fix: adapt to breaking llama.cpp changes

Author: giladgd

llama/addon/addon.cpp

@@ -215,6 +215,7 @@ Napi::Object registerCallback(Napi::Env env, Napi::Object exports) {
     AddonContext::init(exports);
     AddonSampler::init(exports);
 
+    ggml_backend_load_all();
     llama_log_set(addonLlamaCppLogCallback, nullptr);
 
     exports.AddFinalizer(addonFreeLlamaBackend, static_cast<int*>(nullptr));

llama/addon/globals/getGpuInfo.cpp

@@ -103,5 +103,14 @@ Napi::Value getGpuType(const Napi::CallbackInfo& info) {
         }
     }
 
+    for (size_t i = 0; i < ggml_backend_dev_count(); i++) {
+        ggml_backend_dev_t device = ggml_backend_dev_get(i);
+        const auto deviceName = std::string(ggml_backend_dev_name(device));
+        
+        if (deviceName == "CPU") {
+            return Napi::Boolean::New(info.Env(), false);
+        }
+    }
+
     return info.Env().Undefined();
 }

src/bindings/AddonTypes.ts

@@ -69,7 +69,7 @@ export type BindingModule = {
     getGpuDeviceInfo(): {
         deviceNames: string[]
     },
-    getGpuType(): "cuda" | "vulkan" | "metal" | undefined,
+    getGpuType(): "cuda" | "vulkan" | "metal" | false | undefined,
     getSwapInfo(): {
         total: number,
         maxSize: number,

src/bindings/getLlama.ts

@@ -586,7 +586,7 @@ async function loadExistingLlamaBinary({
                 buildMetadata
             });
             const binaryCompatible = shouldTestBinaryBeforeLoading
-                ? await testBindingBinary(localBuildBinPath)
+                ? await testBindingBinary(localBuildBinPath, buildOptions.gpu)
                 : true;
 
             if (binaryCompatible) {
@@ -643,7 +643,7 @@ async function loadExistingLlamaBinary({
                     buildMetadata
                 });
                 const binaryCompatible = shouldTestBinaryBeforeLoading
-                    ? await testBindingBinary(prebuiltBinDetails.binaryPath)
+                    ? await testBindingBinary(prebuiltBinDetails.binaryPath, buildOptions.gpu)
                     : true;
 
                 if (binaryCompatible) {

src/bindings/utils/testBindingBinary.ts

@@ -4,14 +4,15 @@ import {createRequire} from "module";
 import path from "path";
 import {getConsoleLogPrefix} from "../../utils/getConsoleLogPrefix.js";
 import {runningInElectron} from "../../utils/runtime.js";
+import {BuildGpu} from "../types.js";
 import type {BindingModule} from "../AddonTypes.js";
 
 const require = createRequire(import.meta.url);
 const __filename = fileURLToPath(import.meta.url);
 const detectedFileName = path.basename(__filename);
 const expectedFileName = "testBindingBinary";
 
-export async function testBindingBinary(bindingBinaryPath: string, testTimeout: number = 1000 * 60 * 5): Promise<boolean> {
+export async function testBindingBinary(bindingBinaryPath: string, gpu: BuildGpu, testTimeout: number = 1000 * 60 * 5): Promise<boolean> {
     if (!detectedFileName.startsWith(expectedFileName)) {
         console.warn(
             getConsoleLogPrefix() +
@@ -163,10 +164,14 @@ export async function testBindingBinary(bindingBinaryPath: string, testTimeout:
                 onMessage(message: ChildToParentMessage) {
                     if (message.type === "ready") {
                         forkSucceeded = true;
-                        subProcess!.sendMessage({type: "start", bindingBinaryPath} satisfies ParentToChildMessage);
+                        subProcess!.sendMessage({
+                            type: "start",
+                            bindingBinaryPath,
+                            gpu
+                        });
                     } else if (message.type === "done") {
                         testPassed = true;
-                        subProcess!.sendMessage({type: "exit"} satisfies ParentToChildMessage);
+                        subProcess!.sendMessage({type: "exit"});
                     }
                 },
                 onExit(code: number) {
@@ -194,6 +199,12 @@ if (process.env.TEST_BINDING_CP === "true" && (process.parentPort != null || pro
                 await binding.init();
                 binding.getGpuVramInfo();
                 binding.getGpuDeviceInfo();
+
+                const gpuType = binding.getGpuType();
+                void (gpuType as BuildGpu satisfies typeof gpuType);
+                if (gpuType !== message.gpu)
+                    throw new Error(`GPU type mismatch. Expected: ${message.gpu}, got: ${gpuType}`);
+
                 sendMessage({type: "done"});
             } catch (err) {
                 console.error(err);
@@ -214,7 +225,8 @@ if (process.env.TEST_BINDING_CP === "true" && (process.parentPort != null || pro
 
 type ParentToChildMessage = {
     type: "start",
-    bindingBinaryPath: string
+    bindingBinaryPath: string,
+    gpu: BuildGpu
 } | {
     type: "exit"
 };