@shipworthy/ai-sdk-llama-cpp 0.2.4 → 0.2.5

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@shipworthy/ai-sdk-llama-cpp",
-  "version": "0.2.4",
+  "version": "0.2.5",
   "description": "A minimal llama.cpp provider for the Vercel AI SDK implementing LanguageModelV3 and EmbeddingModelV3",
   "type": "module",
   "main": "./dist/index.js",
@@ -16,11 +16,12 @@
     "dist",
     "native",
     "scripts",
-    "CMakeLists.txt"
+    "CMakeLists.txt",
+    "llama.cpp.bundle"
   ],
   "llamaCpp": {
-    "repo": "https://github.com/ggerganov/llama.cpp.git",
-    "commit": "06705fdcb3ef199d2c2c95a5e3cbb7cf9cc5256e"
+    "repo": "https://github.com/ggml-org/llama.cpp.git",
+    "tag": "b7624"
   },
   "scripts": {
     "postinstall": "node scripts/postinstall.cjs",

package/scripts/postinstall.cjs

@@ -1,74 +1,96 @@
+// @ts-check
 const { execSync } = require("child_process");
 const fs = require("fs");
 const path = require("path");
 
-// Step 1: Check platform is macOS
-// if (process.platform !== "darwin") {
-//   console.error("\n===========================================");
-//   console.error("ERROR: ai-sdk-llama-cpp only supports macOS");
-//   console.error("===========================================\n");
-//   console.error(`Detected platform: ${process.platform}`);
-//   console.error("This package requires macOS for native compilation.\n");
-//   process.exit(1);
-// }
-
-// Step 2: Check if git is available
-try {
-  execSync("git --version", { stdio: "ignore" });
-} catch (error) {
-  console.error("\n===========================================");
-  console.error("ERROR: git is required but not found");
-  console.error("===========================================\n");
-  console.error("Please install git to continue.");
-  console.error("  macOS: xcode-select --install");
-  console.error("  or: brew install git\n");
-  process.exit(1);
+if (process.platform !== "darwin") {
+  console.warn("===========================================\n");
+  console.warn(`Detected platform: ${process.platform}`);
+  console.warn("ai-sdk-llama-cpp requires macOS for native compilation.\n");
+  process.exit(0);
 }
 
-// Step 3: Check if llama.cpp already exists (for local development)
-const llamaCppPath = path.join(__dirname, "..", "llama.cpp");
-if (fs.existsSync(llamaCppPath)) {
-  console.log("llama.cpp directory already exists, skipping clone...");
-} else {
-  // Step 4: Read commit hash from package.json
+const ROOT_DIR = path.join(__dirname, "..")
+const LLAMA_CPP_DIR = path.join(ROOT_DIR, "llama.cpp")
+const LLAMA_CPP_BUNDLE = path.join(ROOT_DIR, "llama.cpp.bundle")
+
+function cloneLlamaCpp() {
+  if (fs.existsSync(LLAMA_CPP_DIR)) {
+    console.log("llama.cpp directory already exists, skipping clone");
+    return;
+  }
+
+  if (fs.existsSync(LLAMA_CPP_BUNDLE)) {
+    console.log("Cloning llama.cpp from llama.cpp.bundle");
+    execSync(`git clone ${LLAMA_CPP_BUNDLE} ${LLAMA_CPP_DIR}`, {
+      stdio: "inherit",
+    });
+    return;
+  }
+
+  createLlamaReleaseBundle()
+}
+
+function createLlamaReleaseBundle() {
   const packageJsonPath = path.join(__dirname, "..", "package.json");
   const packageJson = JSON.parse(fs.readFileSync(packageJsonPath, "utf8"));
-  const { repo, commit } = packageJson.llamaCpp;
+  const { repo, tag } = packageJson.llamaCpp;
 
-  if (!repo || !commit) {
+  if (!repo || !tag) {
     console.error(
-      "ERROR: llamaCpp.repo and llamaCpp.commit must be defined in package.json"
+      "ERROR: llamaCpp repo and tag must be defined in package.json"
     );
     process.exit(1);
   }
 
-  // Step 5: Clone llama.cpp at the specific commit
-  console.log(`Cloning llama.cpp at commit ${commit}...`);
-  try {
-    // Clone with depth 1 for faster download, then fetch the specific commit
-    execSync(`git clone --depth 1 ${repo} llama.cpp`, {
-      stdio: "inherit",
-      cwd: path.join(__dirname, ".."),
-    });
+  console.log(`Cloning llama.cpp from release ${tag}`);
 
-    // Fetch the specific commit and checkout
-    execSync(`git fetch --depth 1 origin ${commit}`, {
-      stdio: "inherit",
-      cwd: llamaCppPath,
-    });
-    execSync(`git checkout ${commit}`, {
-      stdio: "inherit",
-      cwd: llamaCppPath,
-    });
+  execSync(`git clone --depth 1 --branch ${tag} ${repo} ${LLAMA_CPP_DIR}`, {
+    stdio: "inherit",
+  });
 
-    console.log("llama.cpp cloned successfully.");
-  } catch (error) {
-    console.error("ERROR: Failed to clone llama.cpp");
-    console.error(error.message);
-    process.exit(1);
-  }
+  execSync(`git fetch --unshallow`, {
+    stdio: "inherit",
+    cwd: LLAMA_CPP_DIR,
+  });
+
+  // Squash all history into a single commit
+  const newSha = execSync(`git commit-tree HEAD^{tree} -m "Squashed"`, {
+    stdio: "pipe",
+    cwd: LLAMA_CPP_DIR,
+  });
+  execSync(`git reset --hard ${newSha}`, {
+    stdio: "inherit",
+    cwd: LLAMA_CPP_DIR,
+  });
+
+  // Delete all tags (reduces bundle size)
+  execSync(`git tag -l | xargs git tag -d`, {
+    stdio: "inherit",
+    cwd: LLAMA_CPP_DIR,
+  });
+
+  // Aggressive garbage collection (major size reduction)
+  execSync(`git gc --aggressive --prune=all`, {
+    stdio: "inherit",
+    cwd: LLAMA_CPP_DIR,
+  });
+
+  execSync(`git bundle create ../llama.cpp.bundle HEAD`, {
+    stdio: "inherit",
+    cwd: LLAMA_CPP_DIR,
+  });
+
+  console.log("llama.cpp cloned and bundled successfully.");
+}
+
+try {
+  cloneLlamaCpp()
+} catch (error) {
+  console.error("ERROR: Failed to clone llama.cpp");
+  console.error(error.message);
+  process.exit(1);
 }
 
-// Step 6: Build native bindings
-console.log("Building native llama.cpp bindings for macOS...");
+console.log("Building native llama.cpp bindings");
 execSync("npx cmake-js compile", { stdio: "inherit" });