineHeight;
        }

-        const identity = identitiesData[nodeInfo.id];
-        if (identity?.osVersion) {
+        const dbgIdentity = identitiesData[nodeInfo.id];
+        if (dbgIdentity?.osVersion) {
+          const osLabel =
+            dbgIdentity.osVersion === "Linux"
+              ? "Linux"
+              : `macOS ${dbgIdentity.osVersion}${dbgIdentity.osBuildVersion ? ` (${dbgIdentity.osBuildVersion})` : ""}`;
          nodeG
            .append("text")
            .attr("x", nodeInfo.x)
@@ -1192,9 +1582,7 @@
            .attr("fill", "rgba(179,179,179,0.7)")
            .attr("font-size", debugFontSize)
            .attr("font-family", "SF Mono, Monaco, monospace")
-            .text(
-              `macOS ${identity.osVersion}${identity.osBuildVersion ? ` (${identity.osBuildVersion})` : ""}`,
-            );
+            .text(osLabel);
        }
      }
    });
@@ -0,0 +1,12 @@
+model_id = "unsloth/DeepSeek-V3.1-GGUF"
+n_layers = 61
+hidden_size = 7168
+supports_tensor = true
+tasks = ["TextGeneration"]
+family = "deepseek"
+quantization = "GGUF"
+base_model = "DeepSeek V3.1"
+capabilities = ["text", "thinking", "thinking_toggle"]
+
+[storage_size]
+in_bytes = 684531386000
@@ -0,0 +1,12 @@
+model_id = "unsloth/GLM-4.7-Flash-GGUF"
+n_layers = 47
+hidden_size = 2048
+supports_tensor = true
+tasks = ["TextGeneration"]
+family = "glm"
+quantization = "GGUF"
+base_model = "GLM 4.7 Flash"
+capabilities = ["text", "thinking", "thinking_toggle"]
+
+[storage_size]
+in_bytes = 35624923488
@@ -0,0 +1,12 @@
+model_id = "unsloth/GLM-5-GGUF"
+n_layers = 78
+hidden_size = 6144
+supports_tensor = true
+tasks = ["TextGeneration"]
+family = "glm"
+quantization = "GGUF"
+base_model = "GLM-5"
+capabilities = ["text", "thinking"]
+
+[storage_size]
+in_bytes = 1487822475264
@@ -0,0 +1,12 @@
+model_id = "unsloth/Llama-3.2-1B-Instruct-GGUF"
+n_layers = 16
+hidden_size = 2048
+supports_tensor = true
+tasks = ["TextGeneration"]
+family = "llama"
+quantization = "GGUF"
+base_model = "Llama 3.2 1B"
+capabilities = ["text"]
+
+[storage_size]
+in_bytes = 2479595264
@@ -0,0 +1,12 @@
+model_id = "unsloth/Llama-3.2-3B-Instruct-GGUF"
+n_layers = 28
+hidden_size = 3072
+supports_tensor = true
+tasks = ["TextGeneration"]
+family = "llama"
+quantization = "GGUF"
+base_model = "Llama 3.2 3B"
+capabilities = ["text"]
+
+[storage_size]
+in_bytes = 6433687744
@@ -0,0 +1,12 @@
+model_id = "unsloth/Llama-3.3-70B-Instruct-GGUF"
+n_layers = 80
+hidden_size = 8192
+supports_tensor = true
+tasks = ["TextGeneration"]
+family = "llama"
+quantization = "GGUF"
+base_model = "Llama 3.3 70B"
+capabilities = ["text"]
+
+[storage_size]
+in_bytes = 144383672320
@@ -0,0 +1,12 @@
+model_id = "unsloth/MiniMax-M2.5-GGUF"
+n_layers = 62
+hidden_size = 3072
+supports_tensor = true
+tasks = ["TextGeneration"]
+family = "minimax"
+quantization = "GGUF"
+base_model = "MiniMax M2.5"
+capabilities = ["text", "thinking"]
+
+[storage_size]
+in_bytes = 228703644928
@@ -0,0 +1,12 @@
+model_id = "unsloth/Qwen3-0.6B-GGUF"
+n_layers = 28
+hidden_size = 1024
+supports_tensor = true
+tasks = ["TextGeneration"]
+family = "qwen"
+quantization = "GGUF"
+base_model = "Qwen3 0.6B"
+capabilities = ["text", "thinking", "thinking_toggle"]
+
+[storage_size]
+in_bytes = 1198182848
@@ -0,0 +1,12 @@
+model_id = "unsloth/Qwen3-30B-A3B-GGUF"
+n_layers = 48
+hidden_size = 2048
+supports_tensor = true
+tasks = ["TextGeneration"]
+family = "qwen"
+quantization = "GGUF"
+base_model = "Qwen3 30B A3B"
+capabilities = ["text", "thinking", "thinking_toggle"]
+
+[storage_size]
+in_bytes = 30532122624
@@ -0,0 +1,12 @@
+model_id = "unsloth/Qwen3-Coder-Next-GGUF"
+n_layers = 48
+hidden_size = 2048
+supports_tensor = true
+tasks = ["TextGeneration"]
+family = "qwen"
+quantization = "GGUF"
+base_model = "Qwen3 Coder Next"
+capabilities = ["text", "code"]
+
+[storage_size]
+in_bytes = 157548627945
@@ -0,0 +1,12 @@
+model_id = "unsloth/Qwen3.5-122B-A10B-GGUF"
+n_layers = 48
+hidden_size = 3072
+supports_tensor = true
+tasks = ["TextGeneration"]
+family = "qwen"
+quantization = "GGUF"
+base_model = "Qwen3.5 122B A10B"
+capabilities = ["text", "thinking", "thinking_toggle"]
+
+[storage_size]
+in_bytes = 245125640160
@@ -0,0 +1,12 @@
+model_id = "unsloth/Qwen3.5-27B-GGUF"
+n_layers = 64
+hidden_size = 5120
+supports_tensor = true
+tasks = ["TextGeneration"]
+family = "qwen"
+quantization = "GGUF"
+base_model = "Qwen3.5 27B"
+capabilities = ["text", "thinking"]
+
+[storage_size]
+in_bytes = 27781427952
@@ -0,0 +1,12 @@
+model_id = "unsloth/Qwen3.5-2B-GGUF"
+n_layers = 24
+hidden_size = 2048
+supports_tensor = true
+tasks = ["TextGeneration"]
+family = "qwen"
+quantization = "GGUF"
+base_model = "Qwen3.5 2B"
+capabilities = ["text", "thinking"]
+
+[storage_size]
+in_bytes = 3775709216
@@ -0,0 +1,12 @@
+model_id = "unsloth/Qwen3.5-35B-A3B-GGUF"
+n_layers = 40
+hidden_size = 2048
+supports_tensor = true
+tasks = ["TextGeneration"]
+family = "qwen"
+quantization = "GGUF"
+base_model = "Qwen3.5 35B A3B"
+capabilities = ["text", "thinking", "thinking_toggle"]
+
+[storage_size]
+in_bytes = 35951822704
@@ -0,0 +1,12 @@
+model_id = "unsloth/Qwen3.5-9B-GGUF"
+n_layers = 32
+hidden_size = 4096
+supports_tensor = true
+tasks = ["TextGeneration"]
+family = "qwen"
+quantization = "GGUF"
+base_model = "Qwen3.5 9B"
+capabilities = ["text", "thinking"]
+
+[storage_size]
+in_bytes = 17920697312
@@ -0,0 +1,12 @@
+model_id = "unsloth/gpt-oss-120b-GGUF"
+n_layers = 36
+hidden_size = 2880
+supports_tensor = true
+tasks = ["TextGeneration"]
+family = "gpt-oss"
+quantization = "GGUF"
+base_model = "GPT-OSS 120B"
+capabilities = ["text", "thinking"]
+
+[storage_size]
+in_bytes = 65369017728
@@ -0,0 +1,12 @@
+model_id = "unsloth/gpt-oss-20b-GGUF"
+n_layers = 24
+hidden_size = 2880
+supports_tensor = true
+tasks = ["TextGeneration"]
+family = "gpt-oss"
+quantization = "GGUF"
+base_model = "GPT-OSS 20B"
+capabilities = ["text", "thinking"]
+
+[storage_size]
+in_bytes = 13792639168
@@ -0,0 +1,265 @@
+import json
+from pathlib import Path
+from typing import Any
+
+import mlx.core as mx
+from loguru import logger
+
+# ---------------------------------------------------------------------------
+# Architecture mapping: GGUF general.architecture → HuggingFace
+# ---------------------------------------------------------------------------
+
+_ARCH_TO_HF: dict[str, list[str]] = {
+    "llama": ["LlamaForCausalLM"],
+    "qwen2": ["Qwen2ForCausalLM"],
+    "qwen3": ["Qwen3ForCausalLM"],
+    "mistral": ["MistralForCausalLM"],
+    "phi3": ["Phi3ForCausalLM"],
+    "gemma": ["GemmaForCausalLM"],
+    "gemma2": ["Gemma2ForCausalLM"],
+    "starcoder2": ["Starcoder2ForCausalLM"],
+}
+
+_ARCH_TO_MODEL_TYPE: dict[str, str] = {
+    "llama": "llama",
+    "qwen2": "qwen2",
+    "qwen3": "qwen3",
+    "mistral": "mistral",
+    "phi3": "phi3",
+    "gemma": "gemma",
+    "gemma2": "gemma2",
+    "starcoder2": "starcoder2",
+}
+
+# GGUF general.file_type → quantization config for mlx_lm
+_FILE_TYPE_TO_QUANT: dict[int, dict[str, int] | None] = {
+    0: None,  # ALL_F32
+    1: None,  # MOSTLY_F16
+    2: {"group_size": 32, "bits": 4},  # MOSTLY_Q4_0
+    3: {"group_size": 32, "bits": 4},  # MOSTLY_Q4_1
+    7: {"group_size": 32, "bits": 8},  # MOSTLY_Q8_0
+}
+
+# Maximum safetensors shard size in bytes (5 GB)
+_MAX_SHARD_BYTES = 5 * 1024 * 1024 * 1024
+
+
+# ---------------------------------------------------------------------------
+# Weight name translation
+# ---------------------------------------------------------------------------
+
+
+def translate_gguf_weight_name(name: str) -> str:
+    """Translate a GGUF tensor name to HuggingFace-style name.
+
+    Vendored from: https://github.com/ml-explore/mlx-examples/blob/main/llms/gguf_llm/models.py
+    (translate_weight_names function)
+
+    Covers the standard Llama-family convention which is shared by
+    Llama, Qwen, Mistral, Gemma, Phi, and most other transformer models.
+    """
+    name = name.replace("blk.", "model.layers.")
+    name = name.replace("ffn_gate", "mlp.gate_proj")
+    name = name.replace("ffn_down", "mlp.down_proj")
+    name = name.replace("ffn_up", "mlp.up_proj")
+    name = name.replace("attn_q", "self_attn.q_proj")
+    name = name.replace("attn_k", "self_attn.k_proj")
+    name = name.replace("attn_v", "self_attn.v_proj")
+    name = name.replace("attn_output", "self_attn.o_proj")
+    name = name.replace("attn_norm", "input_layernorm")
+    name = name.replace("ffn_norm", "post_attention_layernorm")
+    name = name.replace("token_embd", "model.embed_tokens")
+    name = name.replace("output_norm", "model.norm")
+    # "output" must come last — it's a substring of "attn_output" / "output_norm"
+    if name == "output.weight":
+        name = "lm_head.weight"
+    return name
+
+
+# ---------------------------------------------------------------------------
+# Metadata extraction
+# ---------------------------------------------------------------------------
+
+
+def _meta_val(metadata: dict[str, Any], key: str) -> int | float | str:
+    """Get a metadata value, converting mx.array scalars to Python types."""
+    val = metadata[key]  # pyright: ignore[reportAny]
+    if isinstance(val, mx.array):
+        return int(val.item())  # pyright: ignore[reportAny]
+    if isinstance(val, (int, float)):
+        return val
+    return str(val)  # pyright: ignore[reportAny]
+
+
+def extract_config_from_metadata(metadata: dict[str, Any]) -> dict[str, Any]:
+    """Build a HuggingFace-style config.json dict from GGUF metadata."""
+    raw_arch: str = str(metadata.get("general.architecture", "llama"))  # pyright: ignore[reportAny]
+    arch: str = raw_arch.lower()
+    prefix = f"{arch}."
+
+    def get(suffix: str, default: int | float | str | None = None) -> int | float | str | None:
+        key = prefix + suffix
+        if key not in metadata:
+            return default
+        return _meta_val(metadata, key)
+
+    # Vocab size: prefer metadata field, fall back to tokenizer token count
+    vocab_size: int | float | str | None = get("vocab_size")
+    if vocab_size is None:
+        tokens: Any = metadata.get("tokenizer.ggml.tokens")
+        vocab_size = len(tokens) if tokens is not None else 0  # pyright: ignore[reportAny]
+
+    config: dict[str, Any] = {
+        "architectures": _ARCH_TO_HF.get(arch, ["LlamaForCausalLM"]),
+        "model_type": _ARCH_TO_MODEL_TYPE.get(arch, arch),
+        "hidden_size": get("embedding_length", 0),
+        "num_hidden_layers": get("block_count", 0),
+        "num_attention_heads": get("attention.head_count", 0),
+        "num_key_value_heads": get("attention.head_count_kv"),
+        "intermediate_size": get("feed_forward_length", 0),
+        "vocab_size": vocab_size,
+        "rms_norm_eps": get("attention.layer_norm_rms_epsilon", 1e-5),
+        "rope_theta": get("rope.freq_base", 10000.0),
+        "max_position_embeddings": get("context_length", 4096),
+        "tie_word_embeddings": False,
+    }
+
+    # Remove None values
+    config = {k: v for k, v in config.items() if v is not None}  # pyright: ignore[reportAny]
+
+    # Quantization config
+    file_type_raw: Any = metadata.get("general.file_type")
+    if file_type_raw is not None:
+        file_type = int(_meta_val(metadata, "general.file_type"))
+        quant = _FILE_TYPE_TO_QUANT.get(file_type)
+        if quant is not None:
+            config["quantization_config"] = quant
+
+    return config
+
+
+# ---------------------------------------------------------------------------
+# Conversion
+# ---------------------------------------------------------------------------
+
+
+def _load_gguf(gguf_path: Path) -> tuple[dict[str, mx.array], dict[str, Any]]:
+    """Load a GGUF file, returning (weights, metadata)."""
+    result: Any = mx.load(str(gguf_path), return_metadata=True)  # pyright: ignore[reportAny]
+    weights: dict[str, mx.array] = result[0]  # pyright: ignore[reportAny]
+    metadata: dict[str, Any] = result[1]  # pyright: ignore[reportAny]
+    return weights, metadata
+
+
+def convert_gguf_to_safetensors(gguf_path: Path, output_dir: Path) -> None:
+    """Convert a GGUF file to safetensors + config.json in output_dir.
+
+    After conversion:
+    - output_dir contains model*.safetensors, config.json,
+      model.safetensors.index.json, and a .gguf_converted marker
+    - The original .gguf file is deleted to save disk space
+    """
+    logger.info(f"Loading GGUF file: {gguf_path}")
+    weights, metadata = _load_gguf(gguf_path)
+
+    # Translate weight names
+    translated: dict[str, mx.array] = {}
+    for name, tensor in weights.items():
+        new_name = translate_gguf_weight_name(name)
+        translated[new_name] = tensor
+    del weights  # Free memory
+
+    # Extract and write config.json
+    config = extract_config_from_metadata(metadata)
+    config_path = output_dir / "config.json"
+    if not config_path.exists():
+        logger.info(f"Writing config.json to {config_path}")
+        config_path.write_text(json.dumps(config, indent=2, default=str))
+
+    # Shard weights and save as safetensors
+    weight_map: dict[str, str] = {}
+    shards: list[tuple[str, dict[str, mx.array]]] = []
+    current_shard: dict[str, mx.array] = {}
+    current_size = 0
+    shard_idx = 0
+
+    for name, tensor in sorted(translated.items()):
+        tensor_bytes = tensor.nbytes
+        if current_size + tensor_bytes > _MAX_SHARD_BYTES and current_shard:
+            shard_idx += 1
+            shards.append((f"model-{shard_idx:05d}", current_shard))
+            current_shard = {}
+            current_size = 0
+        current_shard[name] = tensor
+        current_size += tensor_bytes
+
+    if current_shard:
+        shard_idx += 1
+        shards.append((f"model-{shard_idx:05d}", current_shard))
+
+    total_shards = len(shards)
+    total_size = 0
+
+    for shard_name, shard_weights in shards:
+        if total_shards == 1:
+            filename = "model.safetensors"
+        else:
+            filename = f"{shard_name}-of-{total_shards:05d}.safetensors"
+
+        shard_path = output_dir / filename
+        logger.info(f"Writing {filename} ({len(shard_weights)} tensors)")
+        mx.save_safetensors(str(shard_path), shard_weights)  # pyright: ignore[reportUnknownMemberType]
+
+        for weight_name in shard_weights:
+            weight_map[weight_name] = filename
+            total_size += shard_weights[weight_name].nbytes
+
+    del translated  # Free memory
+
+    # Write safetensors index
+    index = {
+        "metadata": {"total_size": total_size},
+        "weight_map": weight_map,
+    }
+    index_path = output_dir / "model.safetensors.index.json"
+    logger.info(f"Writing safetensors index to {index_path}")
+    index_path.write_text(json.dumps(index, indent=2))
+
+    # Write marker
+    marker_path = output_dir / ".gguf_converted"
+    marker_path.write_text(f"Converted from {gguf_path.name}\n")
+
+    # Delete original GGUF file
+    logger.info(f"Deleting original GGUF file: {gguf_path}")
+    gguf_path.unlink()
+
+    logger.info("GGUF → safetensors conversion complete")
+
+
+# ---------------------------------------------------------------------------
+# Helpers for model card creation
+# ---------------------------------------------------------------------------
+
+
+def is_gguf_repo(file_paths: list[str]) -> bool:
+    """Check if a file list represents a GGUF-only repo (has .gguf but no .safetensors)."""
+    has_gguf = any(p.endswith(".gguf") for p in file_paths)
+    has_safetensors = any(p.endswith(".safetensors") for p in file_paths)
+    return has_gguf and not has_safetensors
+
+
+def select_gguf_file(file_paths: list[str], file_sizes: dict[str, int | None] | None = None) -> str | None:
+    """Pick the best GGUF file from a file list.
+
+    Prefers the largest file (usually highest quality quant).
+    """
+    gguf_files = [p for p in file_paths if p.endswith(".gguf")]
+    if not gguf_files:
+        return None
+
+    if file_sizes:
+        # Pick largest
+        return max(gguf_files, key=lambda p: file_sizes.get(p, 0) or 0)
+
+    # No size info — just pick first
+    return gguf_files[0]
@@ -115,6 +115,19 @@ class ResumableShardDownloader(ShardDownloader):
            allow_patterns=allow_patterns,
            skip_internet=self.offline,
        )
+
+        if not config_only and target_dir.suffix == ".gguf":
+            output_dir = target_dir.parent
+            marker = output_dir / ".gguf_converted"
+            if not marker.exists():
+                from exo.download.gguf_conversion import convert_gguf_to_safetensors
+
+                logger.info(f"Converting GGUF to safetensors: {target_dir}")
+                await asyncio.to_thread(
+                    convert_gguf_to_safetensors, target_dir, output_dir
+                )
+            target_dir = output_dir
+
        return target_dir

    async def get_shard_download_status(
@@ -126,10 +126,12 @@ class ModelCard(CamelCaseModel):
    @staticmethod
    async def fetch_from_hf(model_id: ModelId) -> "ModelCard":
        """Fetches storage size and number of layers for a Hugging Face model, returns Pydantic ModelMeta."""
-        # TODO: failure if files do not exist
-        config_data = await fetch_config_data(model_id)
+        try:
+            config_data = await fetch_config_data(model_id)
+        except FileNotFoundError:
+            config_data = await fetch_config_data_for_gguf(model_id)
        num_layers = config_data.layer_count
-        mem_size_bytes = await fetch_safetensors_size(model_id)
+        mem_size_bytes = await fetch_model_size(model_id)

        mc = ModelCard(
            model_id=ModelId(model_id),
@@ -244,8 +246,38 @@ async def fetch_config_data(model_id: ModelId) -> ConfigData:
        return ConfigData.model_validate_json(await f.read())


-async def fetch_safetensors_size(model_id: ModelId) -> Memory:
-    """Gets model size from safetensors index or falls back to HF API."""
+async def fetch_config_data_for_gguf(model_id: ModelId) -> ConfigData:
+    # Try to get the base model from HF API
+    info = model_info(model_id)
+    base_model_id: str | None = None
+
+    # card_data has base_model field
+    if info.card_data is not None:
+        base_model_raw: str | list[str] | None = getattr(info.card_data, "base_model", None)  # pyright: ignore[reportAttributeAccessIssue, reportAny]
+        if isinstance(base_model_raw, str):
+            base_model_id = base_model_raw
+        elif isinstance(base_model_raw, list) and base_model_raw:
+            # Pick the first non-quantized base model
+            for bm in base_model_raw:
+                if isinstance(bm, str) and "quantized:" not in bm:
+                    base_model_id = bm
+                    break
+
+    if base_model_id is not None:
+        logger.info(f"GGUF repo {model_id} references base_model={base_model_id}, fetching its config")
+        try:
+            return await fetch_config_data(ModelId(base_model_id))
+        except Exception as exc:
+            logger.warning(f"Failed to fetch config from base model {base_model_id}: {exc}")
+
+    raise ValueError(
+        f"Model {model_id} has no config.json and no base_model reference. "
+        f"Cannot determine model architecture."
+    )
+
+
+async def fetch_model_size(model_id: ModelId) -> Memory:
+    """Gets model size from safetensors index, HF API, or GGUF file sizes."""
    from exo.download.download_utils import (
        download_file_with_retry,
        ensure_models_dir,
@@ -254,23 +286,36 @@ async def fetch_safetensors_size(model_id: ModelId) -> Memory:

    target_dir = (await ensure_models_dir()) / model_id.normalize()
    await aios.makedirs(target_dir, exist_ok=True)
-    index_path = await download_file_with_retry(
-        model_id,
-        "main",
-        "model.safetensors.index.json",
-        target_dir,
-        lambda curr_bytes, total_bytes, is_renamed: logger.debug(
-            f"Downloading model.safetensors.index.json for {model_id}: {curr_bytes}/{total_bytes} ({is_renamed=})"
-        ),
-    )
-    async with aiofiles.open(index_path, "r") as f:
-        index_data = ModelSafetensorsIndex.model_validate_json(await f.read())
+    try:
+        index_path = await download_file_with_retry(
+            model_id,
+            "main",
+            "model.safetensors.index.json",
+            target_dir,
+            lambda curr_bytes, total_bytes, is_renamed: logger.debug(
+                f"Downloading model.safetensors.index.json for {model_id}: {curr_bytes}/{total_bytes} ({is_renamed=})"
+            ),
+        )
+        async with aiofiles.open(index_path, "r") as f:
+            index_data = ModelSafetensorsIndex.model_validate_json(await f.read())

-    metadata = index_data.metadata
-    if metadata is not None:
-        return Memory.from_bytes(metadata.total_size)
+        metadata = index_data.metadata
+        if metadata is not None:
+            return Memory.from_bytes(metadata.total_size)
+    except FileNotFoundError:
+        logger.debug(f"No safetensors index for {model_id}, falling back to HF API")

    info = model_info(model_id)
-    if info.safetensors is None:
-        raise ValueError(f"No safetensors info found for {model_id}")
-    return Memory.from_bytes(info.safetensors.total)
+    if info.safetensors is not None:
+        return Memory.from_bytes(info.safetensors.total)
+
+    # GGUF fallback: sum .gguf file sizes
+    gguf_size = sum(
+        s.size
+        for s in (info.siblings or [])
+        if s.rfilename.endswith(".gguf") and s.size is not None
+    )
+    if gguf_size > 0:
+        return Memory.from_bytes(gguf_size)
+
+    raise ValueError(f"No safetensors or GGUF files found for {model_id}")
@@ -1,6 +1,7 @@
 import platform
 import socket
 import sys
+from pathlib import Path
 from subprocess import CalledProcessError

 import psutil
@@ -117,12 +118,97 @@ async def get_network_interfaces() -> list[NetworkInterfaceInfo]:
    return interfaces_info


+def _read_dmi_field(name: str) -> str | None:
+    """Read a single DMI sysfs field, returning ``None`` on failure."""
+    try:
+        path = Path(f"/sys/class/dmi/id/{name}")
+        if path.exists():
+            return path.read_text().strip()
+    except (OSError, PermissionError):
+        pass
+    return None
+
+
+async def _get_linux_model_and_chip() -> tuple[str, str]:
+    """Get Linux system information using DMI and /proc/cpuinfo.
+
+    Detects NVIDIA DGX Spark (DMI product_name ``"DGX_Spark"``) and other
+    NVIDIA systems via the ``sys_vendor`` DMI field, falling back to generic
+    Linux identification.
+    """
+    model = "Linux"
+    chip = "Unknown Chip"
+
+    product_name = _read_dmi_field("product_name")
+    sys_vendor = _read_dmi_field("sys_vendor")
+
+    # DGX Spark: DMI product_name may be "DGX_Spark" or "gx10" variant
+    product_lower = (product_name or "").lower()
+    if product_name and ("dgx" in product_lower or "gx10" in product_lower):
+        model = "DGX Spark"
+        try:
+            process = await run_process(
+                ["nvidia-smi", "--query-gpu=name", "--format=csv,noheader"]
+            )
+            gpu_name = process.stdout.decode().strip().split("\n")[0]
+            chip = gpu_name if gpu_name and gpu_name != "[N/A]" else "NVIDIA GB10"
+        except (CalledProcessError, FileNotFoundError):
+            chip = "NVIDIA GB10"
+        return (model, chip)
+
+    # Other NVIDIA systems (sys_vendor contains "NVIDIA")
+    if sys_vendor and "NVIDIA" in sys_vendor:
+        model = product_name.replace("_", " ") if product_name else "NVIDIA System"
+        try:
+            process = await run_process(
+                ["nvidia-smi", "--query-gpu=name", "--format=csv,noheader"]
+            )
+            gpu_name = process.stdout.decode().strip().split("\n")[0]
+            if gpu_name and gpu_name != "[N/A]":
+                chip = gpu_name
+        except (CalledProcessError, FileNotFoundError):
+            pass
+        return (model, chip)
+
+    # Generic Linux — detect laptop vs desktop via chassis_type
+    # SMBIOS chassis types: 8,9,10,14,31,32 = portable/laptop
+    chassis_type = _read_dmi_field("chassis_type")
+    laptop_chassis_types = {"8", "9", "10", "14", "31", "32"}
+    if chassis_type in laptop_chassis_types:
+        model = "Linux Laptop"
+    elif chassis_type is not None:
+        model = "Linux Desktop"
+
+    # Also check for battery as a fallback laptop indicator
+    if model == "Linux" and Path("/sys/class/power_supply/BAT0").exists():
+        model = "Linux Laptop"
+
+    # Use /proc/cpuinfo for chip
+    cpuinfo_path = Path("/proc/cpuinfo")
+    if cpuinfo_path.exists():
+        try:
+            for line in cpuinfo_path.read_text().splitlines():
+                if line.startswith("model name"):
+                    chip = line.split(":", 1)[1].strip()
+                    break
+        except OSError:
+            pass
+
+    return (model, chip)
+
+
 async def get_model_and_chip() -> tuple[str, str]:
-    """Get Mac system information using system_profiler."""
+    """Get system model and chip information.
+
+    On macOS, uses ``system_profiler``.  On Linux, reads DMI data from
+    sysfs and CPU info from ``/proc/cpuinfo``.
+    """
    model = "Unknown Model"
    chip = "Unknown Chip"

-    # TODO: better non mac support
+    if sys.platform == "linux":
+        return await _get_linux_model_and_chip()
+
    if sys.platform != "darwin":
        return (model, chip)

@@ -568,14 +568,21 @@ def apply_chat_template(
                    "Patched lossy chat template (removed inner_type length guard)"
                )

-    prompt: str = tokenizer.apply_chat_template(
-        formatted_messages,
-        tokenize=False,
-        add_generation_prompt=True,
-        tools=task_params.tools,
-        **({"chat_template": patched_template} if patched_template is not None else {}),
-        **extra_kwargs,
-    )
+    try:
+        prompt: str = tokenizer.apply_chat_template(
+            formatted_messages,
+            tokenize=False,
+            add_generation_prompt=True,
+            tools=task_params.tools,
+            **({"chat_template": patched_template} if patched_template is not None else {}),
+            **extra_kwargs,
+        )
+    except ValueError:
+        logger.warning("No chat template found, falling back to raw message concatenation")
+        prompt = "\n".join(
+            f"{msg.get('role', 'user')}: {msg.get('content', '')}"
+            for msg in formatted_messages
+        )

    if task_params.tools and _schemas_lost_in_prompt(prompt, task_params.tools):
        logger.warning("Chat template lost nested tool schemas even after patching")
@@ -0,0 +1,52 @@
+#!/usr/bin/env bash
+set -Eeuo pipefail
+
+msg() {
+  printf '\n==> %s\n' "$*"
+}
+
+msg "Bringing down old Apple USB fallback interfaces"
+sudo ifconfig en4 down 2>/dev/null || true
+sudo ifconfig en5 down 2>/dev/null || true
+sudo ifconfig en6 down 2>/dev/null || true
+
+msg "Bringing down current anpi interfaces"
+sudo ifconfig anpi0 down 2>/dev/null || true
+sudo ifconfig anpi1 down 2>/dev/null || true
+sudo ifconfig anpi2 down 2>/dev/null || true
+
+msg "Stopping likely bad exporter services"
+sudo launchctl bootout system/com.apple.usbmuxd 2>/dev/null || true
+sudo launchctl bootout system/com.apple.remoted 2>/dev/null || true
+launchctl bootout "gui/$(id -u)/com.apple.remoted" 2>/dev/null || true
+
+sudo pkill -x usbmuxd 2>/dev/null || true
+sudo pkill -x remoted 2>/dev/null || true
+
+cat <<'EOF'
+
+Now unplug and replug the cable.
+
+After reconnect, press Enter to continue.
+EOF
+
+read -r _
+
+msg "Bringing up anpi interfaces and requesting DHCP"
+for ifn in anpi0 anpi1 anpi2; do
+  sudo ifconfig "$ifn" up 2>/dev/null || true
+  sudo ipconfig set "$ifn" DHCP 2>/dev/null || true
+done
+
+sleep 3
+
+msg "Resulting interface state"
+for ifn in anpi0 anpi1 anpi2; do
+  echo
+  echo "--- $ifn ---"
+  ifconfig "$ifn" 2>/dev/null || true
+  ipconfig getifaddr "$ifn" 2>/dev/null || true
+done
+
+echo
+echo "Use whichever interface got 10.42.0.x or 10.43.0.x"
@@ -0,0 +1,180 @@
+#!/usr/bin/env bash
+set -Eeuo pipefail
+
+KVER="$(uname -r)"
+WORKDIR="${HOME}/apple1905-cdcncm"
+SRCVER="6.14.0-1015.15"
+SRCNAME="linux-nvidia-6.14"
+BASE_URL="https://launchpad.net/ubuntu/+archive/primary/+sourcefiles/${SRCNAME}/${SRCVER}"
+
+ORIG="${SRCNAME}_6.14.0.orig.tar.gz"
+DIFF="${SRCNAME}_${SRCVER}.diff.gz"
+DSC="${SRCNAME}_${SRCVER}.dsc"
+
+KBUILD="/lib/modules/${KVER}/build"
+
+msg() { printf '\n==> %s\n' "$*"; }
+die() {
+  echo "ERROR: $*" >&2
+  exit 1
+}
+need_cmd() { command -v "$1" >/dev/null 2>&1 || die "Missing command: $1"; }
+
+[[ $EUID -eq 0 ]] || die "Run with sudo"
+[[ -d $KBUILD ]] || die "Missing kernel headers/build dir: $KBUILD"
+
+need_cmd curl
+need_cmd dpkg-source
+need_cmd make
+need_cmd python3
+need_cmd modprobe
+need_cmd insmod
+need_cmd modinfo
+need_cmd ip
+need_cmd dmesg
+need_cmd find
+
+msg "Installing build deps"
+export DEBIAN_FRONTEND=noninteractive
+apt-get update
+apt-get install -y --no-install-recommends \
+  build-essential \
+  linux-headers-"${KVER}" \
+  dpkg-dev \
+  ca-certificates \
+  curl \
+  kmod \
+  zstd \
+  python3
+
+msg "Preparing workspace"
+rm -rf "$WORKDIR"
+mkdir -p "$WORKDIR"
+cd "$WORKDIR"
+
+msg "Downloading exact source package for ${SRCNAME} ${SRCVER}"
+curl -fL "${BASE_URL}/${ORIG}" -o "${ORIG}"
+curl -fL "${BASE_URL}/${DIFF}" -o "${DIFF}"
+curl -fL "${BASE_URL}/${DSC}" -o "${DSC}"
+
+msg "Extracting source"
+dpkg-source -x "${DSC}"
+
+SRCDIR="$(find . -maxdepth 1 -mindepth 1 -type d -name "${SRCNAME}-*" | head -n1)"
+[[ -n ${SRCDIR} ]] || die "Could not find extracted source directory"
+cd "${SRCDIR}"
+
+[[ -f drivers/net/usb/cdc_ncm.c ]] || die "cdc_ncm.c not found in source tree"
+
+msg "Patching cdc_ncm.c"
+python3 <<'PY'
+from pathlib import Path
+import re
+
+p = Path("drivers/net/usb/cdc_ncm.c")
+s = p.read_text()
+
+if "0x05ac, 0x1905, 0" not in s:
+    block = (
+        '\t/* Apple Mac direct USB-C networking quirk */\n'
+        '\t{ USB_DEVICE_INTERFACE_NUMBER(0x05ac, 0x1905, 0),\n'
+        '\t  .driver_info = (unsigned long)&apple_private_interface_info,\n'
+        '\t},\n'
+        '\t{ USB_DEVICE_INTERFACE_NUMBER(0x05ac, 0x1905, 2),\n'
+        '\t  .driver_info = (unsigned long)&apple_private_interface_info,\n'
+        '\t},\n'
+    )
+
+    pat = re.compile(
+        r'(\{\s*USB_INTERFACE_INFO\s*\(\s*USB_CLASS_COMM\s*,\s*USB_CDC_SUBCLASS_NCM\s*,\s*USB_CDC_PROTO_NONE\s*\)\s*,\s*\.driver_info\s*=\s*\(unsigned long\)&cdc_ncm_info\s*,\s*\},)',
+        re.S,
+    )
+    s2, n = pat.subn(block + r'\1', s, count=1)
+    if n != 1:
+        raise SystemExit("Could not find generic CDC NCM class-match entry to patch")
+    s = s2
+
+# Patch old bind check if needed.
+# Old style:
+#   if (!dev->in || !dev->out || !dev->status)
+# New behavior for this Mac:
+#   allow missing status endpoint for 05ac:1905
+old = re.compile(
+    r'if\s*\(\s*!dev->in\s*\|\|\s*!dev->out\s*\|\|\s*!dev->status\s*\)',
+    re.S,
+)
+repl = (
+    'if (!dev->in || !dev->out || '
+    '(!dev->status && '
+    '!(le16_to_cpu(dev->udev->descriptor.idVendor) == 0x05ac && '
+    'le16_to_cpu(dev->udev->descriptor.idProduct) == 0x1905)))'
+)
+s, n = old.subn(repl, s, count=1)
+
+# If the tree already has the newer FLAG_LINK_INTR logic, leave it alone.
+if n == 0 and "FLAG_LINK_INTR" not in s:
+    raise SystemExit("Could not patch bind_common() missing-status check")
+
+# Add a loud marker so we know the patched module actually loaded.
+if 'Apple 05ac:1905 quirk test module loaded' not in s:
+    m = re.search(r'static\s+int\s+__init\s+cdc_ncm_init\s*\(\s*void\s*\)\s*\{', s)
+    if m:
+        insert_at = m.end()
+        s = s[:insert_at] + '\n\tpr_info("cdc_ncm: Apple 05ac:1905 quirk test module loaded\\n");' + s[insert_at:]
+
+p.write_text(s)
+print("Patched", p)
+PY
+
+msg "Preparing out-of-tree build dir"
+mkdir -p buildmod
+cp drivers/net/usb/cdc_ncm.c buildmod/
+
+cat >buildmod/Makefile <<'EOF'
+obj-m += cdc_ncm.o
+ccflags-y += -Wno-error
+
+all:
+	$(MAKE) -C /lib/modules/$(shell uname -r)/build M=$(PWD) modules
+
+clean:
+	$(MAKE) -C /lib/modules/$(shell uname -r)/build M=$(PWD) clean
+EOF
+
+msg "Building patched module"
+cd buildmod
+make -C "$KBUILD" M="$PWD" modules
+PATCHED_KO="$PWD/cdc_ncm.ko"
+[[ -f $PATCHED_KO ]] || die "Patched cdc_ncm.ko was not built"
+
+msg "Patched module info"
+modinfo "$PATCHED_KO" | sed -n '1,20p'
+
+msg "Reloading module stack"
+modprobe -r cdc_mbim 2>/dev/null || true
+modprobe -r cdc_ncm 2>/dev/null || true
+insmod "$PATCHED_KO"
+
+msg "Recent dmesg"
+dmesg | tail -n 60
+
+cat <<EOF
+
+Done.
+
+Now:
+  1. run: sudo dmesg -w
+  2. unplug and replug the USB-C cable to the Mac
+  3. then run:
+       ip -br link
+       lsusb | grep 05ac:1905
+
+You want to see:
+  - cdc_ncm: Apple 05ac:1905 quirk test module loaded
+  - no bind() failure for 05ac:1905
+  - a new interface appearing
+
+Workspace:
+  $WORKDIR
+
+EOF
@@ -0,0 +1,155 @@
+#!/usr/bin/env bash
+set -Eeuo pipefail
+
+PATCHED_CDC_NCM="/root/apple1905-cdcncm/linux-nvidia-6.14-6.14.0/buildmod/cdc_ncm.ko"
+
+need() {
+  command -v "$1" >/dev/null 2>&1 || {
+    echo "Missing command: $1" >&2
+    exit 1
+  }
+}
+
+msg() {
+  printf '\n==> %s\n' "$*"
+}
+
+[[ $EUID -eq 0 ]] || {
+  echo "Run with sudo." >&2
+  exit 1
+}
+
+need modprobe
+need insmod
+need nmcli
+need lsusb
+need ip
+need awk
+need grep
+
+[[ -f $PATCHED_CDC_NCM ]] || {
+  echo "Patched module not found: $PATCHED_CDC_NCM" >&2
+  exit 1
+}
+
+msg "Removing old hard-block config if present"
+rm -f /etc/modprobe.d/zz-kill-mac-usb-net.conf
+depmod -a
+
+msg "Loading USB4 / Type-C / Thunderbolt support"
+modprobe typec || true
+modprobe typec_ucsi || true
+modprobe ucsi_acpi || true
+modprobe thunderbolt || true
+modprobe typec_thunderbolt || true
+modprobe thunderbolt_net || true
+
+msg "Unloading stock USB network stack"
+modprobe -r cdc_mbim cdc_wdm cdc_ncm cdc_ether usbnet 2>/dev/null || true
+
+msg "Loading dependency modules"
+modprobe usbnet
+modprobe cdc_ether
+
+msg "Loading patched cdc_ncm"
+if lsmod | grep -q '^cdc_ncm '; then
+  echo "cdc_ncm already loaded"
+else
+  insmod "$PATCHED_CDC_NCM"
+fi
+
+msg "Current module state"
+lsmod | grep -E 'typec|ucsi|thunderbolt|cdc_ncm|cdc_ether|usbnet' || true
+
+cat <<'EOF'
+
+Now do this on the Mac:
+  1. Run the Mac script below
+  2. Replug the cable
+  3. Wait for the Mac to land on the fast bus
+
+Then run this same script again with:
+  sudo ./spark-usbc-setup.sh finalize
+
+EOF
+
+if [[ ${1:-} != "finalize" ]]; then
+  exit 0
+fi
+
+msg "Waiting for Apple Mac on fast bus (Bus 004 @ 10000M or equivalent)"
+for _ in $(seq 1 60); do
+  if lsusb -t | grep -q '05ac:1905\|Driver=\[none\], 10000M'; then
+    break
+  fi
+  sleep 1
+done
+
+msg "lsusb -t snapshot"
+lsusb -t
+echo
+lsusb | grep '05ac:1905' || true
+
+msg "Waiting for new Apple USB NICs"
+for _ in $(seq 1 30); do
+  mapfile -t APPLE_IFS < <(
+    ip -o link | awk -F': ' '{print $2}' |
+      grep '^enx36be1bab12' || true
+  )
+  if [[ ${#APPLE_IFS[@]} -ge 2 ]]; then
+    break
+  fi
+  sleep 1
+done
+
+if [[ ${#APPLE_IFS[@]} -lt 2 ]]; then
+  echo "Did not find two Apple USB interfaces." >&2
+  ip -br link
+  exit 1
+fi
+
+IF0="${APPLE_IFS[0]}"
+IF1="${APPLE_IFS[1]}"
+
+msg "Found interfaces: $IF0 and $IF1"
+
+msg "Resetting old NetworkManager profiles"
+nmcli connection delete mac-usb-dhcp-0 2>/dev/null || true
+nmcli connection delete mac-usb-dhcp-1 2>/dev/null || true
+
+msg "Bringing up shared DHCP on both interfaces"
+nmcli connection add type ethernet ifname "$IF0" con-name mac-usb-dhcp-0
+nmcli connection modify mac-usb-dhcp-0 \
+  ipv4.method shared \
+  ipv6.method disabled \
+  ipv4.addresses 10.42.0.1/24
+nmcli connection up mac-usb-dhcp-0
+
+nmcli connection add type ethernet ifname "$IF1" con-name mac-usb-dhcp-1
+nmcli connection modify mac-usb-dhcp-1 \
+  ipv4.method shared \
+  ipv6.method disabled \
+  ipv4.addresses 10.43.0.1/24
+nmcli connection up mac-usb-dhcp-1
+
+msg "Current addresses"
+ip -br addr show dev "$IF0"
+ip -br addr show dev "$IF1"
+
+msg "Recent NetworkManager log"
+journalctl -u NetworkManager -n 60 --no-pager || true
+
+cat <<EOF
+
+Done.
+
+DGX shared-DHCP interfaces:
+  $IF0 -> 10.42.0.1/24
+  $IF1 -> 10.43.0.1/24
+
+Next on the Mac:
+  sudo ipconfig set anpi0 DHCP
+  sudo ipconfig set anpi1 DHCP
+  sudo ipconfig set anpi2 DHCP
+
+EOF
Author	SHA1	Message	Date
Ryuichi Leo Takashige	ce67529524	vibe coding gguf	2026-03-10 18:32:48 +00:00
Ryuichi Leo Takashige	c3f3d0eeb9	Vibe coding design baby	2026-03-10 17:21:28 +00:00
Ryuichi Leo Takashige	7199e556ef	Some Linux Laptop/Desktop detection and goodbye penguin	2026-03-10 17:03:56 +00:00
Ryuichi Leo Takashige	f376c24895	Fix placement preview	2026-03-10 16:55:28 +00:00
Ryuichi Leo Takashige	d63277edbd	Show Sparks and Linux in topology	2026-03-10 16:08:03 +00:00
Ryuichi Leo Takashige	dd121ff966	Fast direct USB connectivity	2026-03-10 16:08:03 +00:00