@realtimex/node-llama-cpp 0.1.0

package/dist/evaluator/LlamaContext/utils/batchItemsPrioritizationStrategies/maximumParallelismStrategy.js

@@ -0,0 +1,42 @@
+export function maximumParallelismStrategy({ items, size }) {
+    let leftFreeTokens = size;
+    const minTokensForEachItem = Math.floor(leftFreeTokens / items.length);
+    const res = [];
+    const clippedItems = [];
+    for (const item of items) {
+        const processAmount = Math.min(item.tokens.length, leftFreeTokens, minTokensForEachItem);
+        const prioritizeItem = { item, processAmount };
+        res.push(prioritizeItem);
+        leftFreeTokens -= processAmount;
+        if (processAmount < item.tokens.length)
+            clippedItems.push(prioritizeItem);
+        if (leftFreeTokens === 0)
+            break;
+    }
+    for (let passesLeft = 3; leftFreeTokens > 0 && clippedItems.length > 0 && passesLeft > 0; passesLeft--) {
+        const minIncreaseAmount = Math.ceil(leftFreeTokens / clippedItems.length);
+        for (let i = 0; i < clippedItems.length && leftFreeTokens > 0; i++) {
+            const prioritizeItem = clippedItems[i];
+            const unprocessedAmount = prioritizeItem.item.tokens.length - prioritizeItem.processAmount;
+            const increaseAmount = Math.min(unprocessedAmount, leftFreeTokens, minIncreaseAmount);
+            prioritizeItem.processAmount += increaseAmount;
+            if (increaseAmount === unprocessedAmount) {
+                clippedItems.splice(i, 1);
+                i--;
+            }
+        }
+    }
+    clippedItems.sort((a, b) => b.item.evaluationPriority - a.item.evaluationPriority);
+    for (let i = 0; i < clippedItems.length && leftFreeTokens > 0; i++) {
+        const prioritizeItem = clippedItems[i];
+        const unprocessedAmount = prioritizeItem.item.tokens.length - prioritizeItem.processAmount;
+        const increaseAmount = Math.min(unprocessedAmount, leftFreeTokens);
+        prioritizeItem.processAmount += increaseAmount;
+        if (increaseAmount === unprocessedAmount) {
+            clippedItems.splice(i, 1);
+            i--;
+        }
+    }
+    return res;
+}
+//# sourceMappingURL=maximumParallelismStrategy.js.map

package/dist/evaluator/LlamaContext/utils/batchItemsPrioritizationStrategies/maximumParallelismStrategy.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"maximumParallelismStrategy.js","sourceRoot":"","sources":["../../../../../src/evaluator/LlamaContext/utils/batchItemsPrioritizationStrategies/maximumParallelismStrategy.ts"],"names":[],"mappings":"AAEA,MAAM,UAAU,0BAA0B,CAAC,EAAC,KAAK,EAAE,IAAI,EAA8C;IACjG,IAAI,cAAc,GAAG,IAAI,CAAC;IAC1B,MAAM,oBAAoB,GAAG,IAAI,CAAC,KAAK,CAAC,cAAc,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC;IAEvE,MAAM,GAAG,GAA2B,EAAE,CAAC;IACvC,MAAM,YAAY,GAA2B,EAAE,CAAC;IAEhD,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACvB,MAAM,aAAa,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,MAAM,EAAE,cAAc,EAAE,oBAAoB,CAAC,CAAC;QACzF,MAAM,cAAc,GAAG,EAAC,IAAI,EAAE,aAAa,EAAC,CAAC;QAE7C,GAAG,CAAC,IAAI,CAAC,cAAc,CAAC,CAAC;QACzB,cAAc,IAAI,aAAa,CAAC;QAEhC,IAAI,aAAa,GAAG,IAAI,CAAC,MAAM,CAAC,MAAM;YAClC,YAAY,CAAC,IAAI,CAAC,cAAc,CAAC,CAAC;QAEtC,IAAI,cAAc,KAAK,CAAC;YACpB,MAAM;IACd,CAAC;IAED,KAAK,IAAI,UAAU,GAAG,CAAC,EAAE,cAAc,GAAG,CAAC,IAAI,YAAY,CAAC,MAAM,GAAG,CAAC,IAAI,UAAU,GAAG,CAAC,EAAE,UAAU,EAAE,EAAE,CAAC;QACrG,MAAM,iBAAiB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,GAAG,YAAY,CAAC,MAAM,CAAC,CAAC;QAE1E,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,YAAY,CAAC,MAAM,IAAI,cAAc,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;YACjE,MAAM,cAAc,GAAG,YAAY,CAAC,CAAC,CAAE,CAAC;YACxC,MAAM,iBAAiB,GAAG,cAAc,CAAC,IAAI,CAAC,MAAM,CAAC,MAAM,GAAG,cAAc,CAAC,aAAa,CAAC;YAC3F,MAAM,cAAc,GAAG,IAAI,CAAC,GAAG,CAAC,iBAAiB,EAAE,cAAc,EAAE,iBAAiB,CAAC,CAAC;YACtF,cAAc,CAAC,aAAa,IAAI,cAAc,CAAC;YAE/C,IAAI,cAAc,KAAK,iBAAiB,EAAE,CAAC;gBACvC,YAAY,CAAC,MAAM,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;gBAC1B,CAAC,EAAE,CAAC;YACR,CAAC;QACL,CAAC;IACL,CAAC;IAED,YAAY,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,kBAAkB,GAAG,CAAC,CAAC,IAAI,CAAC,kBAAkB,CAAC,CAAC;IAEnF,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,YAAY,CAAC,MAAM,IAAI,cAAc,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QACjE,MAAM,cAAc,GAAG,YAAY,CAAC,CAAC,CAAE,CAAC;QACxC,MAAM,iBAAiB,GAAG,cAAc,CAAC,IAAI,CAAC,MAAM,CAAC,MAAM,GAAG,cAAc,CAAC,aAAa,CAAC;QAC3F,MAAM,cAAc,GAAG,IAAI,CAAC,GAAG,CAAC,iBAAiB,EAAE,cAAc,CAAC,CAAC;QACnE,cAAc,CAAC,aAAa,IAAI,cAAc,CAAC;QAE/C,IAAI,cAAc,KAAK,iBAAiB,EAAE,CAAC;YACvC,YAAY,CAAC,MAAM,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;YAC1B,CAAC,EAAE,CAAC;QACR,CAAC;IACL,CAAC;IAED,OAAO,GAAG,CAAC;AACf,CAAC"}

package/dist/evaluator/LlamaContext/utils/padSafeContextSize.d.ts

@@ -0,0 +1 @@
+export declare function padSafeContextSize(value: number, padDirection: "up" | "down", padding?: number): number;

package/dist/evaluator/LlamaContext/utils/padSafeContextSize.js

@@ -0,0 +1,18 @@
+import { contextSizePad } from "../../../config.js";
+export function padSafeContextSize(value, padDirection, padding = contextSizePad) {
+    const paddedSize = ggmlPad(value, padding);
+    if (paddedSize === value)
+        return value;
+    else if (padDirection === "up")
+        return paddedSize;
+    else if (padDirection === "down") {
+        const smallerPaddedSize = ggmlPad(value - padding, padding);
+        if (smallerPaddedSize >= padding)
+            return smallerPaddedSize;
+    }
+    return paddedSize;
+}
+function ggmlPad(value, padding) {
+    return ((value + padding - 1) & ~(padding - 1));
+}
+//# sourceMappingURL=padSafeContextSize.js.map

package/dist/evaluator/LlamaContext/utils/padSafeContextSize.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"padSafeContextSize.js","sourceRoot":"","sources":["../../../../src/evaluator/LlamaContext/utils/padSafeContextSize.ts"],"names":[],"mappings":"AAAA,OAAO,EAAC,cAAc,EAAC,MAAM,oBAAoB,CAAC;AAElD,MAAM,UAAU,kBAAkB,CAAC,KAAa,EAAE,YAA2B,EAAE,UAAkB,cAAc;IAC3G,MAAM,UAAU,GAAG,OAAO,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC;IAE3C,IAAI,UAAU,KAAK,KAAK;QACpB,OAAO,KAAK,CAAC;SACZ,IAAI,YAAY,KAAK,IAAI;QAC1B,OAAO,UAAU,CAAC;SACjB,IAAI,YAAY,KAAK,MAAM,EAAE,CAAC;QAC/B,MAAM,iBAAiB,GAAG,OAAO,CAAC,KAAK,GAAG,OAAO,EAAE,OAAO,CAAC,CAAC;QAC5D,IAAI,iBAAiB,IAAI,OAAO;YAC5B,OAAO,iBAAiB,CAAC;IACjC,CAAC;IAED,OAAO,UAAU,CAAC;AACtB,CAAC;AACD,SAAS,OAAO,CAAC,KAAa,EAAE,OAAe;IAC3C,OAAO,CAAC,CAAC,KAAK,GAAG,OAAO,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,OAAO,GAAG,CAAC,CAAC,CAAC,CAAC;AACpD,CAAC"}

package/dist/evaluator/LlamaContext/utils/resolveBatchItemsPrioritizationStrategy.d.ts

@@ -0,0 +1,2 @@
+import { BatchingOptions } from "../types.js";
+export declare function resolveBatchItemsPrioritizationStrategy(strategy: Required<BatchingOptions>["itemPrioritizationStrategy"]): import("../types.js").CustomBatchingPrioritizationStrategy;

package/dist/evaluator/LlamaContext/utils/resolveBatchItemsPrioritizationStrategy.js

@@ -0,0 +1,13 @@
+import { maximumParallelismStrategy } from "./batchItemsPrioritizationStrategies/maximumParallelismStrategy.js";
+import { firstInFirstOutStrategy } from "./batchItemsPrioritizationStrategies/firstInFirstOutStrategy.js";
+export function resolveBatchItemsPrioritizationStrategy(strategy) {
+    if (strategy instanceof Function)
+        return strategy;
+    else if (strategy === "maximumParallelism")
+        return maximumParallelismStrategy;
+    else if (strategy === "firstInFirstOut")
+        return firstInFirstOutStrategy;
+    void strategy;
+    throw new Error(`Unknown batch items prioritize strategy: ${strategy}`);
+}
+//# sourceMappingURL=resolveBatchItemsPrioritizationStrategy.js.map

package/dist/evaluator/LlamaContext/utils/resolveBatchItemsPrioritizationStrategy.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"resolveBatchItemsPrioritizationStrategy.js","sourceRoot":"","sources":["../../../../src/evaluator/LlamaContext/utils/resolveBatchItemsPrioritizationStrategy.ts"],"names":[],"mappings":"AACA,OAAO,EAAC,0BAA0B,EAAC,MAAM,oEAAoE,CAAC;AAC9G,OAAO,EAAC,uBAAuB,EAAC,MAAM,iEAAiE,CAAC;AAExG,MAAM,UAAU,uCAAuC,CAAC,QAAiE;IACrH,IAAI,QAAQ,YAAY,QAAQ;QAC5B,OAAO,QAAQ,CAAC;SACf,IAAI,QAAQ,KAAK,oBAAoB;QACtC,OAAO,0BAA0B,CAAC;SACjC,IAAI,QAAQ,KAAK,iBAAiB;QACnC,OAAO,uBAAuB,CAAC;IAEnC,KAAM,QAAyB,CAAC;IAEhC,MAAM,IAAI,KAAK,CAAC,4CAA4C,QAAQ,EAAE,CAAC,CAAC;AAC5E,CAAC"}

package/dist/evaluator/LlamaEmbedding.d.ts

@@ -0,0 +1,21 @@
+export type LlamaEmbeddingOptions = {
+    vector: readonly number[];
+};
+export type LlamaEmbeddingJSON = {
+    type: "embedding";
+    vector: readonly number[];
+};
+export declare class LlamaEmbedding {
+    readonly vector: readonly number[];
+    constructor(options: LlamaEmbeddingOptions);
+    toJSON(): LlamaEmbeddingJSON;
+    /**
+     * Calculates the cosine similarity between this embedding and another embedding.
+     *
+     * Note that you should only compare embeddings created by the exact same model file.
+     * @returns A value between 0 and 1 representing the similarity between the embedding vectors,
+     * where 1 means the embeddings are identical.
+     */
+    calculateCosineSimilarity(other: LlamaEmbedding | LlamaEmbeddingJSON | readonly number[]): number;
+    static fromJSON(json: LlamaEmbeddingJSON): LlamaEmbedding;
+}

package/dist/evaluator/LlamaEmbedding.js

@@ -0,0 +1,53 @@
+export class LlamaEmbedding {
+    vector;
+    constructor(options) {
+        this.vector = Object.freeze(options.vector.slice());
+    }
+    toJSON() {
+        return {
+            type: "embedding",
+            vector: this.vector
+        };
+    }
+    /**
+     * Calculates the cosine similarity between this embedding and another embedding.
+     *
+     * Note that you should only compare embeddings created by the exact same model file.
+     * @returns A value between 0 and 1 representing the similarity between the embedding vectors,
+     * where 1 means the embeddings are identical.
+     */
+    calculateCosineSimilarity(other) {
+        const otherVector = other instanceof Array
+            ? other
+            : other.vector;
+        if (otherVector == null)
+            throw new Error("Other vector is null");
+        else if (otherVector.length !== this.vector.length) {
+            if (otherVector.length === 0 || this.vector.length === 0)
+                return 0;
+            else
+                throw new Error("Vectors have different lengths");
+        }
+        let dotProduct = 0;
+        let thisMagnitude = 0;
+        let otherMagnitude = 0;
+        for (let i = 0; i < this.vector.length; i++) {
+            dotProduct += this.vector[i] * otherVector[i];
+            thisMagnitude += Math.pow(this.vector[i], 2);
+            otherMagnitude += Math.pow(otherVector[i], 2);
+        }
+        if (thisMagnitude === 0 && otherMagnitude === 0)
+            return 1;
+        else if (thisMagnitude === 0 || otherMagnitude === 0)
+            return 0;
+        const thisNorm = Math.sqrt(thisMagnitude);
+        const otherNorm = Math.sqrt(otherMagnitude);
+        return dotProduct / (thisNorm * otherNorm);
+    }
+    static fromJSON(json) {
+        return new LlamaEmbedding({
+            vector: json.vector
+        });
+    }
+}
+//# sourceMappingURL=LlamaEmbedding.js.map

package/dist/evaluator/LlamaEmbedding.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"LlamaEmbedding.js","sourceRoot":"","sources":["../../src/evaluator/LlamaEmbedding.ts"],"names":[],"mappings":"AASA,MAAM,OAAO,cAAc;IACP,MAAM,CAAoB;IAE1C,YAAmB,OAA8B;QAC7C,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC;IACxD,CAAC;IAEM,MAAM;QACT,OAAO;YACH,IAAI,EAAE,WAAW;YACjB,MAAM,EAAE,IAAI,CAAC,MAAM;SACtB,CAAC;IACN,CAAC;IAED;;;;;;OAMG;IACI,yBAAyB,CAAC,KAA8D;QAC3F,MAAM,WAAW,GAAG,KAAK,YAAY,KAAK;YACtC,CAAC,CAAC,KAAK;YACP,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC;QAEnB,IAAI,WAAW,IAAI,IAAI;YACnB,MAAM,IAAI,KAAK,CAAC,sBAAsB,CAAC,CAAC;aACvC,IAAI,WAAW,CAAC,MAAM,KAAK,IAAI,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC;YACjD,IAAI,WAAW,CAAC,MAAM,KAAK,CAAC,IAAI,IAAI,CAAC,MAAM,CAAC,MAAM,KAAK,CAAC;gBACpD,OAAO,CAAC,CAAC;;gBAET,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;QAC1D,CAAC;QAED,IAAI,UAAU,GAAG,CAAC,CAAC;QACnB,IAAI,aAAa,GAAG,CAAC,CAAC;QACtB,IAAI,cAAc,GAAG,CAAC,CAAC;QACvB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YAC1C,UAAU,IAAI,IAAI,CAAC,MAAM,CAAC,CAAC,CAAE,GAAG,WAAW,CAAC,CAAC,CAAE,CAAC;YAChD,aAAa,IAAI,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAE,EAAE,CAAC,CAAC,CAAC;YAC9C,cAAc,IAAI,IAAI,CAAC,GAAG,CAAC,WAAW,CAAC,CAAC,CAAE,EAAE,CAAC,CAAC,CAAC;QACnD,CAAC;QAED,IAAI,aAAa,KAAK,CAAC,IAAI,cAAc,KAAK,CAAC;YAC3C,OAAO,CAAC,CAAC;aACR,IAAI,aAAa,KAAK,CAAC,IAAI,cAAc,KAAK,CAAC;YAChD,OAAO,CAAC,CAAC;QAEb,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;QAC1C,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,CAAC,CAAC;QAE5C,OAAO,UAAU,GAAG,CAAC,QAAQ,GAAG,SAAS,CAAC,CAAC;IAC/C,CAAC;IAEM,MAAM,CAAC,QAAQ,CAAC,IAAwB;QAC3C,OAAO,IAAI,cAAc,CAAC;YACtB,MAAM,EAAE,IAAI,CAAC,MAAM;SACtB,CAAC,CAAC;IACP,CAAC;CACJ"}

package/dist/evaluator/LlamaEmbeddingContext.d.ts

@@ -0,0 +1,52 @@
+import { EventRelay } from "lifecycle-utils";
+import { Token } from "../types.js";
+import { LlamaText } from "../utils/LlamaText.js";
+import { LlamaEmbedding } from "./LlamaEmbedding.js";
+import type { LlamaModel } from "./LlamaModel/LlamaModel.js";
+export type LlamaEmbeddingContextOptions = {
+    /**
+     * The number of tokens the model can see at once.
+     * - **`"auto"`** - adapt to the current VRAM state and attemp to set the context size as high as possible up to the size
+     * the model was trained on.
+     * - **`number`** - set the context size to a specific number of tokens.
+     * If there's not enough VRAM, an error will be thrown.
+     * Use with caution.
+     * - **`{min?: number, max?: number}`** - adapt to the current VRAM state and attemp to set the context size as high as possible
+     * up to the size the model was trained on, but at least `min` and at most `max`.
+     *
+     * Defaults to `"auto"`.
+     */
+    contextSize?: "auto" | number | {
+        min?: number;
+        max?: number;
+    };
+    /** prompt processing batch size */
+    batchSize?: number;
+    /**
+     * number of threads to use to evaluate tokens.
+     * set to 0 to use the maximum threads supported by the current machine hardware
+     */
+    threads?: number;
+    /** An abort signal to abort the context creation */
+    createSignal?: AbortSignal;
+    /**
+     * Ignore insufficient memory errors and continue with the context creation.
+     * Can cause the process to crash if there's not enough VRAM for the new context.
+     *
+     * Defaults to `false`.
+     */
+    ignoreMemorySafetyChecks?: boolean;
+};
+/**
+ * @see [Using Embedding](https://node-llama-cpp.withcat.ai/guide/embedding) tutorial
+ */
+export declare class LlamaEmbeddingContext {
+    readonly onDispose: EventRelay<void>;
+    private constructor();
+    getEmbeddingFor(input: Token[] | string | LlamaText): Promise<LlamaEmbedding>;
+    dispose(): Promise<void>;
+    /** @hidden */
+    [Symbol.asyncDispose](): Promise<void>;
+    get disposed(): boolean;
+    get model(): LlamaModel;
+}

package/dist/evaluator/LlamaEmbeddingContext.js

@@ -0,0 +1,86 @@
+import { AsyncDisposeAggregator, EventRelay, withLock } from "lifecycle-utils";
+import { tokenizeInput } from "../utils/tokenizeInput.js";
+import { resolveBeginningTokenToPrepend, resolveEndTokenToAppend } from "../utils/tokenizerUtils.js";
+import { LlamaEmbedding } from "./LlamaEmbedding.js";
+/**
+ * @see [Using Embedding](https://node-llama-cpp.withcat.ai/guide/embedding) tutorial
+ */
+export class LlamaEmbeddingContext {
+    /** @internal */ _llamaContext;
+    /** @internal */ _sequence;
+    /** @internal */ _disposeAggregator = new AsyncDisposeAggregator();
+    onDispose = new EventRelay();
+    constructor({ _llamaContext }) {
+        this._llamaContext = _llamaContext;
+        this._sequence = this._llamaContext.getSequence();
+        this._disposeAggregator.add(this._llamaContext.onDispose.createListener(() => {
+            void this._disposeAggregator.dispose();
+        }));
+        this._disposeAggregator.add(this.onDispose.dispatchEvent);
+        this._disposeAggregator.add(async () => {
+            await this._llamaContext.dispose();
+        });
+    }
+    async getEmbeddingFor(input) {
+        const resolvedInput = tokenizeInput(input, this._llamaContext.model.tokenizer, undefined, true);
+        if (resolvedInput.length > this._llamaContext.contextSize)
+            throw new Error("Input is longer than the context size. " +
+                "Try to increase the context size or use another model that supports longer contexts.");
+        else if (resolvedInput.length === 0)
+            return new LlamaEmbedding({
+                vector: []
+            });
+        const beginningToken = resolveBeginningTokenToPrepend(this.model.vocabularyType, this.model.tokens);
+        if (beginningToken != null && resolvedInput[0] !== beginningToken)
+            resolvedInput.unshift(beginningToken);
+        const endToken = resolveEndTokenToAppend(this.model.vocabularyType, this.model.tokens);
+        if (endToken != null && resolvedInput.at(-1) !== endToken)
+            resolvedInput.push(endToken);
+        return await withLock([this, "evaluate"], async () => {
+            await this._sequence.eraseContextTokenRanges([{
+                    start: 0,
+                    end: this._sequence.nextTokenIndex
+                }]);
+            const iterator = this._sequence.evaluate(resolvedInput, { _noSampling: true });
+            // eslint-disable-next-line @typescript-eslint/no-unused-vars
+            for await (const token of iterator) {
+                break; // only generate one token to get embeddings
+            }
+            const embedding = this._llamaContext._ctx.getEmbedding(resolvedInput.length);
+            const embeddingVector = Array.from(embedding);
+            return new LlamaEmbedding({
+                vector: embeddingVector
+            });
+        });
+    }
+    async dispose() {
+        await this._disposeAggregator.dispose();
+    }
+    /** @hidden */
+    [Symbol.asyncDispose]() {
+        return this.dispose();
+    }
+    get disposed() {
+        return this._llamaContext.disposed;
+    }
+    get model() {
+        return this._llamaContext.model;
+    }
+    /** @internal */
+    static async _create({ _model }, { contextSize, batchSize, threads = 6, createSignal, ignoreMemorySafetyChecks }) {
+        if (_model.fileInsights.hasEncoder && _model.fileInsights.hasDecoder)
+            throw new Error("Computing embeddings is not supported for encoder-decoder models.");
+        const llamaContext = await _model.createContext({
+            contextSize,
+            batchSize,
+            threads,
+            createSignal,
+            ignoreMemorySafetyChecks,
+            _embeddings: true
+        });
+        return new LlamaEmbeddingContext({
+            _llamaContext: llamaContext
+        });
+    }
+}
+//# sourceMappingURL=LlamaEmbeddingContext.js.map

package/dist/evaluator/LlamaEmbeddingContext.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"LlamaEmbeddingContext.js","sourceRoot":"","sources":["../../src/evaluator/LlamaEmbeddingContext.ts"],"names":[],"mappings":"AAAA,OAAO,EAAC,sBAAsB,EAAE,UAAU,EAAE,QAAQ,EAAC,MAAM,iBAAiB,CAAC;AAG7E,OAAO,EAAC,aAAa,EAAC,MAAM,2BAA2B,CAAC;AACxD,OAAO,EAAC,8BAA8B,EAAE,uBAAuB,EAAC,MAAM,4BAA4B,CAAC;AACnG,OAAO,EAAC,cAAc,EAAC,MAAM,qBAAqB,CAAC;AA2CnD;;GAEG;AACH,MAAM,OAAO,qBAAqB;IAC9B,gBAAgB,CAAkB,aAAa,CAAe;IAC9D,gBAAgB,CAAkB,SAAS,CAAuB;IAClE,gBAAgB,CAAkB,kBAAkB,GAAG,IAAI,sBAAsB,EAAE,CAAC;IAEpE,SAAS,GAAG,IAAI,UAAU,EAAQ,CAAC;IAEnD,YAAoB,EAChB,aAAa,EAGhB;QACG,IAAI,CAAC,aAAa,GAAG,aAAa,CAAC;QACnC,IAAI,CAAC,SAAS,GAAG,IAAI,CAAC,aAAa,CAAC,WAAW,EAAE,CAAC;QAElD,IAAI,CAAC,kBAAkB,CAAC,GAAG,CACvB,IAAI,CAAC,aAAa,CAAC,SAAS,CAAC,cAAc,CAAC,GAAG,EAAE;YAC7C,KAAK,IAAI,CAAC,kBAAkB,CAAC,OAAO,EAAE,CAAC;QAC3C,CAAC,CAAC,CACL,CAAC;QACF,IAAI,CAAC,kBAAkB,CAAC,GAAG,CAAC,IAAI,CAAC,SAAS,CAAC,aAAa,CAAC,CAAC;QAC1D,IAAI,CAAC,kBAAkB,CAAC,GAAG,CAAC,KAAK,IAAI,EAAE;YACnC,MAAM,IAAI,CAAC,aAAa,CAAC,OAAO,EAAE,CAAC;QACvC,CAAC,CAAC,CAAC;IACP,CAAC;IAEM,KAAK,CAAC,eAAe,CAAC,KAAmC;QAC5D,MAAM,aAAa,GAAG,aAAa,CAAC,KAAK,EAAE,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,SAAS,EAAE,SAAS,EAAE,IAAI,CAAC,CAAC;QAEhG,IAAI,aAAa,CAAC,MAAM,GAAG,IAAI,CAAC,aAAa,CAAC,WAAW;YACrD,MAAM,IAAI,KAAK,CACX,yCAAyC;gBACzC,sFAAsF,CACzF,CAAC;aACD,IAAI,aAAa,CAAC,MAAM,KAAK,CAAC;YAC/B,OAAO,IAAI,cAAc,CAAC;gBACtB,MAAM,EAAE,EAAE;aACb,CAAC,CAAC;QAEP,MAAM,cAAc,GAAG,8BAA8B,CAAC,IAAI,CAAC,KAAK,CAAC,cAAc,EAAE,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;QACpG,IAAI,cAAc,IAAI,IAAI,IAAI,aAAa,CAAC,CAAC,CAAC,KAAK,cAAc;YAC7D,aAAa,CAAC,OAAO,CAAC,cAAc,CAAC,CAAC;QAE1C,MAAM,QAAQ,GAAG,uBAAuB,CAAC,IAAI,CAAC,KAAK,CAAC,cAAc,EAAE,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;QACvF,IAAI,QAAQ,IAAI,IAAI,IAAI,aAAa,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,KAAK,QAAQ;YACrD,aAAa,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;QAEjC,OAAO,MAAM,QAAQ,CAAC,CAAC,IAA6B,EAAE,UAAU,CAAC,EAAE,KAAK,IAAI,EAAE;YAC1E,MAAM,IAAI,CAAC,SAAS,CAAC,uBAAuB,CAAC,CAAC;oBAC1C,KAAK,EAAE,CAAC;oBACR,GAAG,EAAE,IAAI,CAAC,SAAS,CAAC,cAAc;iBACrC,CAAC,CAAC,CAAC;YAEJ,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,aAAa,EAAE,EAAC,WAAW,EAAE,IAAI,EAAC,CAAC,CAAC;YAC7E,6DAA6D;YAC7D,IAAI,KAAK,EAAE,MAAM,KAAK,IAAI,QAAQ,EAAE,CAAC;gBACjC,MAAM,CAAC,4CAA4C;YACvD,CAAC;YAED,MAAM,SAAS,GAAG,IAAI,CAAC,aAAa,CAAC,IAAI,CAAC,YAAY,CAAC,aAAa,CAAC,MAAM,CAAC,CAAC;YAC7E,MAAM,eAAe,GAAG,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;YAE9C,OAAO,IAAI,cAAc,CAAC;gBACtB,MAAM,EAAE,eAAe;aAC1B,CAAC,CAAC;QACP,CAAC,CAAC,CAAC;IACP,CAAC;IAEM,KAAK,CAAC,OAAO;QAChB,MAAM,IAAI,CAAC,kBAAkB,CAAC,OAAO,EAAE,CAAC;IAC5C,CAAC;IAED,cAAc;IACP,CAAC,MAAM,CAAC,YAAY,CAAC;QACxB,OAAO,IAAI,CAAC,OAAO,EAAE,CAAC;IAC1B,CAAC;IAED,IAAW,QAAQ;QACf,OAAO,IAAI,CAAC,aAAa,CAAC,QAAQ,CAAC;IACvC,CAAC;IAED,IAAW,KAAK;QACZ,OAAO,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC;IACpC,CAAC;IAED,gBAAgB;IACT,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,EACxB,MAAM,EAGT,EAAE,EACC,WAAW,EACX,SAAS,EACT,OAAO,GAAG,CAAC,EACX,YAAY,EACZ,wBAAwB,EACG;QAC3B,IAAI,MAAM,CAAC,YAAY,CAAC,UAAU,IAAI,MAAM,CAAC,YAAY,CAAC,UAAU;YAChE,MAAM,IAAI,KAAK,CAAC,mEAAmE,CAAC,CAAC;QAEzF,MAAM,YAAY,GAAG,MAAM,MAAM,CAAC,aAAa,CAAC;YAC5C,WAAW;YACX,SAAS;YACT,OAAO;YACP,YAAY;YACZ,wBAAwB;YACxB,WAAW,EAAE,IAAI;SACpB,CAAC,CAAC;QAEH,OAAO,IAAI,qBAAqB,CAAC;YAC7B,aAAa,EAAE,YAAY;SAC9B,CAAC,CAAC;IACP,CAAC;CACJ"}

package/dist/evaluator/LlamaGrammar.d.ts

@@ -0,0 +1,39 @@
+import { LlamaText } from "../utils/LlamaText.js";
+import { Llama } from "../bindings/Llama.js";
+import { Token } from "../types.js";
+export type LlamaGrammarOptions = {
+    /** GBNF grammar */
+    grammar: string;
+    /** Consider any of these as EOS for the generated text. Only supported by `LlamaChat` and `LlamaChatSession` */
+    stopGenerationTriggers?: readonly (LlamaText | string | readonly (string | Token)[])[];
+    /** Trim whitespace from the end of the generated text. Only supported by `LlamaChat` and `LlamaChatSession` */
+    trimWhitespaceSuffix?: boolean;
+    /**
+     * Root rule name.
+     *
+     * Defaults to `"root"`.
+     */
+    rootRuleName?: string;
+};
+/**
+ * @see [Using Grammar](https://node-llama-cpp.withcat.ai/guide/grammar) tutorial
+ */
+export declare class LlamaGrammar {
+    /**
+     * > GBNF files are supported.
+     * > More info here: [
+     * github:ggml-org/llama.cpp:grammars/README.md
+     * ](https://github.com/ggml-org/llama.cpp/blob/f5fe98d11bdf9e7797bcfb05c0c3601ffc4b9d26/grammars/README.md)
+     *
+     * Prefer to create a new instance of this class by using `llama.createGrammar(...)`.
+     * @deprecated Use `llama.createGrammar(...)` instead.
+     * @param llama
+     * @param options
+     */
+    constructor(llama: Llama, { grammar, stopGenerationTriggers, trimWhitespaceSuffix, rootRuleName }: LlamaGrammarOptions);
+    get grammar(): string;
+    get rootRuleName(): string;
+    get stopGenerationTriggers(): readonly (string | import("../utils/LlamaText.js")._LlamaText | readonly (string | Token)[])[];
+    get trimWhitespaceSuffix(): boolean;
+    static getFor(llama: Llama, type: "json" | "json_arr" | "english" | "list" | "c" | "arithmetic" | "japanese" | "chess"): Promise<LlamaGrammar>;
+}

package/dist/evaluator/LlamaGrammar.js

@@ -0,0 +1,72 @@
+import path from "path";
+import fs from "fs-extra";
+import { getGrammarsFolder } from "../utils/getGrammarsFolder.js";
+import { LlamaText } from "../utils/LlamaText.js";
+/**
+ * @see [Using Grammar](https://node-llama-cpp.withcat.ai/guide/grammar) tutorial
+ */
+export class LlamaGrammar {
+    /** @internal */ _llama;
+    /** @internal */ _grammar;
+    /** @internal */ _stopGenerationTriggers;
+    /** @internal */ _trimWhitespaceSuffix;
+    /** @internal */ _grammarText;
+    /** @internal */ _rootRuleName;
+    /**
+     * > GBNF files are supported.
+     * > More info here: [
+     * github:ggml-org/llama.cpp:grammars/README.md
+     * ](https://github.com/ggml-org/llama.cpp/blob/f5fe98d11bdf9e7797bcfb05c0c3601ffc4b9d26/grammars/README.md)
+     *
+     * Prefer to create a new instance of this class by using `llama.createGrammar(...)`.
+     * @deprecated Use `llama.createGrammar(...)` instead.
+     * @param llama
+     * @param options
+     */
+    constructor(llama, { grammar, stopGenerationTriggers = [], trimWhitespaceSuffix = false, rootRuleName = "root" }) {
+        this._llama = llama;
+        this._grammar = new this._llama._bindings.AddonGrammar(grammar, {
+            addonExports: this._llama._bindings,
+            rootRuleName
+        });
+        this._stopGenerationTriggers = stopGenerationTriggers ?? [];
+        this._trimWhitespaceSuffix = trimWhitespaceSuffix;
+        this._grammarText = grammar;
+        this._rootRuleName = rootRuleName;
+    }
+    get grammar() {
+        return this._grammarText;
+    }
+    get rootRuleName() {
+        return this._rootRuleName;
+    }
+    get stopGenerationTriggers() {
+        return this._stopGenerationTriggers;
+    }
+    get trimWhitespaceSuffix() {
+        return this._trimWhitespaceSuffix;
+    }
+    /**
+     * Test if the given text is compatible with the grammar.
+     * @internal
+     */
+    _testText(text) {
+        return this._grammar.isTextCompatible(String(text));
+    }
+    static async getFor(llama, type) {
+        const grammarsFolder = await getGrammarsFolder(llama.buildType);
+        const grammarFile = path.join(grammarsFolder, type + ".gbnf");
+        if (await fs.pathExists(grammarFile)) {
+            const grammar = await fs.readFile(grammarFile, "utf8");
+            return new LlamaGrammar(llama, {
+                grammar,
+                stopGenerationTriggers: [LlamaText(["\n".repeat((type === "json" || type === "json_arr")
+                            ? 4
+                            : 10)])], // this is a workaround for the model not stopping to generate text,
+                trimWhitespaceSuffix: true
+            });
+        }
+        throw new Error(`Grammar file for type "${type}" was not found in "${grammarsFolder}"`);
+    }
+}
+//# sourceMappingURL=LlamaGrammar.js.map

package/dist/evaluator/LlamaGrammar.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"LlamaGrammar.js","sourceRoot":"","sources":["../../src/evaluator/LlamaGrammar.ts"],"names":[],"mappings":"AAAA,OAAO,IAAI,MAAM,MAAM,CAAC;AACxB,OAAO,EAAE,MAAM,UAAU,CAAC;AAC1B,OAAO,EAAC,iBAAiB,EAAC,MAAM,+BAA+B,CAAC;AAChE,OAAO,EAAC,SAAS,EAAC,MAAM,uBAAuB,CAAC;AAwBhD;;GAEG;AACH,MAAM,OAAO,YAAY;IACrB,gBAAgB,CAAiB,MAAM,CAAQ;IAC/C,gBAAgB,CAAiB,QAAQ,CAAe;IACxD,gBAAgB,CAAkB,uBAAuB,CAAgE;IACzH,gBAAgB,CAAkB,qBAAqB,CAAU;IACjE,gBAAgB,CAAkB,YAAY,CAAS;IACvD,gBAAgB,CAAkB,aAAa,CAAS;IAExD;;;;;;;;;;OAUG;IACH,YAAmB,KAAY,EAAE,EAC7B,OAAO,EAAE,sBAAsB,GAAG,EAAE,EAAE,oBAAoB,GAAG,KAAK,EAAE,YAAY,GAAG,MAAM,EACvE;QAClB,IAAI,CAAC,MAAM,GAAG,KAAK,CAAC;QACpB,IAAI,CAAC,QAAQ,GAAG,IAAI,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,YAAY,CAAC,OAAO,EAAE;YAC5D,YAAY,EAAE,IAAI,CAAC,MAAM,CAAC,SAAS;YACnC,YAAY;SACf,CAAC,CAAC;QACH,IAAI,CAAC,uBAAuB,GAAG,sBAAsB,IAAI,EAAE,CAAC;QAC5D,IAAI,CAAC,qBAAqB,GAAG,oBAAoB,CAAC;QAClD,IAAI,CAAC,YAAY,GAAG,OAAO,CAAC;QAC5B,IAAI,CAAC,aAAa,GAAG,YAAY,CAAC;IACtC,CAAC;IAED,IAAW,OAAO;QACd,OAAO,IAAI,CAAC,YAAY,CAAC;IAC7B,CAAC;IAED,IAAW,YAAY;QACnB,OAAO,IAAI,CAAC,aAAa,CAAC;IAC9B,CAAC;IAED,IAAW,sBAAsB;QAC7B,OAAO,IAAI,CAAC,uBAAuB,CAAC;IACxC,CAAC;IAED,IAAW,oBAAoB;QAC3B,OAAO,IAAI,CAAC,qBAAqB,CAAC;IACtC,CAAC;IAED;;;OAGG;IACI,SAAS,CAAC,IAAY;QACzB,OAAO,IAAI,CAAC,QAAQ,CAAC,gBAAgB,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC;IACxD,CAAC;IAEM,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,KAAY,EAAE,IAA0F;QAC/H,MAAM,cAAc,GAAG,MAAM,iBAAiB,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;QAEhE,MAAM,WAAW,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,IAAI,GAAG,OAAO,CAAC,CAAC;QAE9D,IAAI,MAAM,EAAE,CAAC,UAAU,CAAC,WAAW,CAAC,EAAE,CAAC;YACnC,MAAM,OAAO,GAAG,MAAM,EAAE,CAAC,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC,CAAC;YACvD,OAAO,IAAI,YAAY,CAAC,KAAK,EAAE;gBAC3B,OAAO;gBACP,sBAAsB,EAAE,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,MAAM,CAC3C,CAAC,IAAI,KAAK,MAAM,IAAI,IAAI,KAAK,UAAU,CAAC;4BACpC,CAAC,CAAC,CAAC;4BACH,CAAC,CAAC,EAAE,CACX,CAAC,CAAC,CAAC,EAAE,oEAAoE;gBAC1E,oBAAoB,EAAE,IAAI;aAC7B,CAAC,CAAC;QACP,CAAC;QAED,MAAM,IAAI,KAAK,CAAC,0BAA0B,IAAI,uBAAuB,cAAc,GAAG,CAAC,CAAC;IAC5F,CAAC;CACJ"}

package/dist/evaluator/LlamaGrammarEvaluationState.d.ts

@@ -0,0 +1,19 @@
+import type { LlamaGrammar } from "./LlamaGrammar.js";
+import type { LlamaModel } from "./LlamaModel/LlamaModel.js";
+export type LlamaGrammarEvaluationStateOptions = {
+    model: LlamaModel;
+    grammar: LlamaGrammar;
+};
+/**
+ * Grammar evaluation state is used to track the model response to determine the next allowed characters for the model to generate.
+ *
+ * Create a new grammar evaluation state for every response you generate with the model.
+ *
+ * This is only needed when using the `LlamaContext` class directly, since `LlamaChatSession` already handles this for you.
+ */
+export declare class LlamaGrammarEvaluationState {
+    constructor(options: LlamaGrammarEvaluationStateOptions);
+    constructor(existingState: LlamaGrammarEvaluationState);
+    /** Clone the grammar evaluation state */
+    clone(): LlamaGrammarEvaluationState;
+}

package/dist/evaluator/LlamaGrammarEvaluationState.js

@@ -0,0 +1,29 @@
+/**
+ * Grammar evaluation state is used to track the model response to determine the next allowed characters for the model to generate.
+ *
+ * Create a new grammar evaluation state for every response you generate with the model.
+ *
+ * This is only needed when using the `LlamaContext` class directly, since `LlamaChatSession` already handles this for you.
+ */
+export class LlamaGrammarEvaluationState {
+    /** @internal */ _llama;
+    /** @internal */ _state;
+    constructor(existingStateOrOptions) {
+        if (existingStateOrOptions instanceof LlamaGrammarEvaluationState) {
+            this._llama = existingStateOrOptions._llama;
+            this._state = new this._llama._bindings.AddonGrammarEvaluationState(existingStateOrOptions._state);
+        }
+        else {
+            const { model, grammar } = existingStateOrOptions;
+            this._llama = model._llama;
+            if (model._llama !== grammar._llama)
+                throw new Error("The given LlamaModel and LlamaGrammar must be from the same Llama instance");
+            this._state = new model._llama._bindings.AddonGrammarEvaluationState(model._model, grammar._grammar);
+        }
+    }
+    /** Clone the grammar evaluation state */
+    clone() {
+        return new LlamaGrammarEvaluationState(this);
+    }
+}
+//# sourceMappingURL=LlamaGrammarEvaluationState.js.map

package/dist/evaluator/LlamaGrammarEvaluationState.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"LlamaGrammarEvaluationState.js","sourceRoot":"","sources":["../../src/evaluator/LlamaGrammarEvaluationState.ts"],"names":[],"mappings":"AAWA;;;;;;GAMG;AACH,MAAM,OAAO,2BAA2B;IACpC,gBAAgB,CAAiB,MAAM,CAAQ;IAC/C,gBAAgB,CAAiB,MAAM,CAA8B;IAIrE,YAAmB,sBAAwF;QACvG,IAAI,sBAAsB,YAAY,2BAA2B,EAAE,CAAC;YAChE,IAAI,CAAC,MAAM,GAAG,sBAAsB,CAAC,MAAM,CAAC;YAC5C,IAAI,CAAC,MAAM,GAAG,IAAI,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,2BAA2B,CAAC,sBAAsB,CAAC,MAAM,CAAC,CAAC;QACvG,CAAC;aAAM,CAAC;YACJ,MAAM,EAAC,KAAK,EAAE,OAAO,EAAC,GAAG,sBAAsB,CAAC;YAChD,IAAI,CAAC,MAAM,GAAG,KAAK,CAAC,MAAM,CAAC;YAE3B,IAAI,KAAK,CAAC,MAAM,KAAK,OAAO,CAAC,MAAM;gBAC/B,MAAM,IAAI,KAAK,CAAC,4EAA4E,CAAC,CAAC;YAElG,IAAI,CAAC,MAAM,GAAG,IAAI,KAAK,CAAC,MAAM,CAAC,SAAS,CAAC,2BAA2B,CAAC,KAAK,CAAC,MAAM,EAAE,OAAO,CAAC,QAAQ,CAAC,CAAC;QACzG,CAAC;IACL,CAAC;IAED,yCAAyC;IAClC,KAAK;QACR,OAAO,IAAI,2BAA2B,CAAC,IAAI,CAAC,CAAC;IACjD,CAAC;CACJ"}

package/dist/evaluator/LlamaJsonSchemaGrammar.d.ts

@@ -0,0 +1,17 @@
+import { GbnfJsonDefList, GbnfJsonSchema, GbnfJsonSchemaToType } from "../utils/gbnfJson/types.js";
+import { Llama } from "../bindings/Llama.js";
+import { LlamaGrammar } from "./LlamaGrammar.js";
+/**
+ * @see [Using a JSON Schema Grammar](https://node-llama-cpp.withcat.ai/guide/grammar#json-schema) tutorial
+ * @see [Reducing Hallucinations When Using JSON Schema Grammar](https://node-llama-cpp.withcat.ai/guide/grammar#reducing-json-schema-hallucinations) tutorial
+ */
+export declare class LlamaJsonSchemaGrammar<const T extends GbnfJsonSchema<Defs>, const Defs extends GbnfJsonDefList<Defs> = Record<any, any>> extends LlamaGrammar {
+    private readonly _schema;
+    /**
+     * Prefer to create a new instance of this class by using `llama.createGrammarForJsonSchema(...)`.
+     * @deprecated Use `llama.createGrammarForJsonSchema(...)` instead.
+     */
+    constructor(llama: Llama, schema: Readonly<T> & GbnfJsonSchema<Defs>);
+    get schema(): Readonly<T>;
+    parse(json: string): GbnfJsonSchemaToType<T>;
+}

package/dist/evaluator/LlamaJsonSchemaGrammar.js

@@ -0,0 +1,35 @@
+import { getGbnfGrammarForGbnfJsonSchema } from "../utils/gbnfJson/getGbnfGrammarForGbnfJsonSchema.js";
+import { validateObjectAgainstGbnfSchema } from "../utils/gbnfJson/utils/validateObjectAgainstGbnfSchema.js";
+import { LlamaText } from "../utils/LlamaText.js";
+import { LlamaGrammar } from "./LlamaGrammar.js";
+/* eslint-disable @stylistic/max-len */
+/**
+ * @see [Using a JSON Schema Grammar](https://node-llama-cpp.withcat.ai/guide/grammar#json-schema) tutorial
+ * @see [Reducing Hallucinations When Using JSON Schema Grammar](https://node-llama-cpp.withcat.ai/guide/grammar#reducing-json-schema-hallucinations) tutorial
+ */
+export class LlamaJsonSchemaGrammar extends LlamaGrammar {
+    _schema;
+    /**
+     * Prefer to create a new instance of this class by using `llama.createGrammarForJsonSchema(...)`.
+     * @deprecated Use `llama.createGrammarForJsonSchema(...)` instead.
+     */
+    constructor(llama, schema) {
+        const grammar = getGbnfGrammarForGbnfJsonSchema(schema);
+        super(llama, {
+            grammar,
+            stopGenerationTriggers: [LlamaText(["\n".repeat(4)])],
+            trimWhitespaceSuffix: true
+        });
+        this._schema = schema;
+    }
+    get schema() {
+        return this._schema;
+    }
+    parse(json) {
+        const parsedJson = JSON.parse(json);
+        validateObjectAgainstGbnfSchema(parsedJson, this._schema);
+        return parsedJson;
+    }
+}
+/* eslint-enable @stylistic/max-len */
+//# sourceMappingURL=LlamaJsonSchemaGrammar.js.map

package/dist/evaluator/LlamaJsonSchemaGrammar.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"LlamaJsonSchemaGrammar.js","sourceRoot":"","sources":["../../src/evaluator/LlamaJsonSchemaGrammar.ts"],"names":[],"mappings":"AACA,OAAO,EAAC,+BAA+B,EAAC,MAAM,sDAAsD,CAAC;AACrG,OAAO,EAAC,+BAA+B,EAAC,MAAM,4DAA4D,CAAC;AAC3G,OAAO,EAAC,SAAS,EAAC,MAAM,uBAAuB,CAAC;AAEhD,OAAO,EAAC,YAAY,EAAC,MAAM,mBAAmB,CAAC;AAE/C,uCAAuC;AACvC;;;GAGG;AACH,MAAM,OAAO,sBAGX,SAAQ,YAAY;IACD,OAAO,CAAI;IAE5B;;;OAGG;IACH,YAAmB,KAAY,EAAE,MAA0C;QACvE,MAAM,OAAO,GAAG,+BAA+B,CAAC,MAAM,CAAC,CAAC;QAExD,KAAK,CAAC,KAAK,EAAE;YACT,OAAO;YACP,sBAAsB,EAAE,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YACrD,oBAAoB,EAAE,IAAI;SAC7B,CAAC,CAAC;QAEH,IAAI,CAAC,OAAO,GAAG,MAAM,CAAC;IAC1B,CAAC;IAED,IAAW,MAAM;QACb,OAAO,IAAI,CAAC,OAAO,CAAC;IACxB,CAAC;IAEM,KAAK,CAAC,IAAY;QACrB,MAAM,UAAU,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;QAEpC,+BAA+B,CAAC,UAAU,EAAE,IAAI,CAAC,OAAO,CAAC,CAAC;QAE1D,OAAO,UAAU,CAAC;IACtB,CAAC;CACJ;AACD,sCAAsC"}