node-llama-cpp 3.0.0-beta.1 → 3.0.0-beta.10

package/README.md

@@ -15,6 +15,8 @@
 
 </div>
 
+✨ New! [Try the beta of version `3.0.0`](https://github.com/withcatai/node-llama-cpp/pull/105) ✨ (included: function calling, automatic chat wrapper detection, and more)
+
 ## Features
 * Run a text generation model locally on your machine
 * Metal and CUDA support

package/dist/ChatWrapper.d.ts

@@ -0,0 +1,49 @@
+import { ChatHistoryItem, ChatModelFunctions, ChatModelResponse } from "./types.js";
+import { LlamaText } from "./utils/LlamaText.js";
+export type ChatWrapperSettings = {
+    readonly functions: {
+        readonly call: {
+            readonly optionalPrefixSpace: boolean;
+            readonly prefix: string;
+            readonly paramsPrefix: string;
+            readonly suffix: string;
+        };
+        readonly result: {
+            readonly prefix: string;
+            readonly suffix: string;
+        };
+    };
+};
+export declare abstract class ChatWrapper {
+    abstract readonly wrapperName: string;
+    readonly settings: ChatWrapperSettings;
+    generateContextText(history: readonly ChatHistoryItem[], { availableFunctions, documentFunctionParams }?: {
+        availableFunctions?: ChatModelFunctions;
+        documentFunctionParams?: boolean;
+    }): {
+        contextText: LlamaText;
+        stopGenerationTriggers: LlamaText[];
+        ignoreStartText?: LlamaText[];
+        functionCall?: {
+            initiallyEngaged: boolean;
+            disengageInitiallyEngaged: LlamaText[];
+        };
+    };
+    generateFunctionCallAndResult(name: string, params: any, result: any): string;
+    generateFunctionCall(name: string, params: any): string;
+    generateFunctionCallResult(functionName: string, functionParams: any, result: any): string;
+    generateModelResponseText(modelResponse: ChatModelResponse["response"]): string;
+    generateAvailableFunctionsSystemText(availableFunctions: ChatModelFunctions, { documentParams }: {
+        documentParams?: boolean;
+    }): string;
+    addAvailableFunctionsSystemMessageToHistory(history: readonly ChatHistoryItem[], availableFunctions?: ChatModelFunctions, { documentParams }?: {
+        documentParams?: boolean;
+    }): readonly ChatHistoryItem[];
+    /**
+     * Functions that should be made available as part of the function calling grammar and are handled by the chat wrapper
+     * for grammar purposes only
+     */
+    getInternalBuiltinFunctions({ initialFunctionCallEngaged }: {
+        initialFunctionCallEngaged: boolean;
+    }): ChatModelFunctions;
+}

package/dist/ChatWrapper.js

@@ -0,0 +1,120 @@
+import { LlamaText } from "./utils/LlamaText.js";
+import { getTypeScriptTypeStringForGbnfJsonSchema } from "./utils/getTypeScriptTypeStringForGbnfJsonSchema.js";
+export class ChatWrapper {
+    settings = {
+        functions: {
+            call: {
+                optionalPrefixSpace: true,
+                prefix: "[[call: ",
+                paramsPrefix: "(",
+                suffix: ")]]"
+            },
+            result: {
+                prefix: " [[result: ",
+                suffix: "]]"
+            }
+        }
+    };
+    generateContextText(history, { availableFunctions, documentFunctionParams } = {}) {
+        const historyWithFunctions = this.addAvailableFunctionsSystemMessageToHistory(history, availableFunctions, {
+            documentParams: documentFunctionParams
+        });
+        const texts = historyWithFunctions
+            .map((item) => {
+            if (item.type === "system")
+                return LlamaText `system: ${item.text}`;
+            else if (item.type === "user")
+                return LlamaText `user: ${item.text}`;
+            else if (item.type === "model")
+                return LlamaText `model: ${this.generateModelResponseText(item.response)}`;
+            return item;
+        });
+        return {
+            contextText: LlamaText(texts).joinValues("\n"),
+            stopGenerationTriggers: []
+        };
+    }
+    generateFunctionCallAndResult(name, params, result) {
+        return this.generateFunctionCall(name, params) + this.generateFunctionCallResult(name, params, result);
+    }
+    generateFunctionCall(name, params) {
+        return this.settings.functions.call.prefix +
+            name +
+            this.settings.functions.call.paramsPrefix +
+            (params === undefined
+                ? ""
+                : JSON.stringify(params)) +
+            this.settings.functions.call.suffix;
+    }
+    generateFunctionCallResult(functionName, functionParams, result) {
+        const resolveParameters = (text) => text.replaceAll("{{functionName}}", functionName)
+            .replaceAll("{{functionParams}}", functionParams === undefined ? "" : JSON.stringify(functionParams));
+        return resolveParameters(this.settings.functions.result.prefix) +
+            (result === undefined
+                ? "void"
+                : JSON.stringify(result)) +
+            resolveParameters(this.settings.functions.result.suffix);
+    }
+    generateModelResponseText(modelResponse) {
+        return modelResponse
+            .map((item) => {
+            if (typeof item === "string")
+                return item;
+            return item.raw ?? this.generateFunctionCallAndResult(item.name, item.params, item.result);
+        })
+            .join("\n");
+    }
+    generateAvailableFunctionsSystemText(availableFunctions, { documentParams = true }) {
+        const availableFunctionNames = Object.keys(availableFunctions ?? {});
+        if (availableFunctionNames.length === 0)
+            return "";
+        return "The assistant calls the provided functions as needed to retrieve information instead of relying on things it already knows.\n" +
+            "Provided functions:\n```\n" +
+            availableFunctionNames
+                .map((functionName) => {
+                const functionDefinition = availableFunctions[functionName];
+                let res = "";
+                if (functionDefinition?.description != null && functionDefinition.description.trim() !== "")
+                    res += "// " + functionDefinition.description.split("\n").join("\n// ") + "\n";
+                res += "function " + functionName + "(";
+                if (documentParams && functionDefinition?.params != null)
+                    res += "params: " + getTypeScriptTypeStringForGbnfJsonSchema(functionDefinition.params);
+                else if (!documentParams && functionDefinition?.params != null)
+                    res += "params";
+                res += ");";
+                return res;
+            })
+                .join("\n\n") +
+            "\n```\n\n" +
+            "Calling any of the provided functions can be done like this:\n" +
+            this.settings.functions.call.prefix.trimStart() +
+            "functionName" +
+            this.settings.functions.call.paramsPrefix +
+            '{ someKey: "someValue" }' +
+            this.settings.functions.call.suffix + "\n\n" +
+            "After calling a function the result will appear afterwards and be visible only to the assistant, so the assistant has to tell the user about it outside of the function call context.\n" +
+            "The assistant calls the functions in advance before telling the user about the result";
+    }
+    addAvailableFunctionsSystemMessageToHistory(history, availableFunctions, { documentParams = true } = {}) {
+        const availableFunctionNames = Object.keys(availableFunctions ?? {});
+        if (availableFunctions == null || availableFunctionNames.length === 0)
+            return history;
+        const res = history.slice();
+        const firstNonSystemMessageIndex = res.findIndex((item) => item.type !== "system");
+        res.splice(Math.max(0, firstNonSystemMessageIndex), 0, {
+            type: "system",
+            text: this.generateAvailableFunctionsSystemText(availableFunctions, { documentParams })
+        });
+        return res;
+    }
+    /**
+     * Functions that should be made available as part of the function calling grammar and are handled by the chat wrapper
+     * for grammar purposes only
+     */
+    getInternalBuiltinFunctions({ initialFunctionCallEngaged }) {
+        if (initialFunctionCallEngaged)
+            return {};
+        return {};
+    }
+}
+//# sourceMappingURL=ChatWrapper.js.map

package/dist/ChatWrapper.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"ChatWrapper.js","sourceRoot":"","sources":["../src/ChatWrapper.ts"],"names":[],"mappings":"AACA,OAAO,EAAC,SAAS,EAAC,MAAM,sBAAsB,CAAC;AAC/C,OAAO,EAAC,wCAAwC,EAAC,MAAM,qDAAqD,CAAC;AAiB7G,MAAM,OAAgB,WAAW;IAGb,QAAQ,GAAwB;QAC5C,SAAS,EAAE;YACP,IAAI,EAAE;gBACF,mBAAmB,EAAE,IAAI;gBACzB,MAAM,EAAE,UAAU;gBAClB,YAAY,EAAE,GAAG;gBACjB,MAAM,EAAE,KAAK;aAChB;YACD,MAAM,EAAE;gBACJ,MAAM,EAAE,aAAa;gBACrB,MAAM,EAAE,IAAI;aACf;SACJ;KACJ,CAAC;IAEK,mBAAmB,CAAC,OAAmC,EAAE,EAAC,kBAAkB,EAAE,sBAAsB,KAGvG,EAAE;QASF,MAAM,oBAAoB,GAAG,IAAI,CAAC,2CAA2C,CAAC,OAAO,EAAE,kBAAkB,EAAE;YACvG,cAAc,EAAE,sBAAsB;SACzC,CAAC,CAAC;QAEH,MAAM,KAAK,GAAG,oBAAoB;aAC7B,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE;YACV,IAAI,IAAI,CAAC,IAAI,KAAK,QAAQ;gBACtB,OAAO,SAAS,CAAC,WAAW,IAAI,CAAC,IAAI,EAAE,CAAC;iBACvC,IAAI,IAAI,CAAC,IAAI,KAAK,MAAM;gBACzB,OAAO,SAAS,CAAC,SAAS,IAAI,CAAC,IAAI,EAAE,CAAC;iBACrC,IAAI,IAAI,CAAC,IAAI,KAAK,OAAO;gBAC1B,OAAO,SAAS,CAAC,UAAU,IAAI,CAAC,yBAAyB,CAAC,IAAI,CAAC,QAAQ,CAAC,EAAE,CAAC;YAE/E,OAAO,IAAoB,CAAC;QAChC,CAAC,CAAC,CAAC;QAEP,OAAO;YACH,WAAW,EAAE,SAAS,CAAC,KAAK,CAAC,CAAC,UAAU,CAAC,IAAI,CAAC;YAC9C,sBAAsB,EAAE,EAAE;SAC7B,CAAC;IACN,CAAC;IAEM,6BAA6B,CAAC,IAAY,EAAE,MAAW,EAAE,MAAW;QACvE,OAAO,IAAI,CAAC,oBAAoB,CAAC,IAAI,EAAE,MAAM,CAAC,GAAG,IAAI,CAAC,0BAA0B,CAAC,IAAI,EAAE,MAAM,EAAE,MAAM,CAAC,CAAC;IAC3G,CAAC;IAEM,oBAAoB,CAAC,IAAY,EAAE,MAAW;QACjD,OAAO,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,CAAC,MAAM;YACtC,IAAI;YACJ,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,CAAC,YAAY;YACzC,CACI,MAAM,KAAK,SAAS;gBAChB,CAAC,CAAC,EAAE;gBACJ,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,CAC/B;YACD,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,CAAC,MAAM,CAAC;IAC5C,CAAC;IAEM,0BAA0B,CAAC,YAAoB,EAAE,cAAmB,EAAE,MAAW;QACpF,MAAM,iBAAiB,GAAG,CAAC,IAAY,EAAE,EAAE,CACvC,IAAI,CAAC,UAAU,CAAC,kBAAkB,EAAE,YAAY,CAAC;aAC5C,UAAU,CAAC,oBAAoB,EAAE,cAAc,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,cAAc,CAAC,CAAC,CAAC;QAE9G,OAAO,iBAAiB,CAAC,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC,MAAM,CAAC,MAAM,CAAC;YAC3D,CACI,MAAM,KAAK,SAAS;gBAChB,CAAC,CAAC,MAAM;gBACR,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,CAC/B;YACD,iBAAiB,CAAC,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;IACjE,CAAC;IAEM,yBAAyB,CAAC,aAA4C;QACzE,OAAO,aAAa;aACf,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE;YACV,IAAI,OAAO,IAAI,KAAK,QAAQ;gBACxB,OAAO,IAAI,CAAC;YAEhB,OAAO,IAAI,CAAC,GAAG,IAAI,IAAI,CAAC,6BAA6B,CAAC,IAAI,CAAC,IAAI,EAAE,IAAI,CAAC,MAAM,EAAE,IAAI,CAAC,MAAM,CAAC,CAAC;QAC/F,CAAC,CAAC;aACD,IAAI,CAAC,IAAI,CAAC,CAAC;IACpB,CAAC;IAEM,oCAAoC,CAAC,kBAAsC,EAAE,EAAC,cAAc,GAAG,IAAI,EAEzG;QACG,MAAM,sBAAsB,GAAG,MAAM,CAAC,IAAI,CAAC,kBAAkB,IAAI,EAAE,CAAC,CAAC;QAErE,IAAI,sBAAsB,CAAC,MAAM,KAAK,CAAC;YACnC,OAAO,EAAE,CAAC;QAEd,OAAO,+HAA+H;YAClI,4BAA4B;YAC5B,sBAAsB;iBACjB,GAAG,CAAC,CAAC,YAAY,EAAE,EAAE;gBAClB,MAAM,kBAAkB,GAAG,kBAAkB,CAAC,YAAY,CAAC,CAAC;gBAC5D,IAAI,GAAG,GAAG,EAAE,CAAC;gBAEb,IAAI,kBAAkB,EAAE,WAAW,IAAI,IAAI,IAAI,kBAAkB,CAAC,WAAW,CAAC,IAAI,EAAE,KAAK,EAAE;oBACvF,GAAG,IAAI,KAAK,GAAG,kBAAkB,CAAC,WAAW,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,GAAG,IAAI,CAAC;gBAEnF,GAAG,IAAI,WAAW,GAAG,YAAY,GAAG,GAAG,CAAC;gBAExC,IAAI,cAAc,IAAI,kBAAkB,EAAE,MAAM,IAAI,IAAI;oBACpD,GAAG,IAAI,UAAU,GAAG,wCAAwC,CAAC,kBAAkB,CAAC,MAAM,CAAC,CAAC;qBACvF,IAAI,CAAC,cAAc,IAAI,kBAAkB,EAAE,MAAM,IAAI,IAAI;oBAC1D,GAAG,IAAI,QAAQ,CAAC;gBAEpB,GAAG,IAAI,IAAI,CAAC;gBAEZ,OAAO,GAAG,CAAC;YACf,CAAC,CAAC;iBACD,IAAI,CAAC,MAAM,CAAC;YACjB,WAAW;YAEX,gEAAgE;YAChE,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,CAAC,MAAM,CAAC,SAAS,EAAE;YAC/C,cAAc;YACd,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,CAAC,YAAY;YACzC,0BAA0B;YAC1B,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,CAAC,MAAM,GAAG,MAAM;YAE5C,yLAAyL;YACzL,uFAAuF,CAAC;IAChG,CAAC;IAEM,2CAA2C,CAAC,OAAmC,EAAE,kBAAuC,EAAE,EAC7H,cAAc,GAAG,IAAI,KAGrB,EAAE;QACF,MAAM,sBAAsB,GAAG,MAAM,CAAC,IAAI,CAAC,kBAAkB,IAAI,EAAE,CAAC,CAAC;QAErE,IAAI,kBAAkB,IAAI,IAAI,IAAI,sBAAsB,CAAC,MAAM,KAAK,CAAC;YACjE,OAAO,OAAO,CAAC;QAEnB,MAAM,GAAG,GAAG,OAAO,CAAC,KAAK,EAAE,CAAC;QAE5B,MAAM,0BAA0B,GAAG,GAAG,CAAC,SAAS,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC;QACnF,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,0BAA0B,CAAC,EAAE,CAAC,EAAE;YACnD,IAAI,EAAE,QAAQ;YACd,IAAI,EAAE,IAAI,CAAC,oCAAoC,CAAC,kBAAkB,EAAE,EAAC,cAAc,EAAC,CAAC;SACxF,CAAC,CAAC;QAEH,OAAO,GAAG,CAAC;IACf,CAAC;IAED;;;OAGG;IACI,2BAA2B,CAAC,EAAC,0BAA0B,EAAwC;QAClG,IAAI,0BAA0B;YAC1B,OAAO,EAAE,CAAC;QAEd,OAAO,EAAE,CAAC;IACd,CAAC;CACJ"}

package/dist/{utils/getBin.d.ts → bindings/AddonTypes.d.ts}

@@ -1,6 +1,4 @@
 import { Token } from "../types.js";
-export declare function getPrebuildBinPath(): Promise<string | null>;
-export declare function loadBin(): Promise<BindingModule>;
 export type BindingModule = {
     AddonModel: {
         new (modelPath: string, params: {
@@ -15,7 +13,6 @@ export type BindingModule = {
             seed?: number;
             contextSize?: number;
             batchSize?: number;
-            f16Kv?: boolean;
             logitsAll?: boolean;
             embedding?: boolean;
             threads?: number;
@@ -30,10 +27,16 @@ export type BindingModule = {
         new (grammar: AddonGrammar): AddonGrammarEvaluationState;
     };
     systemInfo(): string;
+    setLogger(logger: (level: number, message: string) => void): void;
+    setLoggerLogLevel(level: number): void;
+    getGpuVramInfo(): {
+        total: number;
+        used: number;
+    };
 };
 export type AddonModel = {
     dispose(): void;
-    tokenize(text: string): Uint32Array;
+    tokenize(text: string, specialTokens: boolean): Uint32Array;
     detokenize(tokens: Uint32Array): string;
     getTrainContextSize(): number;
     getTotalSize(): number;
@@ -56,6 +59,7 @@ export type AddonContext = {
     decodeBatch(): Promise<void>;
     sampleToken(batchLogitIndex: BatchLogitIndex, options?: {
         temperature?: number;
+        minP?: number;
         topK?: number;
         topP?: number;
         repeatPenalty?: number;
@@ -67,6 +71,9 @@ export type AddonContext = {
     disposeSequence(sequenceId: number): void;
     removeTokenCellsFromSequence(sequenceId: number, startPos: number, endPos: number): void;
     shiftSequenceTokenCells(sequenceId: number, startPos: number, endPos: number, shiftDelta: number): void;
+    acceptGrammarEvaluationStateToken(grammarEvaluationState: AddonGrammarEvaluationState, token: Token): void;
+    getEmbedding(): Float64Array;
+    printTimings(): void;
 };
 export type BatchLogitIndex = number & {
     __batchLogitIndex: never;

package/dist/bindings/AddonTypes.js

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=AddonTypes.js.map

package/dist/bindings/AddonTypes.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"AddonTypes.js","sourceRoot":"","sources":["../../src/bindings/AddonTypes.ts"],"names":[],"mappings":""}

package/dist/bindings/Llama.d.ts

@@ -0,0 +1,23 @@
+import { LlamaLogLevel } from "./types.js";
+export declare class Llama {
+    private constructor();
+    get metal(): boolean;
+    get cuda(): boolean;
+    get logLevel(): LlamaLogLevel;
+    set logLevel(value: LlamaLogLevel);
+    get logger(): (level: LlamaLogLevel, message: string) => void;
+    set logger(value: (level: LlamaLogLevel, message: string) => void);
+    get buildType(): "localBuild" | "prebuilt";
+    get cmakeOptions(): Readonly<Record<string, string>>;
+    get llamaCppRelease(): {
+        readonly repo: string;
+        readonly release: string;
+    };
+    get systemInfo(): string;
+    getVramState(): {
+        total: number;
+        used: number;
+        free: number;
+    };
+    static defaultConsoleLogger(level: LlamaLogLevel, message: string): void;
+}

package/dist/bindings/Llama.js

@@ -0,0 +1,225 @@
+import chalk from "chalk";
+import { getConsoleLogPrefix } from "../utils/getConsoleLogPrefix.js";
+import { LlamaLogLevel } from "./types.js";
+const LlamaLogLevelToAddonLogLevel = new Map([
+    [LlamaLogLevel.disabled, 0],
+    [LlamaLogLevel.fatal, 1],
+    [LlamaLogLevel.error, 2],
+    [LlamaLogLevel.warn, 3],
+    [LlamaLogLevel.info, 4],
+    [LlamaLogLevel.debug, 5]
+]);
+const addonLogLevelToLlamaLogLevel = new Map([...LlamaLogLevelToAddonLogLevel.entries()].map(([key, value]) => [value, key]));
+const defaultLogLevel = 5;
+export class Llama {
+    /** @internal */ _bindings;
+    /** @internal */ _metal;
+    /** @internal */ _cuda;
+    /** @internal */ _buildType;
+    /** @internal */ _cmakeOptions;
+    /** @internal */ _llamaCppRelease;
+    /** @internal */ _logger;
+    /** @internal */ _logLevel;
+    /** @internal */ _pendingLog = null;
+    /** @internal */ _pendingLogLevel = null;
+    /** @internal */ _logDispatchQueuedMicrotasks = 0;
+    /** @internal */ _previousLog = null;
+    /** @internal */ _previousLogLevel = null;
+    /** @internal */ _nextLogNeedNewLine = false;
+    constructor({ bindings, metal, cuda, logLevel, logger, buildType, cmakeOptions, llamaCppRelease }) {
+        this._bindings = bindings;
+        this._metal = metal;
+        this._cuda = cuda;
+        this._logLevel = logLevel ?? LlamaLogLevel.debug;
+        this._logger = logger;
+        this._buildType = buildType;
+        this._cmakeOptions = Object.freeze({ ...cmakeOptions });
+        this._llamaCppRelease = Object.freeze({
+            repo: llamaCppRelease.repo,
+            release: llamaCppRelease.release
+        });
+        this._dispatchPendingLogMicrotask = this._dispatchPendingLogMicrotask.bind(this);
+        this._onAddonLog = this._onAddonLog.bind(this);
+        this._bindings.setLogger(this._onAddonLog);
+        this._bindings.setLoggerLogLevel(LlamaLogLevelToAddonLogLevel.get(this._logLevel) ?? defaultLogLevel);
+    }
+    get metal() {
+        return this._metal;
+    }
+    get cuda() {
+        return this._cuda;
+    }
+    get logLevel() {
+        return this._logLevel;
+    }
+    set logLevel(value) {
+        if (value === this._logLevel)
+            return;
+        this._bindings.setLoggerLogLevel(LlamaLogLevelToAddonLogLevel.get(value) ?? defaultLogLevel);
+        this._logLevel = value;
+    }
+    get logger() {
+        return this._logger;
+    }
+    set logger(value) {
+        this._logger = value;
+        if (value !== Llama.defaultConsoleLogger)
+            this._nextLogNeedNewLine = false;
+    }
+    get buildType() {
+        return this._buildType;
+    }
+    get cmakeOptions() {
+        return this._cmakeOptions;
+    }
+    get llamaCppRelease() {
+        return this._llamaCppRelease;
+    }
+    get systemInfo() {
+        return this._bindings.systemInfo();
+    }
+    getVramState() {
+        const { total, used } = this._bindings.getGpuVramInfo();
+        return {
+            total,
+            used,
+            free: Math.max(0, total - used)
+        };
+    }
+    /** @internal */
+    _onAddonLog(level, message) {
+        const llamaLogLevel = addonLogLevelToLlamaLogLevel.get(level) ?? LlamaLogLevel.fatal;
+        if (this._pendingLog != null && this._pendingLogLevel != null && this._pendingLogLevel != llamaLogLevel) {
+            this._callLogger(this._pendingLogLevel, this._pendingLog);
+            this._pendingLog = null;
+        }
+        const sourceMessage = (this._pendingLog ?? "") + message;
+        const lastNewLineIndex = sourceMessage.lastIndexOf("\n");
+        const currentLog = lastNewLineIndex < 0
+            ? sourceMessage
+            : sourceMessage.slice(0, lastNewLineIndex);
+        const nextLog = lastNewLineIndex < 0
+            ? ""
+            : sourceMessage.slice(lastNewLineIndex + 1);
+        if (currentLog !== "")
+            this._callLogger(llamaLogLevel, currentLog);
+        if (nextLog !== "") {
+            this._pendingLog = nextLog;
+            this._pendingLogLevel = llamaLogLevel;
+            queueMicrotask(this._dispatchPendingLogMicrotask);
+            this._logDispatchQueuedMicrotasks++;
+        }
+        else
+            this._pendingLog = null;
+    }
+    /** @internal */
+    _dispatchPendingLogMicrotask() {
+        this._logDispatchQueuedMicrotasks--;
+        if (this._logDispatchQueuedMicrotasks !== 0)
+            return;
+        if (this._pendingLog != null && this._pendingLogLevel != null) {
+            this._callLogger(this._pendingLogLevel, this._pendingLog);
+            this._pendingLog = null;
+        }
+    }
+    /** @internal */
+    _callLogger(level, message) {
+        // llama.cpp uses dots to indicate progress, so we don't want to print them as different lines,
+        // and instead, append to the same log line
+        if (logMessageIsOnlyDots(message) && this._logger === Llama.defaultConsoleLogger) {
+            if (logMessageIsOnlyDots(this._previousLog) && level === this._previousLogLevel) {
+                process.stdout.write(message);
+            }
+            else {
+                this._nextLogNeedNewLine = true;
+                process.stdout.write(prefixAndColorMessage(message, getColorForLogLevel(level)));
+            }
+        }
+        else {
+            if (this._nextLogNeedNewLine) {
+                process.stdout.write("\n");
+                this._nextLogNeedNewLine = false;
+            }
+            try {
+                this._logger(level, message);
+            }
+            catch (err) {
+                // the native addon code calls this function, so there's no use to throw an error here
+            }
+        }
+        this._previousLog = message;
+        this._previousLogLevel = level;
+    }
+    /** @internal */
+    static async _create({ bindings, buildType, buildMetadata, logLevel, logger }) {
+        return new Llama({
+            bindings,
+            buildType,
+            metal: buildMetadata.buildOptions.computeLayers.metal,
+            cuda: buildMetadata.buildOptions.computeLayers.cuda,
+            cmakeOptions: buildMetadata.buildOptions.customCmakeOptions,
+            llamaCppRelease: {
+                repo: buildMetadata.buildOptions.llamaCpp.repo,
+                release: buildMetadata.buildOptions.llamaCpp.release
+            },
+            logLevel,
+            logger
+        });
+    }
+    static defaultConsoleLogger(level, message) {
+        switch (level) {
+            case LlamaLogLevel.disabled:
+                break;
+            case LlamaLogLevel.fatal:
+                // we don't use console.error here because it prints the stack trace
+                console.warn(prefixAndColorMessage(message, getColorForLogLevel(level)));
+                break;
+            case LlamaLogLevel.error:
+                // we don't use console.error here because it prints the stack trace
+                console.warn(prefixAndColorMessage(message, getColorForLogLevel(level)));
+                break;
+            case LlamaLogLevel.warn:
+                console.warn(prefixAndColorMessage(message, getColorForLogLevel(level)));
+                break;
+            case LlamaLogLevel.info:
+                console.info(prefixAndColorMessage(message, getColorForLogLevel(level)));
+                break;
+            case LlamaLogLevel.debug:
+                console.debug(prefixAndColorMessage(message, getColorForLogLevel(level)));
+                break;
+            default:
+                void (level);
+                console.warn(getConsoleLogPrefix() + getColorForLogLevel(LlamaLogLevel.warn)(`Unknown log level: ${level}`));
+                console.log(prefixAndColorMessage(message, getColorForLogLevel(level)));
+        }
+    }
+}
+function getColorForLogLevel(level) {
+    switch (level) {
+        case LlamaLogLevel.disabled: return chalk.whiteBright;
+        case LlamaLogLevel.fatal: return chalk.redBright;
+        case LlamaLogLevel.error: return chalk.red;
+        case LlamaLogLevel.warn: return chalk.yellow;
+        case LlamaLogLevel.info: return chalk.whiteBright;
+        case LlamaLogLevel.debug: return chalk.white;
+        default:
+            void (level);
+            return chalk.whiteBright;
+    }
+}
+function prefixAndColorMessage(message, color) {
+    return getConsoleLogPrefix() + (message
+        .split("\n")
+        .map(line => color(line))
+        .join("\n" + getConsoleLogPrefix()));
+}
+function logMessageIsOnlyDots(message) {
+    if (message == null)
+        return false;
+    for (let i = 0; i < message.length; i++) {
+        if (message[i] !== ".")
+            return false;
+    }
+    return true;
+}
+//# sourceMappingURL=Llama.js.map

package/dist/bindings/Llama.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"Llama.js","sourceRoot":"","sources":["../../src/bindings/Llama.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,MAAM,OAAO,CAAC;AAC1B,OAAO,EAAC,mBAAmB,EAAC,MAAM,iCAAiC,CAAC;AAEpE,OAAO,EAAoB,aAAa,EAAC,MAAM,YAAY,CAAC;AAE5D,MAAM,4BAA4B,GAAuC,IAAI,GAAG,CAAC;IAC7E,CAAC,aAAa,CAAC,QAAQ,EAAE,CAAC,CAAC;IAC3B,CAAC,aAAa,CAAC,KAAK,EAAE,CAAC,CAAC;IACxB,CAAC,aAAa,CAAC,KAAK,EAAE,CAAC,CAAC;IACxB,CAAC,aAAa,CAAC,IAAI,EAAE,CAAC,CAAC;IACvB,CAAC,aAAa,CAAC,IAAI,EAAE,CAAC,CAAC;IACvB,CAAC,aAAa,CAAC,KAAK,EAAE,CAAC,CAAC;CAC3B,CAAC,CAAC;AACH,MAAM,4BAA4B,GAAuC,IAAI,GAAG,CAC5E,CAAC,GAAG,4BAA4B,CAAC,OAAO,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,EAAE,KAAK,CAAC,EAAE,EAAE,CAAC,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC,CAClF,CAAC;AACF,MAAM,eAAe,GAAG,CAAC,CAAC;AAE1B,MAAM,OAAO,KAAK;IACd,gBAAgB,CAAiB,SAAS,CAAgB;IAC1D,gBAAgB,CAAkB,MAAM,CAAU;IAClD,gBAAgB,CAAkB,KAAK,CAAU;IACjD,gBAAgB,CAAkB,UAAU,CAA4B;IACxE,gBAAgB,CAAkB,aAAa,CAAmC;IAClF,gBAAgB,CAAkB,gBAAgB,CAGhD;IACF,gBAAgB,CAAS,OAAO,CAAoD;IACpF,gBAAgB,CAAS,SAAS,CAAgB;IAClD,gBAAgB,CAAS,WAAW,GAAkB,IAAI,CAAC;IAC3D,gBAAgB,CAAS,gBAAgB,GAAyB,IAAI,CAAC;IACvE,gBAAgB,CAAS,4BAA4B,GAAW,CAAC,CAAC;IAClE,gBAAgB,CAAS,YAAY,GAAkB,IAAI,CAAC;IAC5D,gBAAgB,CAAS,iBAAiB,GAAyB,IAAI,CAAC;IACxE,gBAAgB,CAAS,mBAAmB,GAAY,KAAK,CAAC;IAE9D,YAAoB,EAChB,QAAQ,EAAE,KAAK,EAAE,IAAI,EAAE,QAAQ,EAAE,MAAM,EAAE,SAAS,EAAE,YAAY,EAAE,eAAe,EAapF;QACG,IAAI,CAAC,SAAS,GAAG,QAAQ,CAAC;QAC1B,IAAI,CAAC,MAAM,GAAG,KAAK,CAAC;QACpB,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC;QAClB,IAAI,CAAC,SAAS,GAAG,QAAQ,IAAI,aAAa,CAAC,KAAK,CAAC;QACjD,IAAI,CAAC,OAAO,GAAG,MAAM,CAAC;QACtB,IAAI,CAAC,UAAU,GAAG,SAAS,CAAC;QAC5B,IAAI,CAAC,aAAa,GAAG,MAAM,CAAC,MAAM,CAAC,EAAC,GAAG,YAAY,EAAC,CAAC,CAAC;QACtD,IAAI,CAAC,gBAAgB,GAAG,MAAM,CAAC,MAAM,CAAC;YAClC,IAAI,EAAE,eAAe,CAAC,IAAI;YAC1B,OAAO,EAAE,eAAe,CAAC,OAAO;SACnC,CAAC,CAAC;QAEH,IAAI,CAAC,4BAA4B,GAAG,IAAI,CAAC,4BAA4B,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACjF,IAAI,CAAC,WAAW,GAAG,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAE/C,IAAI,CAAC,SAAS,CAAC,SAAS,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;QAC3C,IAAI,CAAC,SAAS,CAAC,iBAAiB,CAAC,4BAA4B,CAAC,GAAG,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,eAAe,CAAC,CAAC;IAC1G,CAAC;IAED,IAAW,KAAK;QACZ,OAAO,IAAI,CAAC,MAAM,CAAC;IACvB,CAAC;IAED,IAAW,IAAI;QACX,OAAO,IAAI,CAAC,KAAK,CAAC;IACtB,CAAC;IAED,IAAW,QAAQ;QACf,OAAO,IAAI,CAAC,SAAS,CAAC;IAC1B,CAAC;IAED,IAAW,QAAQ,CAAC,KAAoB;QACpC,IAAI,KAAK,KAAK,IAAI,CAAC,SAAS;YACxB,OAAO;QAEX,IAAI,CAAC,SAAS,CAAC,iBAAiB,CAAC,4BAA4B,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,eAAe,CAAC,CAAC;QAC7F,IAAI,CAAC,SAAS,GAAG,KAAK,CAAC;IAC3B,CAAC;IAED,IAAW,MAAM;QACb,OAAO,IAAI,CAAC,OAAO,CAAC;IACxB,CAAC;IAED,IAAW,MAAM,CAAC,KAAsD;QACpE,IAAI,CAAC,OAAO,GAAG,KAAK,CAAC;QAErB,IAAI,KAAK,KAAK,KAAK,CAAC,oBAAoB;YACpC,IAAI,CAAC,mBAAmB,GAAG,KAAK,CAAC;IACzC,CAAC;IAED,IAAW,SAAS;QAChB,OAAO,IAAI,CAAC,UAAU,CAAC;IAC3B,CAAC;IAED,IAAW,YAAY;QACnB,OAAO,IAAI,CAAC,aAAa,CAAC;IAC9B,CAAC;IAED,IAAW,eAAe;QACtB,OAAO,IAAI,CAAC,gBAAgB,CAAC;IACjC,CAAC;IAED,IAAW,UAAU;QACjB,OAAO,IAAI,CAAC,SAAS,CAAC,UAAU,EAAE,CAAC;IACvC,CAAC;IAEM,YAAY;QACf,MAAM,EAAC,KAAK,EAAE,IAAI,EAAC,GAAG,IAAI,CAAC,SAAS,CAAC,cAAc,EAAE,CAAC;QAEtD,OAAO;YACH,KAAK;YACL,IAAI;YACJ,IAAI,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,KAAK,GAAG,IAAI,CAAC;SAClC,CAAC;IACN,CAAC;IAED,gBAAgB;IACR,WAAW,CAAC,KAAa,EAAE,OAAe;QAC9C,MAAM,aAAa,GAAG,4BAA4B,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,aAAa,CAAC,KAAK,CAAC;QAErF,IAAI,IAAI,CAAC,WAAW,IAAI,IAAI,IAAI,IAAI,CAAC,gBAAgB,IAAI,IAAI,IAAI,IAAI,CAAC,gBAAgB,IAAI,aAAa,EAAE;YACrG,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC,gBAAgB,EAAE,IAAI,CAAC,WAAW,CAAC,CAAC;YAC1D,IAAI,CAAC,WAAW,GAAG,IAAI,CAAC;SAC3B;QAED,MAAM,aAAa,GAAG,CAAC,IAAI,CAAC,WAAW,IAAI,EAAE,CAAC,GAAG,OAAO,CAAC;QAEzD,MAAM,gBAAgB,GAAG,aAAa,CAAC,WAAW,CAAC,IAAI,CAAC,CAAC;QACzD,MAAM,UAAU,GAAG,gBAAgB,GAAG,CAAC;YACnC,CAAC,CAAC,aAAa;YACf,CAAC,CAAC,aAAa,CAAC,KAAK,CAAC,CAAC,EAAE,gBAAgB,CAAC,CAAC;QAC/C,MAAM,OAAO,GAAG,gBAAgB,GAAG,CAAC;YAChC,CAAC,CAAC,EAAE;YACJ,CAAC,CAAC,aAAa,CAAC,KAAK,CAAC,gBAAgB,GAAG,CAAC,CAAC,CAAC;QAEhD,IAAI,UAAU,KAAK,EAAE;YACjB,IAAI,CAAC,WAAW,CAAC,aAAa,EAAE,UAAU,CAAC,CAAC;QAEhD,IAAI,OAAO,KAAK,EAAE,EAAE;YAChB,IAAI,CAAC,WAAW,GAAG,OAAO,CAAC;YAC3B,IAAI,CAAC,gBAAgB,GAAG,aAAa,CAAC;YAEtC,cAAc,CAAC,IAAI,CAAC,4BAA4B,CAAC,CAAC;YAClD,IAAI,CAAC,4BAA4B,EAAE,CAAC;SACvC;;YACG,IAAI,CAAC,WAAW,GAAG,IAAI,CAAC;IAChC,CAAC;IAED,gBAAgB;IACR,4BAA4B;QAChC,IAAI,CAAC,4BAA4B,EAAE,CAAC;QACpC,IAAI,IAAI,CAAC,4BAA4B,KAAK,CAAC;YACvC,OAAO;QAEX,IAAI,IAAI,CAAC,WAAW,IAAI,IAAI,IAAI,IAAI,CAAC,gBAAgB,IAAI,IAAI,EAAE;YAC3D,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC,gBAAgB,EAAE,IAAI,CAAC,WAAW,CAAC,CAAC;YAC1D,IAAI,CAAC,WAAW,GAAG,IAAI,CAAC;SAC3B;IACL,CAAC;IAED,gBAAgB;IACR,WAAW,CAAC,KAAoB,EAAE,OAAe;QACrD,+FAA+F;QAC/F,2CAA2C;QAC3C,IAAI,oBAAoB,CAAC,OAAO,CAAC,IAAI,IAAI,CAAC,OAAO,KAAK,KAAK,CAAC,oBAAoB,EAAE;YAC9E,IAAI,oBAAoB,CAAC,IAAI,CAAC,YAAY,CAAC,IAAI,KAAK,KAAK,IAAI,CAAC,iBAAiB,EAAE;gBAC7E,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;aACjC;iBAAM;gBACH,IAAI,CAAC,mBAAmB,GAAG,IAAI,CAAC;gBAChC,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,qBAAqB,CAAC,OAAO,EAAE,mBAAmB,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;aACpF;SACJ;aAAM;YACH,IAAI,IAAI,CAAC,mBAAmB,EAAE;gBAC1B,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;gBAC3B,IAAI,CAAC,mBAAmB,GAAG,KAAK,CAAC;aACpC;YAED,IAAI;gBACA,IAAI,CAAC,OAAO,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC;aAChC;YAAC,OAAO,GAAG,EAAE;gBACV,sFAAsF;aACzF;SACJ;QAED,IAAI,CAAC,YAAY,GAAG,OAAO,CAAC;QAC5B,IAAI,CAAC,iBAAiB,GAAG,KAAK,CAAC;IACnC,CAAC;IAED,gBAAgB;IACT,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,EACxB,QAAQ,EAAE,SAAS,EAAE,aAAa,EAAE,QAAQ,EAAE,MAAM,EAOvD;QACG,OAAO,IAAI,KAAK,CAAC;YACb,QAAQ;YACR,SAAS;YACT,KAAK,EAAE,aAAa,CAAC,YAAY,CAAC,aAAa,CAAC,KAAK;YACrD,IAAI,EAAE,aAAa,CAAC,YAAY,CAAC,aAAa,CAAC,IAAI;YACnD,YAAY,EAAE,aAAa,CAAC,YAAY,CAAC,kBAAkB;YAC3D,eAAe,EAAE;gBACb,IAAI,EAAE,aAAa,CAAC,YAAY,CAAC,QAAQ,CAAC,IAAI;gBAC9C,OAAO,EAAE,aAAa,CAAC,YAAY,CAAC,QAAQ,CAAC,OAAO;aACvD;YACD,QAAQ;YACR,MAAM;SACT,CAAC,CAAC;IACP,CAAC;IAEM,MAAM,CAAC,oBAAoB,CAAC,KAAoB,EAAE,OAAe;QACpE,QAAQ,KAAK,EAAE;YACX,KAAK,aAAa,CAAC,QAAQ;gBACvB,MAAM;YACV,KAAK,aAAa,CAAC,KAAK;gBACpB,oEAAoE;gBACpE,OAAO,CAAC,IAAI,CAAC,qBAAqB,CAAC,OAAO,EAAE,mBAAmB,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;gBACzE,MAAM;YACV,KAAK,aAAa,CAAC,KAAK;gBACpB,oEAAoE;gBACpE,OAAO,CAAC,IAAI,CAAC,qBAAqB,CAAC,OAAO,EAAE,mBAAmB,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;gBACzE,MAAM;YACV,KAAK,aAAa,CAAC,IAAI;gBACnB,OAAO,CAAC,IAAI,CAAC,qBAAqB,CAAC,OAAO,EAAE,mBAAmB,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;gBACzE,MAAM;YACV,KAAK,aAAa,CAAC,IAAI;gBACnB,OAAO,CAAC,IAAI,CAAC,qBAAqB,CAAC,OAAO,EAAE,mBAAmB,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;gBACzE,MAAM;YACV,KAAK,aAAa,CAAC,KAAK;gBACpB,OAAO,CAAC,KAAK,CAAC,qBAAqB,CAAC,OAAO,EAAE,mBAAmB,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;gBAC1E,MAAM;YACV;gBACI,KAAK,CAAC,KAAqB,CAAC,CAAC;gBAC7B,OAAO,CAAC,IAAI,CAAC,mBAAmB,EAAE,GAAG,mBAAmB,CAAC,aAAa,CAAC,IAAI,CAAC,CAAC,sBAAsB,KAAK,EAAE,CAAC,CAAC,CAAC;gBAC7G,OAAO,CAAC,GAAG,CAAC,qBAAqB,CAAC,OAAO,EAAE,mBAAmB,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;SAC/E;IAEL,CAAC;CACJ;AAED,SAAS,mBAAmB,CAAC,KAAoB;IAC7C,QAAQ,KAAK,EAAE;QACX,KAAK,aAAa,CAAC,QAAQ,CAAC,CAAC,OAAO,KAAK,CAAC,WAAW,CAAC;QACtD,KAAK,aAAa,CAAC,KAAK,CAAC,CAAC,OAAO,KAAK,CAAC,SAAS,CAAC;QACjD,KAAK,aAAa,CAAC,KAAK,CAAC,CAAC,OAAO,KAAK,CAAC,GAAG,CAAC;QAC3C,KAAK,aAAa,CAAC,IAAI,CAAC,CAAC,OAAO,KAAK,CAAC,MAAM,CAAC;QAC7C,KAAK,aAAa,CAAC,IAAI,CAAC,CAAC,OAAO,KAAK,CAAC,WAAW,CAAC;QAClD,KAAK,aAAa,CAAC,KAAK,CAAC,CAAC,OAAO,KAAK,CAAC,KAAK,CAAC;QAC7C;YACI,KAAK,CAAC,KAAqB,CAAC,CAAC;YAC7B,OAAO,KAAK,CAAC,WAAW,CAAC;KAChC;AACL,CAAC;AAED,SAAS,qBAAqB,CAAC,OAAe,EAAE,KAAkC;IAC9E,OAAO,mBAAmB,EAAE,GAAG,CAC3B,OAAO;SACF,KAAK,CAAC,IAAI,CAAC;SACX,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;SACxB,IAAI,CAAC,IAAI,GAAG,mBAAmB,EAAE,CAAC,CAC1C,CAAC;AACN,CAAC;AAED,SAAS,oBAAoB,CAAC,OAAsB;IAChD,IAAI,OAAO,IAAI,IAAI;QACf,OAAO,KAAK,CAAC;IAEjB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,OAAO,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE;QACrC,IAAI,OAAO,CAAC,CAAC,CAAC,KAAK,GAAG;YAClB,OAAO,KAAK,CAAC;KACpB;IAED,OAAO,IAAI,CAAC;AAChB,CAAC"}

package/dist/bindings/getLlama.d.ts

@@ -0,0 +1,86 @@
+import { 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.
+     */
+    cuda?: boolean;
+    /**
+     * Set the minimum log level for llama.cpp.
+     * Defaults to "debug".
+     */
+    logLevel?: LlamaLogLevel;
+    /**
+     * Set a custom logger for llama.cpp logs.
+     */
+    logger?: (level: LlamaLogLevel, message: string) => void;
+    /**
+     * Set what build method to use.
+     * - **`"auto"`**: If a local build is found, use it.
+     * Otherwise, if a prebuilt binary is found, use it.
+     * Otherwise, build from source.
+     * - **`"never"`**: If a local build is found, use it.
+     * Otherwise, if a prebuilt binary is found, use it.
+     * Otherwise, throw a `NoBinaryFoundError` error.
+     * - **`"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".
+     */
+    build?: "auto" | "never" | "forceRebuild";
+    /**
+     * Set custom CMake options for llama.cpp
+     */
+    cmakeOptions?: Record<string, string>;
+    /**
+     * When a prebuilt binary is found, only use it if it was built with the same build options as the ones specified in `buildOptions`.
+     * Disabled by default.
+     */
+    existingPrebuiltBinaryMustMatchBuildOptions?: boolean;
+    /**
+     * Use prebuilt binaries if they match the build options.
+     * Enabled by default.
+     */
+    usePrebuiltBinaries?: boolean;
+    /**
+     * Print binary compilation progress logs.
+     * Enabled by default.
+     */
+    progressLogs?: boolean;
+    /**
+     * Don't download llama.cpp source if it's not found.
+     * When set to `true`, and llama.cpp source is not found, a `NoBinaryFoundError` error will be thrown.
+     * Disabled by default.
+     */
+    skipDownload?: boolean;
+};
+export type LastBuildOptions = {
+    /**
+     * Set the minimum log level for llama.cpp.
+     * Defaults to "debug".
+     */
+    logLevel?: LlamaLogLevel;
+    /**
+     * Set a custom logger for llama.cpp logs.
+     */
+    logger?: (level: LlamaLogLevel, message: string) => void;
+};
+export declare const getLlamaFunctionName = "getLlama";
+/**
+ * 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.
+ */
+export declare function getLlama(): Promise<Llama>;
+export declare function getLlama(options: LlamaOptions, lastBuildOptions?: never): Promise<Llama>;
+export declare function getLlama(type: "lastBuild", lastBuildOptions?: LastBuildOptions): Promise<Llama>;
+export declare function getLlamaForOptions({ metal, cuda, logLevel, logger, build, cmakeOptions, existingPrebuiltBinaryMustMatchBuildOptions, usePrebuiltBinaries, progressLogs, skipDownload }: LlamaOptions, { updateLastBuildInfoOnCompile }?: {
+    updateLastBuildInfoOnCompile?: boolean;
+}): Promise<Llama>;