Add Metal CI and point kernel-builder to ChipFlow fork

- Add GitHub Actions workflow for Metal kernel CI (macOS 15 + 26)
- Point flake.nix to ChipFlow/kernels metal-stack branch for
macOS 15 toolchain detection support

Co-developed-by: Claude Code v2.1.50 (claude-opus-4-6)

.github/workflows/test_metal.yaml

@@ -0,0 +1,39 @@
+name: "Build and test (Metal)"
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+  workflow_dispatch:
+
+jobs:
+  test:
+    name: Build and test (${{ matrix.os }})
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - os: macos-15-xlarge
+            xcode: "/Applications/Xcode_16.2.app"
+          - os: macos-26-xlarge
+            xcode: "/Applications/Xcode_26.0.app"
+    steps:
+      - name: "Select Xcode"
+        run: sudo xcrun xcode-select -s ${{ matrix.xcode }}
+      - name: "Install Metal Toolchain"
+        if: matrix.os == 'macos-26-xlarge'
+        run: xcodebuild -downloadComponent metalToolchain
+      - uses: actions/checkout@v6
+      - uses: cachix/install-nix-action@v31
+        with:
+          extra_nix_config: |
+            sandbox = relaxed
+      - uses: cachix/cachix-action@v16
+        with:
+          name: huggingface
+
+      - name: Build kernel
+        run: nix build .#redistributable.torch29-metal-aarch64-darwin -L
+      - name: Test kernel
+        run: nix develop .#test --command pytest tests/ -v

flake.lock

@@ -37,19 +37,21 @@
       "inputs": {
         "flake-compat": "flake-compat",
         "flake-utils": "flake-utils",
-        "nixpkgs": "nixpkgs"
+        "nixpkgs": "nixpkgs",
+        "rust-overlay": "rust-overlay"
       },
       "locked": {
-        "lastModified": 1769448133,
-        "narHash": "sha256-XOp8+8u7fmXn1f63mJ40dPj/OHPMKtL9o4q7y0CUZFU=",
-        "owner": "huggingface",
-        "repo": "kernel-builder",
-        "rev": "078351df6e0fddb4a1a41ba3ffb8b804f58c4c6a",
+        "lastModified": 1772650055,
+        "narHash": "sha256-6R8dJEPH+uHJyvr3nZPZ/xFwULzR4UCsLQGSjLRsxQE=",
+        "owner": "ChipFlow",
+        "repo": "kernels",
+        "rev": "f85b1d195c115acdb3f92c061a0dafcc0f9bfe79",
         "type": "github"
       },
       "original": {
-        "owner": "huggingface",
-        "repo": "kernel-builder",
+        "owner": "ChipFlow",
+        "ref": "metal-stack",
+        "repo": "kernels",
         "type": "github"
       }
     },
@@ -74,6 +76,27 @@
         "kernel-builder": "kernel-builder"
       }
     },
+    "rust-overlay": {
+      "inputs": {
+        "nixpkgs": [
+          "kernel-builder",
+          "nixpkgs"
+        ]
+      },
+      "locked": {
+        "lastModified": 1769050281,
+        "narHash": "sha256-1H8DN4UZgEUqPUA5ecHOufLZMscJ4IlcGaEftaPtpBY=",
+        "owner": "oxalica",
+        "repo": "rust-overlay",
+        "rev": "6deef0585c52d9e70f96b6121207e1496d4b0c49",
+        "type": "github"
+      },
+      "original": {
+        "owner": "oxalica",
+        "repo": "rust-overlay",
+        "type": "github"
+      }
+    },
     "systems": {
       "locked": {
         "lastModified": 1681028828,

flake.nix

@@ -2,7 +2,7 @@
   description = "Flake for rotary embedding kernel";
 
   inputs = {
-    kernel-builder.url = "github:huggingface/kernel-builder";
+    kernel-builder.url = "github:ChipFlow/kernels/metal-stack";
   };
 
   outputs =