Nix flake support

.github/workflows/update_release.yaml

@@ -0,0 +1,38 @@
+
+name: Update Flake assets on Release
+
+on:
+  release:
+    types: [published] # Triggers when a release is fully published
+
+permissions:
+  contents: write # Required to push commits back to the repo
+
+jobs:
+  update-sources:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+        with:
+          persist-credentials: true
+
+      - name: Run update script
+        run: |
+          git checkout ${{github.ref_name}}
+          chmod +x ./update.sh
+          ./update.sh
+
+      - name: Commit and push changes
+        run: |
+          git config --global user.name "github-actions[bot]"
+          git config --global user.email "41898282+github-actions[bot]@users.noreply"
+
+          # Only commit if sources.json actually changed
+          if ! git diff --exit-code .; then
+            git add .
+            git commit -m "chore: update nix flake assets for ${{ github.event.release.tag_name }}"
+            git push origin ${{github.ref_name}}
+          else
+            echo "No changes detected in sources.json"
+          fi

flake.nix

@@ -0,0 +1,72 @@
+{
+  description = "llama.cpp CUDA binaries from ai-dock (Patched for NixOS)";
+
+  inputs = {
+    nixpkgs.url = "github:NixOS/nixpkgs/nixos-unstable";
+    flake-utils.url = "github:numtide/flake-utils";
+  };
+
+  outputs =
+    {
+      self,
+      nixpkgs,
+      flake-utils,
+    }:
+    flake-utils.lib.eachDefaultSystem (
+      system:
+      let
+        pkgs = import nixpkgs {
+          inherit system;
+          config.allowUnfree = true; # Required for CUDA
+        };
+        release = builtins.fromJSON (builtins.readFile "${self}/latest_release.json");
+        sha256 = builtins.convertHash {
+          hash = release.hash;
+          toHashFormat = "sri";
+          hashAlgo = "sha256";
+        };
+      in
+      {
+        packages.default =
+          with release;
+          pkgs.stdenv.mkDerivation {
+            pname = "aidock_llama-cpp-cuda";
+            inherit version;
+            src = pkgs.fetchurl { inherit url sha256; };
+            nativeBuildInputs = with pkgs; [ autoPatchelfHook ];
+            buildInputs = with pkgs; [
+              stdenv.cc.cc.lib
+              cudaPackages.cudatoolkit
+              cudaPackages.nccl
+              linuxPackages.nvidia_x11
+            ];
+
+            appendRunpaths = [
+              "/run/opengl-driver/lib"
+              "${pkgs.linuxPackages.nvidia_x11}/lib"
+              "$out/lib"
+            ];
+
+            sourceRoot = ".";
+
+            installPhase = ''
+              mkdir -p $out/bin $out/lib
+              find . -name "*.so*" -exec cp -vP {} $out/lib/ \;
+              find . -type f -executable \
+                ! -name "*.so*" \
+                ! -name "*.txt" \
+                ! -name "*.md" \
+                -exec cp -v {} $out/bin/ \;
+              [ "$(ls -A $out/bin)" ] || { echo "Error: No binaries found!"; exit 1; }
+            '';
+            autoPatchelfIgnoreMissingDeps = false;
+          };
+        devShells.default = pkgs.mkShell {
+          buildInputs = [ self.packages.${system}.default ];
+          shellHook = ''
+            export LD_LIBRARY_PATH="/run/opengl-driver/lib:${pkgs.linuxPackages.nvidia_x11}/lib:$LD_LIBRARY_PATH"
+          '';
+        };
+      }
+    );
+}

latest_release.json

@@ -0,0 +1,5 @@
+{
+  "version": "b9190",
+  "hash": "sha256:988920b5fb99250861577bd38a5ee639ff051183ce22cab170001208ae8f481b",
+  "url": "https://github.com/ai-dock/llama.cpp-cuda/releases/download/b9190/llama.cpp-b9190-cuda-12.8.tar.gz"
+}

update.sh

@@ -0,0 +1,3 @@
+#!/usr/bin/env bash
+curl -s https://api.github.com/repos/ai-dock/llama.cpp-cuda/releases/latest | jq '{version:(.tag_name),hash:(.assets[0].digest),url:(.assets[0].browser_download_url)}' > latest_release.json
+nix flake update