node-llama-cpp 3.0.0-beta.9 → 3.0.1

package/dist/bindings/getLlama.d.ts

@@ -1,20 +1,25 @@
-import { LlamaLogLevel } from "./types.js";
+import { LlamaGpuType, LlamaLogLevel } from "./types.js";
 import { Llama } from "./Llama.js";
 export type LlamaOptions = {
     /**
-     * Toggle Metal support in llama.cpp.
-     * Only supported on macOS.
-     * Enabled by default on Apple Silicon Macs.
-     */
-    metal?: boolean;
-    /**
-     * Toggle CUDA support on llama.cpp.
-     * Disabled by default.
+     * The compute layer implementation type to use for llama.cpp.
+     * - **`"auto"`**: Automatically detect and use the best GPU available (Metal on macOS, and CUDA or Vulkan on Windows and Linux)
+     * - **`"metal"`**: Use Metal.
+     *   Only supported on macOS.
+     *   Enabled by default on Apple Silicon Macs.
+     * - **`"cuda"`**: Use CUDA.
+     * - **`"vulkan"`**: Use Vulkan.
+     * - **`false`**: Disable any GPU support and only use the CPU.
+     *
+     * `"auto"` by default.
      */
-    cuda?: boolean;
+    gpu?: "auto" | LlamaGpuType | {
+        type: "auto";
+        exclude?: LlamaGpuType[];
+    };
     /**
      * Set the minimum log level for llama.cpp.
-     * Defaults to "debug".
+     * Defaults to `"warn"`.
      */
     logLevel?: LlamaLogLevel;
     /**
@@ -32,7 +37,11 @@ export type LlamaOptions = {
      * - **`"forceRebuild"`**: Always build from source.
      * Be cautious with this option, as it will cause the build to fail on Windows when the binaries are in use by another process.
      *
-     * Defaults to "auto".
+     * When running from inside an Asar archive in Electron, building from source is not possible, so it'll never build from source.
+     * To allow building from source in Electron apps, make sure you ship `node-llama-cpp` as an unpacked module.
+     *
+     * Defaults to `"auto"`.
+     * On Electron, defaults to `"never"`.
      */
     build?: "auto" | "never" | "forceRebuild";
     /**
@@ -60,24 +69,120 @@ export type LlamaOptions = {
      * Disabled by default.
      */
     skipDownload?: boolean;
+    /**
+     * The maximum number of threads to use for the Llama instance.
+     *
+     * Set to `0` to have no thread limit.
+     *
+     * When not using a GPU, defaults to the number of CPU cores that are useful for math (`.cpuMathCores`), or `4`, whichever is higher.
+     *
+     * When using a GPU, there's no limit by default.
+     */
+    maxThreads?: number;
+    /**
+     * Pad the available VRAM for the memory size calculations, as these calculations are not always accurate.
+     * Recommended to ensure stability.
+     * This only affects the calculations of `"auto"` in function options and is not reflected in the `getVramState` function.
+     *
+     * Defaults to `6%` of the total VRAM or 1GB, whichever is lower.
+     * Set to `0` to disable.
+     */
+    vramPadding?: number | ((totalVram: number) => number);
+    /**
+     * Enable debug mode to find issues with llama.cpp.
+     * Makes logs print directly to the console from `llama.cpp` and not through the provided logger.
+     *
+     * Defaults to `false`.
+     *
+     * The default can be set using the `NODE_LLAMA_CPP_DEBUG` environment variable.
+     */
+    debug?: boolean;
 };
 export type LastBuildOptions = {
     /**
      * Set the minimum log level for llama.cpp.
-     * Defaults to "debug".
+     * Defaults to "warn".
      */
     logLevel?: LlamaLogLevel;
     /**
      * Set a custom logger for llama.cpp logs.
      */
     logger?: (level: LlamaLogLevel, message: string) => void;
+    /**
+     * If a local build is not found, use prebuilt binaries.
+     * Enabled by default.
+     */
+    usePrebuiltBinaries?: boolean;
+    /**
+     * If a local build is not found, and prebuilt binaries are not found, when building from source,
+     * print binary compilation progress logs.
+     * Enabled by default.
+     */
+    progressLogs?: boolean;
+    /**
+     * If a local build is not found, and prebuilt binaries are not found, don't download llama.cpp source if it's not found.
+     * When set to `true`, and llama.cpp source is needed but is not found, a `NoBinaryFoundError` error will be thrown.
+     * Disabled by default.
+     */
+    skipDownload?: boolean;
+    /**
+     * The maximum number of threads to use for the Llama instance.
+     *
+     * Set to `0` to have no thread limit.
+     *
+     * When not using a GPU, defaults to the number of CPU cores that are useful for math (`.cpuMathCores`), or `4`, whichever is higher.
+     *
+     * When using a GPU, there's no limit by default.
+     */
+    maxThreads?: number;
+    /**
+     * Pad the available VRAM for the memory size calculations, as these calculations are not always accurate.
+     * Recommended to ensure stability.
+     * This only affects the calculations of `"auto"` in function options and is not reflected in the `getVramState` function.
+     *
+     * Defaults to `6%` of the total VRAM or 1GB, whichever is lower.
+     * Set to `0` to disable.
+     */
+    vramPadding?: number | ((totalVram: number) => number);
+    /**
+     * Enable debug mode to find issues with llama.cpp.
+     * Makes logs print directly to the console from `llama.cpp` and not through the provided logger.
+     *
+     * Defaults to `false`.
+     *
+     * The default can be set using the `NODE_LLAMA_CPP_DEBUG` environment variable.
+     */
+    debug?: boolean;
 };
 export declare const getLlamaFunctionName = "getLlama";
+export declare const defaultLlamaVramPadding: (totalVram: number) => number;
 /**
- * Get a llama.cpp binding.
- * Defaults to prefer a prebuilt binary, and fallbacks to building from source if a prebuilt binary is not found.
- * Pass `"lastCliBuild"` to default to use the last successful build created using the `download` or `build` CLI commands if one exists.
+ * Get a `llama.cpp` binding.
+ *
+ * Defaults to use a local binary built using the `source download` or `source build` CLI commands if one exists,
+ * otherwise, uses a prebuilt binary, and fallbacks to building from source if a prebuilt binary is not found.
+ *
+ * Pass `"lastBuild"` to default to use the last successful build created
+ * using the `source download` or `source build` CLI commands if one exists.
+ *
+ * The difference between using `"lastBuild"` and not using it is that `"lastBuild"` will use the binary built using a CLI command
+ * with the configuration used to build that binary (like using its GPU type),
+ * while not using `"lastBuild"` will only attempt to only use a binary that complies with the given options.
+ *
+ * For example, if your machine supports both CUDA and Vulkan, and you run the `source download --gpu vulkan` command,
+ * calling `getLlama("lastBuild")` will return the binary you built with Vulkan,
+ * while calling `getLlama()` will return a binding from a pre-built binary with CUDA,
+ * since CUDA is preferable on systems that support it.
+ *
+ * For example, if your machine supports CUDA, and you run the `source download --gpu cuda` command,
+ * calling `getLlama("lastBuild")` will return the binary you built with CUDA,
+ * and calling `getLlama()` will also return that same binary you built with CUDA.
+ *
+ * You should prefer to use `getLlama()` without `"lastBuild"` unless you have a specific reason to use the last build.
  */
-export declare function getLlama(): Promise<Llama>;
-export declare function getLlama(options: LlamaOptions, lastBuildOptions?: never): Promise<Llama>;
+export declare function getLlama(options?: LlamaOptions): Promise<Llama>;
 export declare function getLlama(type: "lastBuild", lastBuildOptions?: LastBuildOptions): Promise<Llama>;
+export declare function getLlamaForOptions({ gpu, logLevel, logger, build, cmakeOptions, existingPrebuiltBinaryMustMatchBuildOptions, usePrebuiltBinaries, progressLogs, skipDownload, maxThreads, vramPadding, debug }: LlamaOptions, { updateLastBuildInfoOnCompile, skipLlamaInit }?: {
+    updateLastBuildInfoOnCompile?: boolean;
+    skipLlamaInit?: boolean;
+}): Promise<Llama>;

package/dist/bindings/getLlama.js

@@ -2,10 +2,11 @@ import process from "process";
 import path from "path";
 import console from "console";
 import { createRequire } from "module";
-import { builtinLlamaCppGitHubRepo, builtinLlamaCppRelease, defaultLlamaCppCudaSupport, defaultLlamaCppDebugLogs, defaultLlamaCppGitHubRepo, defaultLlamaCppMetalSupport, defaultLlamaCppRelease, defaultSkipDownload, llamaLocalBuildBinsDirectory } from "../config.js";
+import { builtinLlamaCppGitHubRepo, builtinLlamaCppRelease, defaultLlamaCppLogLevel, defaultLlamaCppGitHubRepo, defaultLlamaCppGpuSupport, defaultLlamaCppRelease, defaultSkipDownload, llamaLocalBuildBinsDirectory, recommendedBaseDockerImage, defaultLlamaCppDebugMode } from "../config.js";
 import { getConsoleLogPrefix } from "../utils/getConsoleLogPrefix.js";
 import { waitForLockfileRelease } from "../utils/waitForLockfileRelease.js";
 import { isGithubReleaseNeedsResolving, resolveGithubRelease } from "../utils/resolveGithubRelease.js";
+import { runningInsideAsar, runningInElectron } from "../utils/runtime.js";
 import { compileLlamaCpp, getLocalBuildBinaryBuildMetadata, getLocalBuildBinaryPath, getPrebuiltBinaryBuildMetadata, getPrebuiltBinaryPath } from "./utils/compileLLamaCpp.js";
 import { getLastBuildInfo } from "./utils/lastBuildInfo.js";
 import { getClonedLlamaCppRepoReleaseInfo, isLlamaCppRepoCloned } from "./utils/cloneLlamaCppRepo.js";
@@ -15,14 +16,31 @@ import { resolveCustomCmakeOptions } from "./utils/resolveCustomCmakeOptions.js"
 import { getCanUsePrebuiltBinaries } from "./utils/getCanUsePrebuiltBinaries.js";
 import { NoBinaryFoundError } from "./utils/NoBinaryFoundError.js";
 import { Llama } from "./Llama.js";
+import { getGpuTypesToUseForOption } from "./utils/getGpuTypesToUseForOption.js";
+import { getPrettyBuildGpuName } from "./consts.js";
+import { detectGlibc } from "./utils/detectGlibc.js";
+import { getLinuxDistroInfo, isDistroAlpineLinux } from "./utils/getLinuxDistroInfo.js";
+import { testBindingBinary } from "./utils/testBindingBinary.js";
+import { getPlatformInfo } from "./utils/getPlatformInfo.js";
+import { hasBuildingFromSourceDependenciesInstalled } from "./utils/hasBuildingFromSourceDependenciesInstalled.js";
 const require = createRequire(import.meta.url);
 export const getLlamaFunctionName = "getLlama";
+export const defaultLlamaVramPadding = (totalVram) => Math.floor(Math.min(totalVram * 0.06, 1024 * 1024 * 1024));
+const defaultBuildOption = runningInElectron
+    ? "never"
+    : "auto";
 export async function getLlama(options, lastBuildOptions) {
     if (options === "lastBuild") {
         const lastBuildInfo = await getLastBuildInfo();
         const getLlamaOptions = {
-            logLevel: lastBuildOptions?.logLevel ?? defaultLlamaCppDebugLogs,
-            logger: lastBuildOptions?.logger ?? Llama.defaultConsoleLogger
+            logLevel: lastBuildOptions?.logLevel ?? defaultLlamaCppLogLevel,
+            logger: lastBuildOptions?.logger ?? Llama.defaultConsoleLogger,
+            usePrebuiltBinaries: lastBuildOptions?.usePrebuiltBinaries ?? true,
+            progressLogs: lastBuildOptions?.progressLogs ?? true,
+            skipDownload: lastBuildOptions?.skipDownload ?? defaultSkipDownload,
+            maxThreads: lastBuildOptions?.maxThreads,
+            vramPadding: lastBuildOptions?.vramPadding ?? defaultLlamaVramPadding,
+            debug: lastBuildOptions?.debug ?? defaultLlamaCppDebugMode
         };
         if (lastBuildInfo == null)
             return getLlamaForOptions(getLlamaOptions);
@@ -38,7 +56,10 @@ export async function getLlama(options, lastBuildOptions) {
                     buildType: "localBuild",
                     buildMetadata,
                     logger: lastBuildOptions?.logger ?? Llama.defaultConsoleLogger,
-                    logLevel: lastBuildOptions?.logLevel ?? defaultLlamaCppDebugLogs
+                    logLevel: lastBuildOptions?.logLevel ?? defaultLlamaCppLogLevel,
+                    maxThreads: lastBuildOptions?.maxThreads,
+                    vramPadding: lastBuildOptions?.vramPadding ?? defaultLlamaVramPadding,
+                    debug: lastBuildOptions?.debug ?? defaultLlamaCppDebugMode
                 });
             }
             catch (err) {
@@ -50,19 +71,15 @@ export async function getLlama(options, lastBuildOptions) {
     }
     return getLlamaForOptions(options ?? {});
 }
-async function getLlamaForOptions({ metal = defaultLlamaCppMetalSupport, cuda = defaultLlamaCppCudaSupport, logLevel = defaultLlamaCppDebugLogs, logger = Llama.defaultConsoleLogger, build = "auto", cmakeOptions = {}, existingPrebuiltBinaryMustMatchBuildOptions = false, usePrebuiltBinaries = true, progressLogs = true, skipDownload = defaultSkipDownload }) {
+export async function getLlamaForOptions({ gpu = defaultLlamaCppGpuSupport, logLevel = defaultLlamaCppLogLevel, logger = Llama.defaultConsoleLogger, build = defaultBuildOption, cmakeOptions = {}, existingPrebuiltBinaryMustMatchBuildOptions = false, usePrebuiltBinaries = true, progressLogs = true, skipDownload = defaultSkipDownload, maxThreads, vramPadding = defaultLlamaVramPadding, debug = defaultLlamaCppDebugMode }, { updateLastBuildInfoOnCompile = false, skipLlamaInit = false } = {}) {
     const platform = getPlatform();
     const arch = process.arch;
-    if (platform !== "mac")
-        metal = false;
-    else if (platform === "mac" && arch === "arm64")
-        cuda = false;
     if (logLevel == null)
-        logLevel = defaultLlamaCppDebugLogs;
+        logLevel = defaultLlamaCppLogLevel;
     if (logger == null)
         logger = Llama.defaultConsoleLogger;
     if (build == null)
-        build = "auto";
+        build = defaultBuildOption;
     if (cmakeOptions == null)
         cmakeOptions = {};
     if (existingPrebuiltBinaryMustMatchBuildOptions == null)
@@ -73,99 +90,240 @@ async function getLlamaForOptions({ metal = defaultLlamaCppMetalSupport, cuda =
         progressLogs = true;
     if (skipDownload == null)
         skipDownload = defaultSkipDownload;
+    if (vramPadding == null)
+        vramPadding = defaultLlamaVramPadding;
+    if (debug == null)
+        debug = defaultLlamaCppDebugMode;
     const clonedLlamaCppRepoReleaseInfo = await getClonedLlamaCppRepoReleaseInfo();
-    const buildOptions = {
-        customCmakeOptions: resolveCustomCmakeOptions(cmakeOptions),
-        progressLogs,
-        platform,
-        arch,
-        computeLayers: {
-            metal,
-            cuda
-        },
-        llamaCpp: {
-            repo: clonedLlamaCppRepoReleaseInfo?.llamaCppGithubRepo ?? builtinLlamaCppGitHubRepo,
-            release: clonedLlamaCppRepoReleaseInfo?.tag ?? builtinLlamaCppRelease
-        }
-    };
-    const canUsePrebuiltBinaries = (build === "forceRebuild" || !usePrebuiltBinaries)
+    let canUsePrebuiltBinaries = (build === "forceRebuild" || !usePrebuiltBinaries)
         ? false
         : await getCanUsePrebuiltBinaries();
-    let buildFolderName = await getBuildFolderNameForBuildOptions(buildOptions);
+    const buildGpusToTry = await getGpuTypesToUseForOption(gpu, { platform, arch });
+    const platformInfo = await getPlatformInfo();
+    const llamaCppInfo = {
+        repo: clonedLlamaCppRepoReleaseInfo?.llamaCppGithubRepo ?? builtinLlamaCppGitHubRepo,
+        release: clonedLlamaCppRepoReleaseInfo?.tag ?? builtinLlamaCppRelease
+    };
+    let shouldLogNoGlibcWarningIfNoBuildIsAvailable = false;
+    const canBuild = build !== "never" && !runningInsideAsar &&
+        (!runningInElectron || await hasBuildingFromSourceDependenciesInstalled());
+    if (canUsePrebuiltBinaries && platform === "linux") {
+        if (!(await detectGlibc({ platform }))) {
+            canUsePrebuiltBinaries = false;
+            shouldLogNoGlibcWarningIfNoBuildIsAvailable = true;
+        }
+    }
+    if (buildGpusToTry.length === 0)
+        throw new Error("No GPU types available to try building with");
     if (build === "auto" || build === "never") {
-        const localBuildFolder = path.join(llamaLocalBuildBinsDirectory, buildFolderName.withCustomCmakeOptions);
-        const localBuildBinPath = await getLocalBuildBinaryPath(buildFolderName.withCustomCmakeOptions);
-        await waitForLockfileRelease({ resourcePath: localBuildFolder });
-        if (localBuildBinPath != null) {
-            try {
+        for (let i = 0; i < buildGpusToTry.length; i++) {
+            const gpu = buildGpusToTry[i];
+            const isLastItem = i === buildGpusToTry.length - 1;
+            if (gpu == null)
+                continue;
+            const buildOptions = {
+                customCmakeOptions: resolveCustomCmakeOptions(cmakeOptions),
+                progressLogs,
+                platform,
+                platformInfo,
+                arch,
+                gpu,
+                llamaCpp: llamaCppInfo
+            };
+            const llama = await loadExistingLlamaBinary({
+                buildOptions,
+                canUsePrebuiltBinaries,
+                logLevel,
+                logger,
+                existingPrebuiltBinaryMustMatchBuildOptions,
+                progressLogs,
+                platform,
+                platformInfo,
+                skipLlamaInit,
+                maxThreads,
+                vramPadding,
+                fallbackMessage: !isLastItem
+                    ? `falling back to using ${getPrettyBuildGpuName(buildGpusToTry[i + 1])}`
+                    : (canBuild
+                        ? "falling back to building from source"
+                        : null),
+                debug
+            });
+            if (llama != null)
+                return llama;
+        }
+    }
+    if (shouldLogNoGlibcWarningIfNoBuildIsAvailable && progressLogs)
+        await logNoGlibcWarning();
+    if (!canBuild)
+        throw new NoBinaryFoundError();
+    const llamaCppRepoCloned = await isLlamaCppRepoCloned();
+    if (!llamaCppRepoCloned) {
+        if (skipDownload)
+            throw new NoBinaryFoundError("No prebuilt binaries found, no llama.cpp source found and `skipDownload` or NODE_LLAMA_CPP_SKIP_DOWNLOAD env var is set to true, so llama.cpp cannot be built from source");
+        llamaCppInfo.repo = defaultLlamaCppGitHubRepo;
+        llamaCppInfo.release = defaultLlamaCppRelease;
+        if (isGithubReleaseNeedsResolving(llamaCppInfo.release)) {
+            const [owner, name] = defaultLlamaCppGitHubRepo.split("/");
+            llamaCppInfo.release = await resolveGithubRelease(owner, name, llamaCppInfo.release);
+        }
+    }
+    for (let i = 0; i < buildGpusToTry.length; i++) {
+        const gpu = buildGpusToTry[i];
+        const isLastItem = i === buildGpusToTry.length - 1;
+        if (gpu == null)
+            continue;
+        const buildOptions = {
+            customCmakeOptions: resolveCustomCmakeOptions(cmakeOptions),
+            progressLogs,
+            platform,
+            platformInfo,
+            arch,
+            gpu,
+            llamaCpp: llamaCppInfo
+        };
+        try {
+            return await buildAndLoadLlamaBinary({
+                buildOptions,
+                skipDownload,
+                logLevel,
+                logger,
+                updateLastBuildInfoOnCompile,
+                maxThreads,
+                vramPadding,
+                skipLlamaInit,
+                debug
+            });
+        }
+        catch (err) {
+            console.error(getConsoleLogPrefix() +
+                `Failed to build llama.cpp with ${getPrettyBuildGpuName(gpu)} support. ` +
+                (!isLastItem
+                    ? `falling back to building llama.cpp with ${getPrettyBuildGpuName(buildGpusToTry[i + 1])} support. `
+                    : "") +
+                "Error:", err);
+            if (isLastItem)
+                throw err;
+        }
+    }
+    throw new Error("Failed to build llama.cpp");
+}
+async function loadExistingLlamaBinary({ buildOptions, canUsePrebuiltBinaries, logLevel, logger, existingPrebuiltBinaryMustMatchBuildOptions, progressLogs, platform, platformInfo, skipLlamaInit, maxThreads, vramPadding, fallbackMessage, debug }) {
+    const buildFolderName = await getBuildFolderNameForBuildOptions(buildOptions);
+    const localBuildFolder = path.join(llamaLocalBuildBinsDirectory, buildFolderName.withCustomCmakeOptions);
+    const localBuildBinPath = await getLocalBuildBinaryPath(buildFolderName.withCustomCmakeOptions);
+    await waitForLockfileRelease({ resourcePath: localBuildFolder });
+    if (localBuildBinPath != null) {
+        try {
+            const buildMetadata = await getLocalBuildBinaryBuildMetadata(buildFolderName.withCustomCmakeOptions);
+            const shouldTestBinaryBeforeLoading = getShouldTestBinaryBeforeLoading({
+                isPrebuiltBinary: false,
+                platform,
+                platformInfo,
+                buildMetadata
+            });
+            const binaryCompatible = shouldTestBinaryBeforeLoading
+                ? await testBindingBinary(localBuildBinPath)
+                : true;
+            if (binaryCompatible) {
                 const binding = loadBindingModule(localBuildBinPath);
-                const buildMetadata = await getLocalBuildBinaryBuildMetadata(buildFolderName.withCustomCmakeOptions);
                 return await Llama._create({
                     bindings: binding,
                     buildType: "localBuild",
                     buildMetadata,
                     logLevel,
-                    logger
+                    logger,
+                    maxThreads,
+                    vramPadding,
+                    skipLlamaInit,
+                    debug
                 });
             }
-            catch (err) {
-                const binaryDescription = describeBinary(buildOptions);
-                console.error(getConsoleLogPrefix() + `Failed to load a local build ${binaryDescription}. Error:`, err);
-                console.info(getConsoleLogPrefix() + "Falling back to prebuilt binaries");
+            else if (progressLogs) {
+                console.warn(getConsoleLogPrefix() + "The local build binary was not built in the current system and is incompatible with it");
+                if (canUsePrebuiltBinaries)
+                    console.info(getConsoleLogPrefix() + "Falling back to prebuilt binaries");
+                else if (fallbackMessage != null)
+                    console.info(getConsoleLogPrefix() + fallbackMessage);
             }
         }
-        if (canUsePrebuiltBinaries) {
-            const prebuiltBinPath = await getPrebuiltBinaryPath(existingPrebuiltBinaryMustMatchBuildOptions
-                ? buildFolderName.withCustomCmakeOptions
-                : buildFolderName.withoutCustomCmakeOptions);
-            if (prebuiltBinPath != null) {
-                try {
-                    const binding = loadBindingModule(prebuiltBinPath);
-                    const buildMetadata = await getPrebuiltBinaryBuildMetadata(buildFolderName.withCustomCmakeOptions);
+        catch (err) {
+            const binaryDescription = describeBinary(buildOptions);
+            console.error(getConsoleLogPrefix() + `Failed to load a local build ${binaryDescription}. Error:`, err);
+            if (canUsePrebuiltBinaries)
+                console.info(getConsoleLogPrefix() + "Falling back to prebuilt binaries");
+            else if (fallbackMessage != null)
+                console.info(getConsoleLogPrefix() + fallbackMessage);
+        }
+    }
+    if (canUsePrebuiltBinaries) {
+        const prebuiltBinDetails = await getPrebuiltBinaryPath(buildOptions, existingPrebuiltBinaryMustMatchBuildOptions
+            ? buildFolderName.withCustomCmakeOptions
+            : buildFolderName.withoutCustomCmakeOptions);
+        if (prebuiltBinDetails != null) {
+            try {
+                const buildMetadata = await getPrebuiltBinaryBuildMetadata(prebuiltBinDetails.folderPath, prebuiltBinDetails.folderName);
+                const shouldTestBinaryBeforeLoading = getShouldTestBinaryBeforeLoading({
+                    isPrebuiltBinary: true,
+                    platform,
+                    platformInfo,
+                    buildMetadata
+                });
+                const binaryCompatible = shouldTestBinaryBeforeLoading
+                    ? await testBindingBinary(prebuiltBinDetails.binaryPath)
+                    : true;
+                if (binaryCompatible) {
+                    const binding = loadBindingModule(prebuiltBinDetails.binaryPath);
                     return await Llama._create({
                         bindings: binding,
                         buildType: "prebuilt",
                         buildMetadata,
                         logLevel,
-                        logger
+                        logger,
+                        maxThreads,
+                        vramPadding,
+                        skipLlamaInit,
+                        debug
                     });
                 }
-                catch (err) {
+                else if (progressLogs) {
                     const binaryDescription = describeBinary({
                         ...buildOptions,
                         customCmakeOptions: existingPrebuiltBinaryMustMatchBuildOptions
                             ? buildOptions.customCmakeOptions
                             : new Map()
                     });
-                    console.error(getConsoleLogPrefix() + `Failed to load a prebuilt ${binaryDescription}` + (build !== "never"
-                        ? ", falling back to building from source"
-                        : "") + ". Error:", err);
+                    console.warn(getConsoleLogPrefix() +
+                        `The prebuilt ${binaryDescription} is not compatible with the current system` + (fallbackMessage != null
+                        ? ", " + fallbackMessage
+                        : ""));
                 }
             }
-            else if (progressLogs)
-                console.warn(getConsoleLogPrefix() + "A prebuilt binary was not found" + (build !== "never"
-                    ? ", falling back to building from source"
-                    : ""));
-        }
-    }
-    if (build === "never")
-        throw new NoBinaryFoundError();
-    const llamaCppRepoCloned = await isLlamaCppRepoCloned();
-    if (!llamaCppRepoCloned) {
-        if (skipDownload)
-            throw new NoBinaryFoundError("No prebuilt binaries found, no llama.cpp source found and `skipDownload` or NODE_LLAMA_CPP_SKIP_DOWNLOAD env var is set to true, so llama.cpp cannot be built from source");
-        buildOptions.llamaCpp.repo = defaultLlamaCppGitHubRepo;
-        buildOptions.llamaCpp.release = defaultLlamaCppRelease;
-        if (isGithubReleaseNeedsResolving(buildOptions.llamaCpp.release)) {
-            const [owner, name] = defaultLlamaCppGitHubRepo.split("/");
-            buildOptions.llamaCpp.release = await resolveGithubRelease(owner, name, buildOptions.llamaCpp.release);
-            buildFolderName = await getBuildFolderNameForBuildOptions(buildOptions);
+            catch (err) {
+                const binaryDescription = describeBinary({
+                    ...buildOptions,
+                    customCmakeOptions: existingPrebuiltBinaryMustMatchBuildOptions
+                        ? buildOptions.customCmakeOptions
+                        : new Map()
+                });
+                console.error(getConsoleLogPrefix() + `Failed to load a prebuilt ${binaryDescription}` + (fallbackMessage != null
+                    ? ", " + fallbackMessage
+                    : "") + ". Error:", err);
+            }
         }
+        else if (progressLogs)
+            console.warn(getConsoleLogPrefix() + "A prebuilt binary was not found" + (fallbackMessage != null
+                ? ", " + fallbackMessage
+                : ""));
     }
+    return null;
+}
+async function buildAndLoadLlamaBinary({ buildOptions, skipDownload, logLevel, logger, updateLastBuildInfoOnCompile, maxThreads, vramPadding, skipLlamaInit, debug }) {
+    const buildFolderName = await getBuildFolderNameForBuildOptions(buildOptions);
     await compileLlamaCpp(buildOptions, {
         ensureLlamaCppRepoIsCloned: !skipDownload,
         downloadCmakeIfNeeded: true,
-        updateLastBuildInfo: false
+        updateLastBuildInfo: updateLastBuildInfoOnCompile
     });
     const localBuildFolder = path.join(llamaLocalBuildBinsDirectory, buildFolderName.withCustomCmakeOptions);
     await waitForLockfileRelease({ resourcePath: localBuildFolder });
@@ -180,16 +338,33 @@ async function getLlamaForOptions({ metal = defaultLlamaCppMetalSupport, cuda =
         buildType: "localBuild",
         buildMetadata,
         logLevel,
-        logger
+        logger,
+        maxThreads,
+        vramPadding,
+        skipLlamaInit,
+        debug
     });
 }
+async function logNoGlibcWarning() {
+    console.warn(getConsoleLogPrefix() +
+        "The prebuilt binaries cannot be used in this Linux distro, as `glibc` is not detected");
+    const linuxDistroInfo = await getLinuxDistroInfo();
+    const isAlpineLinux = await isDistroAlpineLinux(linuxDistroInfo);
+    if (isAlpineLinux) {
+        console.warn(getConsoleLogPrefix() +
+            "Using Alpine Linux is not recommended for running LLMs, " +
+            "as using GPU drivers is complicated and suboptimal in this distro at the moment.\n" +
+            getConsoleLogPrefix() +
+            "Consider using a different Linux distro, such as Debian or Ubuntu.\n" +
+            getConsoleLogPrefix() +
+            `If you're trying to run this inside of a Docker container, consider using "${recommendedBaseDockerImage}" image`);
+    }
+}
 function describeBinary(binaryOptions) {
     let res = `binary for platform "${binaryOptions.platform}" "${binaryOptions.arch}"`;
     const additions = [];
-    if (binaryOptions.computeLayers.metal)
-        additions.push("with Metal support");
-    if (binaryOptions.computeLayers.cuda)
-        additions.push("with CUDA support");
+    if (binaryOptions.gpu != false)
+        additions.push(`with ${getPrettyBuildGpuName(binaryOptions.gpu)} support`);
     if (binaryOptions.customCmakeOptions.size > 0)
         additions.push("with custom build options");
     res += additions
@@ -222,4 +397,18 @@ function loadBindingModule(bindingModulePath) {
         catch (err) { }
     }
 }
+function getShouldTestBinaryBeforeLoading({ isPrebuiltBinary, platform, platformInfo, buildMetadata }) {
+    if (platform === "linux") {
+        if (isPrebuiltBinary)
+            return true;
+        if (platformInfo.name !== buildMetadata.buildOptions.platformInfo.name ||
+            platformInfo.version !== buildMetadata.buildOptions.platformInfo.version)
+            return true;
+    }
+    else if (platform === "win") {
+        if (buildMetadata.buildOptions.gpu !== false)
+            return true;
+    }
+    return false;
+}
 //# sourceMappingURL=getLlama.js.map