node-llama-cpp 2.8.6 → 3.0.0-beta.10

package/dist/chatWrappers/generateContextTextFromConversationHistory.d.ts

@@ -1,23 +0,0 @@
-import { ChatPromptWrapper } from "../ChatPromptWrapper.js";
-import { ConversationInteraction } from "../types.js";
-/**
- * Generate context text to load into a model context from a conversation history.
- * @param {ChatPromptWrapper} chatPromptWrapper
- * @param {ConversationInteraction[]} conversationHistory
- * @param {object} [options]
- * @param {string} [options.systemPrompt]
- * @param {number} [options.currentPromptIndex]
- * @param {string | null} [options.lastStopString]
- * @param {string | null} [options.lastStopStringSuffix]
- * @returns {{text: string, stopString: (string | null), stopStringSuffix: (string | null)}}
- */
-export declare function generateContextTextFromConversationHistory(chatPromptWrapper: ChatPromptWrapper, conversationHistory: readonly ConversationInteraction[], { systemPrompt, currentPromptIndex, lastStopString, lastStopStringSuffix }?: {
-    systemPrompt?: string;
-    currentPromptIndex?: number;
-    lastStopString?: string | null;
-    lastStopStringSuffix?: string | null;
-}): {
-    text: string;
-    stopString: string | null;
-    stopStringSuffix: string | null;
-};

package/dist/chatWrappers/generateContextTextFromConversationHistory.js

@@ -1,47 +0,0 @@
-import { defaultChatSystemPrompt } from "../config.js";
-/**
- * Generate context text to load into a model context from a conversation history.
- * @param {ChatPromptWrapper} chatPromptWrapper
- * @param {ConversationInteraction[]} conversationHistory
- * @param {object} [options]
- * @param {string} [options.systemPrompt]
- * @param {number} [options.currentPromptIndex]
- * @param {string | null} [options.lastStopString]
- * @param {string | null} [options.lastStopStringSuffix]
- * @returns {{text: string, stopString: (string | null), stopStringSuffix: (string | null)}}
- */
-export function generateContextTextFromConversationHistory(chatPromptWrapper, conversationHistory, { systemPrompt = defaultChatSystemPrompt, currentPromptIndex = 0, lastStopString = null, lastStopStringSuffix = null } = {}) {
-    let res = "";
-    for (let i = 0; i < conversationHistory.length; i++) {
-        const interaction = conversationHistory[i];
-        const wrappedPrompt = chatPromptWrapper.wrapPrompt(interaction.prompt, {
-            systemPrompt,
-            promptIndex: currentPromptIndex,
-            lastStopString,
-            lastStopStringSuffix
-        });
-        const stopStrings = chatPromptWrapper.getStopStrings();
-        const defaultStopString = chatPromptWrapper.getDefaultStopString();
-        const stopStringsToCheckInResponse = new Set([...stopStrings, defaultStopString]);
-        currentPromptIndex++;
-        lastStopString = null;
-        lastStopStringSuffix = null;
-        res += wrappedPrompt;
-        for (const stopString of stopStringsToCheckInResponse) {
-            if (interaction.response.includes(stopString)) {
-                console.error(`Stop string "${stopString}" was found in model response of conversation interaction index ${i}`, { interaction, stopString });
-                throw new Error("A stop string cannot be in a conversation history interaction model response");
-            }
-        }
-        res += interaction.response;
-        res += defaultStopString;
-        lastStopString = defaultStopString;
-        lastStopStringSuffix = "";
-    }
-    return {
-        text: res,
-        stopString: lastStopString,
-        stopStringSuffix: lastStopStringSuffix
-    };
-}
-//# sourceMappingURL=generateContextTextFromConversationHistory.js.map

package/dist/chatWrappers/generateContextTextFromConversationHistory.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"generateContextTextFromConversationHistory.js","sourceRoot":"","sources":["../../src/chatWrappers/generateContextTextFromConversationHistory.ts"],"names":[],"mappings":"AACA,OAAO,EAAC,uBAAuB,EAAC,MAAM,cAAc,CAAC;AAIrD;;;;;;;;;;GAUG;AACH,MAAM,UAAU,0CAA0C,CACtD,iBAAoC,EACpC,mBAAuD,EACvD,EACI,YAAY,GAAG,uBAAuB,EAAE,kBAAkB,GAAG,CAAC,EAAE,cAAc,GAAG,IAAI,EAAE,oBAAoB,GAAG,IAAI,KAGlH,EAAE;IAMN,IAAI,GAAG,GAAG,EAAE,CAAC;IAEb,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,mBAAmB,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE;QACjD,MAAM,WAAW,GAAG,mBAAmB,CAAC,CAAC,CAAC,CAAC;QAC3C,MAAM,aAAa,GAAG,iBAAiB,CAAC,UAAU,CAAC,WAAW,CAAC,MAAM,EAAE;YACnE,YAAY;YACZ,WAAW,EAAE,kBAAkB;YAC/B,cAAc;YACd,oBAAoB;SACvB,CAAC,CAAC;QACH,MAAM,WAAW,GAAG,iBAAiB,CAAC,cAAc,EAAE,CAAC;QACvD,MAAM,iBAAiB,GAAG,iBAAiB,CAAC,oBAAoB,EAAE,CAAC;QACnE,MAAM,4BAA4B,GAAG,IAAI,GAAG,CAAC,CAAC,GAAG,WAAW,EAAE,iBAAiB,CAAC,CAAC,CAAC;QAElF,kBAAkB,EAAE,CAAC;QACrB,cAAc,GAAG,IAAI,CAAC;QACtB,oBAAoB,GAAG,IAAI,CAAC;QAE5B,GAAG,IAAI,aAAa,CAAC;QAErB,KAAK,MAAM,UAAU,IAAI,4BAA4B,EAAE;YACnD,IAAI,WAAW,CAAC,QAAQ,CAAC,QAAQ,CAAC,UAAU,CAAC,EAAE;gBAC3C,OAAO,CAAC,KAAK,CACT,gBAAgB,UAAU,mEAAmE,CAAC,EAAE,EAChG,EAAC,WAAW,EAAE,UAAU,EAAC,CAC5B,CAAC;gBACF,MAAM,IAAI,KAAK,CAAC,8EAA8E,CAAC,CAAC;aACnG;SACJ;QAED,GAAG,IAAI,WAAW,CAAC,QAAQ,CAAC;QAC5B,GAAG,IAAI,iBAAiB,CAAC;QACzB,cAAc,GAAG,iBAAiB,CAAC;QACnC,oBAAoB,GAAG,EAAE,CAAC;KAC7B;IAED,OAAO;QACH,IAAI,EAAE,GAAG;QACT,UAAU,EAAE,cAAc;QAC1B,gBAAgB,EAAE,oBAAoB;KACzC,CAAC;AACN,CAAC"}

package/dist/llamaEvaluator/LlamaBins.d.ts

@@ -1,4 +0,0 @@
-import { type LLAMAModel, type LLAMAContext, type LLAMAGrammar, type LLAMAGrammarEvaluationState } from "../utils/getBin.js";
-export declare const llamaCppNode: import("../utils/getBin.js").LlamaCppNodeModule;
-declare const LLAMAModel: LLAMAModel, LLAMAContext: LLAMAContext, LLAMAGrammar: LLAMAGrammar, LLAMAGrammarEvaluationState: LLAMAGrammarEvaluationState;
-export { LLAMAModel, LLAMAContext, LLAMAGrammar, LLAMAGrammarEvaluationState };

package/dist/llamaEvaluator/LlamaBins.js

@@ -1,5 +0,0 @@
-import { loadBin } from "../utils/getBin.js";
-export const llamaCppNode = await loadBin();
-const { LLAMAModel, LLAMAContext, LLAMAGrammar, LLAMAGrammarEvaluationState } = llamaCppNode;
-export { LLAMAModel, LLAMAContext, LLAMAGrammar, LLAMAGrammarEvaluationState };
-//# sourceMappingURL=LlamaBins.js.map

package/dist/llamaEvaluator/LlamaBins.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"LlamaBins.js","sourceRoot":"","sources":["../../src/llamaEvaluator/LlamaBins.ts"],"names":[],"mappings":"AAAA,OAAO,EAAC,OAAO,EAA0F,MAAM,oBAAoB,CAAC;AAEpI,MAAM,CAAC,MAAM,YAAY,GAAG,MAAM,OAAO,EAAE,CAAC;AAC5C,MAAM,EAAC,UAAU,EAAE,YAAY,EAAE,YAAY,EAAE,2BAA2B,EAAC,GAAG,YAAY,CAAC;AAE3F,OAAO,EAAC,UAAU,EAAE,YAAY,EAAE,YAAY,EAAE,2BAA2B,EAAC,CAAC"}

package/dist/llamaEvaluator/LlamaChatSession.d.ts

@@ -1,122 +0,0 @@
-/// <reference types="node" />
-import { ChatPromptWrapper } from "../ChatPromptWrapper.js";
-import { ConversationInteraction, Token } from "../types.js";
-import { LlamaContext } from "./LlamaContext.js";
-import { LlamaGrammar } from "./LlamaGrammar.js";
-export type LlamaChatSessionOptions = {
-    context: LlamaContext;
-    printLLamaSystemInfo?: boolean;
-    /** GeneralChatPromptWrapper is ued by default */
-    promptWrapper?: ChatPromptWrapper | "auto";
-    systemPrompt?: string;
-    /** Conversation history to load into the context to continue an existing conversation */
-    conversationHistory?: readonly ConversationInteraction[];
-};
-export type LLamaChatPromptOptions = {
-    onToken?: (tokens: Token[]) => void;
-    signal?: AbortSignal;
-    maxTokens?: number;
-    /**
-     * Temperature is a hyperparameter that controls the randomness of the generated text.
-     * It affects the probability distribution of the model's output tokens.
-     * A higher temperature (e.g., 1.5) makes the output more random and creative,
-     * while a lower temperature (e.g., 0.5) makes the output more focused, deterministic, and conservative.
-     * The suggested temperature is 0.8, which provides a balance between randomness and determinism.
-     * At the extreme, a temperature of 0 will always pick the most likely next token, leading to identical outputs in each run.
-     *
-     * Set to `0` to disable.
-     * Disabled by default (set to `0`).
-     */
-    temperature?: number;
-    /**
-     * Limits the model to consider only the K most likely next tokens for sampling at each step of sequence generation.
-     * An integer number between `1` and the size of the vocabulary.
-     * Set to `0` to disable (which uses the full vocabulary).
-     *
-     * Only relevant when `temperature` is set to a value greater than 0.
-     */
-    topK?: number;
-    /**
-     * Dynamically selects the smallest set of tokens whose cumulative probability exceeds the threshold P,
-     * and samples the next token only from this set.
-     * A float number between `0` and `1`.
-     * Set to `1` to disable.
-     *
-     * Only relevant when `temperature` is set to a value greater than `0`.
-     */
-    topP?: number;
-    grammar?: LlamaGrammar;
-    /**
-     * Trim whitespace from the end of the generated text
-     * Disabled by default.
-     */
-    trimWhitespaceSuffix?: boolean;
-    repeatPenalty?: false | LlamaChatSessionRepeatPenalty;
-};
-export type LlamaChatSessionRepeatPenalty = {
-    /**
-     * Number of recent tokens generated by the model to apply penalties to repetition of.
-     * Defaults to `64`.
-     */
-    lastTokens?: number;
-    punishTokensFilter?: (tokens: Token[]) => Token[];
-    /**
-     * Penalize new line tokens.
-     * Enabled by default.
-     */
-    penalizeNewLine?: boolean;
-    /**
-     * The relative amount to lower the probability of the tokens in `punishTokens` by
-     * Defaults to `1.1`.
-     * Set to `1` to disable.
-     */
-    penalty?: number;
-    /**
-     * For n time a token is in the `punishTokens` array, lower its probability by `n * frequencyPenalty`
-     * Disabled by default (`0`).
-     * Set to a value between `0` and `1` to enable.
-     */
-    frequencyPenalty?: number;
-    /**
-     * Lower the probability of all the tokens in the `punishTokens` array by `presencePenalty`
-     * Disabled by default (`0`).
-     * Set to a value between `0` and `1` to enable.
-     */
-    presencePenalty?: number;
-};
-export declare class LlamaChatSession {
-    private readonly _systemPrompt;
-    private readonly _printLLamaSystemInfo;
-    private readonly _promptWrapper;
-    private _promptIndex;
-    private _initialized;
-    private _lastStopString;
-    private _lastStopStringSuffix;
-    private _conversationHistoryToLoad;
-    private readonly _ctx;
-    /**
-     * @param {LlamaChatSessionOptions} options
-     */
-    constructor({ context, printLLamaSystemInfo, promptWrapper, systemPrompt, conversationHistory }: LlamaChatSessionOptions);
-    get initialized(): boolean;
-    get context(): LlamaContext;
-    init(): Promise<void>;
-    /**
-     * @param {string} prompt
-     * @param {object} options
-     * @returns {Promise<string>}
-     */
-    prompt(prompt: string, { onToken, signal, maxTokens, temperature, topK, topP, grammar, trimWhitespaceSuffix, repeatPenalty }?: LLamaChatPromptOptions): Promise<string>;
-    /**
-     * @param {string} prompt
-     * @param {LLamaChatPromptOptions} options
-     */
-    promptWithMeta(prompt: string, { onToken, signal, maxTokens, temperature, topK, topP, grammar, trimWhitespaceSuffix, repeatPenalty }?: LLamaChatPromptOptions): Promise<{
-        text: string;
-        stopReason: "maxTokens" | "eosToken" | "stopString";
-        stopString: string | null;
-        stopStringSuffix: string | null;
-    }>;
-    private _evalTokens;
-    private _checkStopString;
-}

package/dist/llamaEvaluator/LlamaChatSession.js

@@ -1,236 +0,0 @@
-import { defaultChatSystemPrompt } from "../config.js";
-import { withLock } from "../utils/withLock.js";
-import { AbortError } from "../AbortError.js";
-import { GeneralChatPromptWrapper } from "../chatWrappers/GeneralChatPromptWrapper.js";
-import { getChatWrapperByBos } from "../chatWrappers/createChatWrapperByBos.js";
-import { generateContextTextFromConversationHistory } from "../chatWrappers/generateContextTextFromConversationHistory.js";
-import { removeNullFields } from "../utils/removeNullFields.js";
-import { LlamaModel } from "./LlamaModel.js";
-import { LlamaGrammarEvaluationState } from "./LlamaGrammarEvaluationState.js";
-const UNKNOWN_UNICODE_CHAR = "\ufffd";
-export class LlamaChatSession {
-    _systemPrompt;
-    _printLLamaSystemInfo;
-    _promptWrapper;
-    _promptIndex = 0;
-    _initialized = false;
-    _lastStopString = null;
-    _lastStopStringSuffix = null;
-    _conversationHistoryToLoad = null;
-    _ctx;
-    /**
-     * @param {LlamaChatSessionOptions} options
-     */
-    constructor({ context, printLLamaSystemInfo = false, promptWrapper = new GeneralChatPromptWrapper(), systemPrompt = defaultChatSystemPrompt, conversationHistory }) {
-        this._ctx = context;
-        this._printLLamaSystemInfo = printLLamaSystemInfo;
-        this._systemPrompt = systemPrompt;
-        this._conversationHistoryToLoad = (conversationHistory != null && conversationHistory.length > 0)
-            ? conversationHistory
-            : null;
-        if (promptWrapper === "auto") {
-            const chatWrapper = getChatWrapperByBos(context.getBosString());
-            if (chatWrapper != null)
-                this._promptWrapper = new chatWrapper();
-            else
-                this._promptWrapper = new GeneralChatPromptWrapper();
-        }
-        else
-            this._promptWrapper = promptWrapper;
-    }
-    get initialized() {
-        return this._initialized;
-    }
-    get context() {
-        return this._ctx;
-    }
-    async init() {
-        await withLock(this, "init", async () => {
-            if (this._initialized)
-                return;
-            if (this._printLLamaSystemInfo)
-                console.log("Llama system info", LlamaModel.systemInfo);
-            this._initialized = true;
-        });
-    }
-    /**
-     * @param {string} prompt
-     * @param {object} options
-     * @returns {Promise<string>}
-     */
-    async prompt(prompt, { onToken, signal, maxTokens, temperature, topK, topP, grammar = this.context._chatGrammar, trimWhitespaceSuffix = false, repeatPenalty } = {}) {
-        const { text } = await this.promptWithMeta(prompt, {
-            onToken, signal, maxTokens, temperature, topK, topP, grammar, trimWhitespaceSuffix, repeatPenalty
-        });
-        return text;
-    }
-    /**
-     * @param {string} prompt
-     * @param {LLamaChatPromptOptions} options
-     */
-    async promptWithMeta(prompt, { onToken, signal, maxTokens, temperature, topK, topP, grammar = this.context._chatGrammar, trimWhitespaceSuffix = false, repeatPenalty } = {}) {
-        if (!this.initialized)
-            await this.init();
-        return await withLock(this, "prompt", async () => {
-            let promptText = "";
-            if (this._promptIndex == 0 && this._conversationHistoryToLoad != null) {
-                const { text, stopString, stopStringSuffix } = generateContextTextFromConversationHistory(this._promptWrapper, this._conversationHistoryToLoad, {
-                    systemPrompt: this._systemPrompt,
-                    currentPromptIndex: this._promptIndex,
-                    lastStopString: this._lastStopString,
-                    lastStopStringSuffix: this._promptIndex == 0
-                        ? (this._ctx.prependBos
-                            ? this._ctx.getBosString()
-                            : null)
-                        : this._lastStopStringSuffix
-                });
-                promptText += text;
-                this._lastStopString = stopString;
-                this._lastStopStringSuffix = stopStringSuffix;
-                this._promptIndex += this._conversationHistoryToLoad.length;
-                this._conversationHistoryToLoad = null;
-            }
-            promptText += this._promptWrapper.wrapPrompt(prompt, {
-                systemPrompt: this._systemPrompt,
-                promptIndex: this._promptIndex,
-                lastStopString: this._lastStopString,
-                lastStopStringSuffix: this._promptIndex == 0
-                    ? (this._ctx.prependBos
-                        ? this._ctx.getBosString()
-                        : null)
-                    : this._lastStopStringSuffix
-            });
-            this._promptIndex++;
-            this._lastStopString = null;
-            this._lastStopStringSuffix = null;
-            const { text, stopReason, stopString, stopStringSuffix } = await this._evalTokens(this._ctx.encode(promptText), {
-                onToken, signal, maxTokens, temperature, topK, topP, grammar, trimWhitespaceSuffix,
-                repeatPenalty: repeatPenalty == false ? { lastTokens: 0 } : repeatPenalty
-            });
-            this._lastStopString = stopString;
-            this._lastStopStringSuffix = stopStringSuffix;
-            return {
-                text,
-                stopReason,
-                stopString,
-                stopStringSuffix
-            };
-        });
-    }
-    async _evalTokens(tokens, { onToken, signal, maxTokens, temperature, topK, topP, grammar = this.context._chatGrammar, trimWhitespaceSuffix = false, repeatPenalty: { lastTokens: repeatPenaltyLastTokens = 64, punishTokensFilter, penalizeNewLine, penalty, frequencyPenalty, presencePenalty } = {} } = {}) {
-        let stopStrings = this._promptWrapper.getStopStrings();
-        if (grammar != null)
-            stopStrings = stopStrings.concat(grammar.stopStrings);
-        const stopStringIndexes = Array(stopStrings.length).fill(0);
-        const skippedChunksQueue = [];
-        const res = [];
-        const grammarEvaluationState = grammar != null
-            ? new LlamaGrammarEvaluationState({ grammar })
-            : undefined;
-        const repeatPenaltyEnabled = repeatPenaltyLastTokens > 0;
-        let stopReason = "eosToken";
-        const getPenaltyTokens = () => {
-            let punishTokens = res.slice(-repeatPenaltyLastTokens);
-            if (punishTokensFilter != null)
-                punishTokens = punishTokensFilter(punishTokens);
-            if (!penalizeNewLine) {
-                const nlToken = this.context.getNlToken();
-                if (nlToken != null)
-                    punishTokens = punishTokens.filter(token => token !== nlToken);
-            }
-            return Uint32Array.from(punishTokens);
-        };
-        const evaluationIterator = this._ctx.evaluate(tokens, removeNullFields({
-            temperature, topK, topP, grammarEvaluationState,
-            repeatPenalty: !repeatPenaltyEnabled ? undefined : {
-                punishTokens: getPenaltyTokens,
-                penalty,
-                frequencyPenalty,
-                presencePenalty
-            }
-        }));
-        for await (const chunk of evaluationIterator) {
-            if (signal?.aborted)
-                throw new AbortError();
-            const tokenStr = this._ctx.decode(Uint32Array.from([chunk]));
-            const { shouldReturn, skipTokenEvent, stopString, stopStringSuffix } = this._checkStopString(tokenStr, stopStrings, stopStringIndexes);
-            if (shouldReturn) {
-                skippedChunksQueue.push(chunk);
-                const skippedChunksText = skippedChunksQueue.length > 0
-                    ? this._ctx.decode(Uint32Array.from(skippedChunksQueue))
-                    : "";
-                let [queuedTextBeforeStopString] = skippedChunksText.split(stopString);
-                if (grammar?.trimWhitespaceSuffix || trimWhitespaceSuffix)
-                    queuedTextBeforeStopString = queuedTextBeforeStopString.trimEnd();
-                if (queuedTextBeforeStopString.length > 0) {
-                    const beforeStopStringTokens = Array.from(this._ctx.encode(queuedTextBeforeStopString));
-                    res.push(...beforeStopStringTokens);
-                    onToken?.(beforeStopStringTokens);
-                    skippedChunksQueue.length = 0;
-                }
-                stopReason = "stopString";
-                return {
-                    text: this._ctx.decode(Uint32Array.from(res)),
-                    stopReason,
-                    stopString,
-                    stopStringSuffix
-                };
-            }
-            // if the token is unknown, it means it's not complete character
-            if (tokenStr === UNKNOWN_UNICODE_CHAR || skipTokenEvent || ((grammar?.trimWhitespaceSuffix || trimWhitespaceSuffix) && tokenStr.trim() === "")) {
-                skippedChunksQueue.push(chunk);
-                continue;
-            }
-            if (skippedChunksQueue.length > 0) {
-                res.push(...skippedChunksQueue);
-                onToken?.(skippedChunksQueue);
-                skippedChunksQueue.length = 0;
-            }
-            res.push(chunk);
-            onToken?.([chunk]);
-            if (maxTokens != null && maxTokens > 0 && res.length >= maxTokens) {
-                stopReason = "maxTokens";
-                break;
-            }
-        }
-        let resText = this._ctx.decode(Uint32Array.from(res));
-        if (grammar?.trimWhitespaceSuffix || trimWhitespaceSuffix)
-            resText = resText.trimEnd();
-        return {
-            text: resText,
-            stopReason,
-            stopString: null,
-            stopStringSuffix: null
-        };
-    }
-    _checkStopString(tokenStr, stopStrings, stopStringIndexes) {
-        let skipTokenEvent = false;
-        for (let stopStringIndex = 0; stopStringIndex < stopStrings.length; stopStringIndex++) {
-            const stopString = stopStrings[stopStringIndex];
-            let localShouldSkipTokenEvent = false;
-            let i = 0;
-            for (; i < tokenStr.length && stopStringIndexes[stopStringIndex] !== stopString.length; i++) {
-                if (tokenStr[i] === stopString[stopStringIndexes[stopStringIndex]]) {
-                    stopStringIndexes[stopStringIndex]++;
-                    localShouldSkipTokenEvent = true;
-                }
-                else {
-                    stopStringIndexes[stopStringIndex] = 0;
-                    localShouldSkipTokenEvent = false;
-                }
-            }
-            if (stopStringIndexes[stopStringIndex] === stopString.length) {
-                return {
-                    shouldReturn: true,
-                    stopString,
-                    stopStringSuffix: tokenStr.length === i
-                        ? null
-                        : tokenStr.slice(i)
-                };
-            }
-            skipTokenEvent ||= localShouldSkipTokenEvent;
-        }
-        return { skipTokenEvent };
-    }
-}
-//# sourceMappingURL=LlamaChatSession.js.map

package/dist/llamaEvaluator/LlamaChatSession.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"LlamaChatSession.js","sourceRoot":"","sources":["../../src/llamaEvaluator/LlamaChatSession.ts"],"names":[],"mappings":"AAAA,OAAO,EAAC,uBAAuB,EAAC,MAAM,cAAc,CAAC;AACrD,OAAO,EAAC,QAAQ,EAAC,MAAM,sBAAsB,CAAC;AAE9C,OAAO,EAAC,UAAU,EAAC,MAAM,kBAAkB,CAAC;AAC5C,OAAO,EAAC,wBAAwB,EAAC,MAAM,6CAA6C,CAAC;AACrF,OAAO,EAAC,mBAAmB,EAAC,MAAM,2CAA2C,CAAC;AAE9E,OAAO,EAAC,0CAA0C,EAAC,MAAM,+DAA+D,CAAC;AACzH,OAAO,EAAC,gBAAgB,EAAC,MAAM,8BAA8B,CAAC;AAC9D,OAAO,EAAC,UAAU,EAAC,MAAM,iBAAiB,CAAC;AAG3C,OAAO,EAAC,2BAA2B,EAAC,MAAM,kCAAkC,CAAC;AAE7E,MAAM,oBAAoB,GAAG,QAAQ,CAAC;AAqGtC,MAAM,OAAO,gBAAgB;IACR,aAAa,CAAS;IACtB,qBAAqB,CAAU;IAC/B,cAAc,CAAoB;IAC3C,YAAY,GAAW,CAAC,CAAC;IACzB,YAAY,GAAY,KAAK,CAAC;IAC9B,eAAe,GAAkB,IAAI,CAAC;IACtC,qBAAqB,GAAkB,IAAI,CAAC;IAC5C,0BAA0B,GAA8C,IAAI,CAAC;IACpE,IAAI,CAAe;IAEpC;;OAEG;IACH,YAAmB,EACf,OAAO,EACP,oBAAoB,GAAG,KAAK,EAC5B,aAAa,GAAG,IAAI,wBAAwB,EAAE,EAC9C,YAAY,GAAG,uBAAuB,EACtC,mBAAmB,EACG;QACtB,IAAI,CAAC,IAAI,GAAG,OAAO,CAAC;QACpB,IAAI,CAAC,qBAAqB,GAAG,oBAAoB,CAAC;QAClD,IAAI,CAAC,aAAa,GAAG,YAAY,CAAC;QAClC,IAAI,CAAC,0BAA0B,GAAG,CAAC,mBAAmB,IAAI,IAAI,IAAI,mBAAmB,CAAC,MAAM,GAAG,CAAC,CAAC;YAC7F,CAAC,CAAC,mBAAmB;YACrB,CAAC,CAAC,IAAI,CAAC;QAEX,IAAI,aAAa,KAAK,MAAM,EAAE;YAC1B,MAAM,WAAW,GAAG,mBAAmB,CAAC,OAAO,CAAC,YAAY,EAAE,CAAC,CAAC;YAEhE,IAAI,WAAW,IAAI,IAAI;gBACnB,IAAI,CAAC,cAAc,GAAG,IAAI,WAAW,EAAE,CAAC;;gBAExC,IAAI,CAAC,cAAc,GAAG,IAAI,wBAAwB,EAAE,CAAC;SAC5D;;YACG,IAAI,CAAC,cAAc,GAAG,aAAa,CAAC;IAC5C,CAAC;IAED,IAAW,WAAW;QAClB,OAAO,IAAI,CAAC,YAAY,CAAC;IAC7B,CAAC;IAED,IAAW,OAAO;QACd,OAAO,IAAI,CAAC,IAAI,CAAC;IACrB,CAAC;IAEM,KAAK,CAAC,IAAI;QACb,MAAM,QAAQ,CAAC,IAAI,EAAE,MAAM,EAAE,KAAK,IAAI,EAAE;YACpC,IAAI,IAAI,CAAC,YAAY;gBACjB,OAAO;YAEX,IAAI,IAAI,CAAC,qBAAqB;gBAC1B,OAAO,CAAC,GAAG,CAAC,mBAAmB,EAAE,UAAU,CAAC,UAAU,CAAC,CAAC;YAE5D,IAAI,CAAC,YAAY,GAAG,IAAI,CAAC;QAC7B,CAAC,CAAC,CAAC;IACP,CAAC;IAED;;;;OAIG;IACI,KAAK,CAAC,MAAM,CAAC,MAAc,EAAE,EAChC,OAAO,EACP,MAAM,EACN,SAAS,EACT,WAAW,EACX,IAAI,EACJ,IAAI,EACJ,OAAO,GAAG,IAAI,CAAC,OAAO,CAAC,YAAY,EACnC,oBAAoB,GAAG,KAAK,EAC5B,aAAa,KACW,EAAE;QAC1B,MAAM,EAAC,IAAI,EAAC,GAAG,MAAM,IAAI,CAAC,cAAc,CAAC,MAAM,EAAE;YAC7C,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE,WAAW,EAAE,IAAI,EAAE,IAAI,EAAE,OAAO,EAAE,oBAAoB,EAAE,aAAa;SACpG,CAAC,CAAC;QAEH,OAAO,IAAI,CAAC;IAChB,CAAC;IAED;;;OAGG;IACI,KAAK,CAAC,cAAc,CAAC,MAAc,EAAE,EACxC,OAAO,EACP,MAAM,EACN,SAAS,EACT,WAAW,EACX,IAAI,EACJ,IAAI,EACJ,OAAO,GAAG,IAAI,CAAC,OAAO,CAAC,YAAY,EACnC,oBAAoB,GAAG,KAAK,EAC5B,aAAa,KACW,EAAE;QAC1B,IAAI,CAAC,IAAI,CAAC,WAAW;YACjB,MAAM,IAAI,CAAC,IAAI,EAAE,CAAC;QAEtB,OAAO,MAAM,QAAQ,CAAC,IAAI,EAAE,QAAQ,EAAE,KAAK,IAAI,EAAE;YAC7C,IAAI,UAAU,GAAG,EAAE,CAAC;YAEpB,IAAI,IAAI,CAAC,YAAY,IAAI,CAAC,IAAI,IAAI,CAAC,0BAA0B,IAAI,IAAI,EAAE;gBACnE,MAAM,EAAC,IAAI,EAAE,UAAU,EAAE,gBAAgB,EAAC,GACtC,0CAA0C,CAAC,IAAI,CAAC,cAAc,EAAE,IAAI,CAAC,0BAA0B,EAAE;oBAC7F,YAAY,EAAE,IAAI,CAAC,aAAa;oBAChC,kBAAkB,EAAE,IAAI,CAAC,YAAY;oBACrC,cAAc,EAAE,IAAI,CAAC,eAAe;oBACpC,oBAAoB,EAAE,IAAI,CAAC,YAAY,IAAI,CAAC;wBACxC,CAAC,CAAC,CACE,IAAI,CAAC,IAAI,CAAC,UAAU;4BAChB,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE;4BAC1B,CAAC,CAAC,IAAI,CACb;wBACD,CAAC,CAAC,IAAI,CAAC,qBAAqB;iBACnC,CAAC,CAAC;gBAEP,UAAU,IAAI,IAAI,CAAC;gBACnB,IAAI,CAAC,eAAe,GAAG,UAAU,CAAC;gBAClC,IAAI,CAAC,qBAAqB,GAAG,gBAAgB,CAAC;gBAC9C,IAAI,CAAC,YAAY,IAAI,IAAI,CAAC,0BAA0B,CAAC,MAAM,CAAC;gBAE5D,IAAI,CAAC,0BAA0B,GAAG,IAAI,CAAC;aAC1C;YAED,UAAU,IAAI,IAAI,CAAC,cAAc,CAAC,UAAU,CAAC,MAAM,EAAE;gBACjD,YAAY,EAAE,IAAI,CAAC,aAAa;gBAChC,WAAW,EAAE,IAAI,CAAC,YAAY;gBAC9B,cAAc,EAAE,IAAI,CAAC,eAAe;gBACpC,oBAAoB,EAAE,IAAI,CAAC,YAAY,IAAI,CAAC;oBACxC,CAAC,CAAC,CACE,IAAI,CAAC,IAAI,CAAC,UAAU;wBAChB,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE;wBAC1B,CAAC,CAAC,IAAI,CACb;oBACD,CAAC,CAAC,IAAI,CAAC,qBAAqB;aACnC,CAAC,CAAC;YACH,IAAI,CAAC,YAAY,EAAE,CAAC;YACpB,IAAI,CAAC,eAAe,GAAG,IAAI,CAAC;YAC5B,IAAI,CAAC,qBAAqB,GAAG,IAAI,CAAC;YAElC,MAAM,EAAC,IAAI,EAAE,UAAU,EAAE,UAAU,EAAE,gBAAgB,EAAC,GAClD,MAAM,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,UAAU,CAAC,EAAE;gBACjD,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE,WAAW,EAAE,IAAI,EAAE,IAAI,EAAE,OAAO,EAAE,oBAAoB;gBAClF,aAAa,EAAE,aAAa,IAAI,KAAK,CAAC,CAAC,CAAC,EAAC,UAAU,EAAE,CAAC,EAAC,CAAC,CAAC,CAAC,aAAa;aAC1E,CAAC,CAAC;YACP,IAAI,CAAC,eAAe,GAAG,UAAU,CAAC;YAClC,IAAI,CAAC,qBAAqB,GAAG,gBAAgB,CAAC;YAE9C,OAAO;gBACH,IAAI;gBACJ,UAAU;gBACV,UAAU;gBACV,gBAAgB;aACnB,CAAC;QACN,CAAC,CAAC,CAAC;IACP,CAAC;IAEO,KAAK,CAAC,WAAW,CAAC,MAAmB,EAAE,EAC3C,OAAO,EACP,MAAM,EACN,SAAS,EACT,WAAW,EACX,IAAI,EACJ,IAAI,EACJ,OAAO,GAAG,IAAI,CAAC,OAAO,CAAC,YAAY,EACnC,oBAAoB,GAAG,KAAK,EAC5B,aAAa,EAAE,EACX,UAAU,EAAE,uBAAuB,GAAG,EAAE,EACxC,kBAAkB,EAClB,eAAe,EACf,OAAO,EACP,gBAAgB,EAChB,eAAe,EAClB,GAAG,EAAE,KAWN,EAAE;QACF,IAAI,WAAW,GAAG,IAAI,CAAC,cAAc,CAAC,cAAc,EAAE,CAAC;QAEvD,IAAI,OAAO,IAAI,IAAI;YACf,WAAW,GAAG,WAAW,CAAC,MAAM,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC;QAE1D,MAAM,iBAAiB,GAAa,KAAK,CAAC,WAAW,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACtE,MAAM,kBAAkB,GAAY,EAAE,CAAC;QACvC,MAAM,GAAG,GAAY,EAAE,CAAC;QACxB,MAAM,sBAAsB,GAAG,OAAO,IAAI,IAAI;YAC1C,CAAC,CAAC,IAAI,2BAA2B,CAAC,EAAC,OAAO,EAAC,CAAC;YAC5C,CAAC,CAAC,SAAS,CAAC;QAChB,MAAM,oBAAoB,GAAG,uBAAuB,GAAG,CAAC,CAAC;QACzD,IAAI,UAAU,GAA4C,UAAU,CAAC;QAErE,MAAM,gBAAgB,GAAG,GAAG,EAAE;YAC1B,IAAI,YAAY,GAAG,GAAG,CAAC,KAAK,CAAC,CAAC,uBAAuB,CAAC,CAAC;YAEvD,IAAI,kBAAkB,IAAI,IAAI;gBAC1B,YAAY,GAAG,kBAAkB,CAAC,YAAY,CAAC,CAAC;YAEpD,IAAI,CAAC,eAAe,EAAE;gBAClB,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,CAAC,UAAU,EAAE,CAAC;gBAE1C,IAAI,OAAO,IAAI,IAAI;oBACf,YAAY,GAAG,YAAY,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC,KAAK,KAAK,OAAO,CAAC,CAAC;aACtE;YAED,OAAO,WAAW,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;QAC1C,CAAC,CAAC;QAEF,MAAM,kBAAkB,GAAG,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,MAAM,EAAE,gBAAgB,CAAC;YACnE,WAAW,EAAE,IAAI,EAAE,IAAI,EAAE,sBAAsB;YAC/C,aAAa,EAAE,CAAC,oBAAoB,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC;gBAC/C,YAAY,EAAE,gBAAgB;gBAC9B,OAAO;gBACP,gBAAgB;gBAChB,eAAe;aAClB;SACJ,CAAC,CAAC,CAAC;QAEJ,IAAI,KAAK,EAAE,MAAM,KAAK,IAAI,kBAAkB,EAAE;YAC1C,IAAI,MAAM,EAAE,OAAO;gBACf,MAAM,IAAI,UAAU,EAAE,CAAC;YAE3B,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,WAAW,CAAC,IAAI,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;YAC7D,MAAM,EACF,YAAY,EAAE,cAAc,EAAE,UAAU,EAAE,gBAAgB,EAC7D,GAAG,IAAI,CAAC,gBAAgB,CAAC,QAAQ,EAAE,WAAW,EAAE,iBAAiB,CAAC,CAAC;YAEpE,IAAI,YAAY,EAAE;gBACd,kBAAkB,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;gBAC/B,MAAM,iBAAiB,GAAG,kBAAkB,CAAC,MAAM,GAAG,CAAC;oBACnD,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,WAAW,CAAC,IAAI,CAAC,kBAAkB,CAAC,CAAC;oBACxD,CAAC,CAAC,EAAE,CAAC;gBAET,IAAI,CAAC,0BAA0B,CAAC,GAAG,iBAAiB,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC;gBAEvE,IAAI,OAAO,EAAE,oBAAoB,IAAI,oBAAoB;oBACrD,0BAA0B,GAAG,0BAA0B,CAAC,OAAO,EAAE,CAAC;gBAEtE,IAAI,0BAA0B,CAAC,MAAM,GAAG,CAAC,EAAE;oBACvC,MAAM,sBAAsB,GAAY,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,0BAA0B,CAAC,CAAC,CAAC;oBAEjG,GAAG,CAAC,IAAI,CAAC,GAAG,sBAAsB,CAAC,CAAC;oBACpC,OAAO,EAAE,CAAC,sBAAsB,CAAC,CAAC;oBAClC,kBAAkB,CAAC,MAAM,GAAG,CAAC,CAAC;iBACjC;gBAED,UAAU,GAAG,YAAY,CAAC;gBAE1B,OAAO;oBACH,IAAI,EAAE,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,WAAW,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;oBAC7C,UAAU;oBACV,UAAU;oBACV,gBAAgB;iBACnB,CAAC;aACL;YAED,gEAAgE;YAChE,IAAI,QAAQ,KAAK,oBAAoB,IAAI,cAAc,IAAI,CACvD,CAAC,OAAO,EAAE,oBAAoB,IAAI,oBAAoB,CAAC,IAAI,QAAQ,CAAC,IAAI,EAAE,KAAK,EAAE,CACpF,EAAE;gBACC,kBAAkB,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;gBAC/B,SAAS;aACZ;YAED,IAAI,kBAAkB,CAAC,MAAM,GAAG,CAAC,EAAE;gBAC/B,GAAG,CAAC,IAAI,CAAC,GAAG,kBAAkB,CAAC,CAAC;gBAChC,OAAO,EAAE,CAAC,kBAAkB,CAAC,CAAC;gBAC9B,kBAAkB,CAAC,MAAM,GAAG,CAAC,CAAC;aACjC;YAED,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;YAChB,OAAO,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC;YAEnB,IAAI,SAAS,IAAI,IAAI,IAAI,SAAS,GAAG,CAAC,IAAI,GAAG,CAAC,MAAM,IAAI,SAAS,EAAE;gBAC/D,UAAU,GAAG,WAAW,CAAC;gBACzB,MAAM;aACT;SACJ;QAED,IAAI,OAAO,GAAG,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,WAAW,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;QAEtD,IAAI,OAAO,EAAE,oBAAoB,IAAI,oBAAoB;YACrD,OAAO,GAAG,OAAO,CAAC,OAAO,EAAE,CAAC;QAEhC,OAAO;YACH,IAAI,EAAE,OAAO;YACb,UAAU;YACV,UAAU,EAAE,IAAI;YAChB,gBAAgB,EAAE,IAAI;SACzB,CAAC;IACN,CAAC;IAEO,gBAAgB,CAAC,QAAgB,EAAE,WAAqB,EAAE,iBAA2B;QACzF,IAAI,cAAc,GAAG,KAAK,CAAC;QAE3B,KAAK,IAAI,eAAe,GAAG,CAAC,EAAE,eAAe,GAAG,WAAW,CAAC,MAAM,EAAE,eAAe,EAAE,EAAE;YACnF,MAAM,UAAU,GAAG,WAAW,CAAC,eAAe,CAAC,CAAC;YAEhD,IAAI,yBAAyB,GAAG,KAAK,CAAC;YACtC,IAAI,CAAC,GAAG,CAAC,CAAC;YACV,OAAO,CAAC,GAAG,QAAQ,CAAC,MAAM,IAAI,iBAAiB,CAAC,eAAe,CAAC,KAAK,UAAU,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE;gBACzF,IAAI,QAAQ,CAAC,CAAC,CAAC,KAAK,UAAU,CAAC,iBAAiB,CAAC,eAAe,CAAC,CAAC,EAAE;oBAChE,iBAAiB,CAAC,eAAe,CAAC,EAAE,CAAC;oBACrC,yBAAyB,GAAG,IAAI,CAAC;iBACpC;qBAAM;oBACH,iBAAiB,CAAC,eAAe,CAAC,GAAG,CAAC,CAAC;oBACvC,yBAAyB,GAAG,KAAK,CAAC;iBACrC;aACJ;YAED,IAAI,iBAAiB,CAAC,eAAe,CAAC,KAAK,UAAU,CAAC,MAAM,EAAE;gBAC1D,OAAO;oBACH,YAAY,EAAE,IAAI;oBAClB,UAAU;oBACV,gBAAgB,EAAE,QAAQ,CAAC,MAAM,KAAK,CAAC;wBACnC,CAAC,CAAC,IAAI;wBACN,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC;iBAC1B,CAAC;aACL;YAED,cAAc,KAAK,yBAAyB,CAAC;SAChD;QAED,OAAO,EAAC,cAAc,EAAC,CAAC;IAC5B,CAAC;CACJ"}

package/dist/llamaEvaluator/LlamaContext.d.ts

@@ -1,99 +0,0 @@
-import { Token } from "../types.js";
-import { LlamaModel } from "./LlamaModel.js";
-import { LlamaGrammarEvaluationState } from "./LlamaGrammarEvaluationState.js";
-import { LlamaGrammar } from "./LlamaGrammar.js";
-export type LlamaContextOptions = {
-    model: LlamaModel;
-    prependBos?: boolean;
-    /**
-     * @deprecated use the `grammar` option on `LlamaChatSession`'s `prompt` function
-     * or the `grammarEvaluationState` option on `LlamaContext`'s `evaluate` function instead
-     * @hidden
-     */
-    grammar?: LlamaGrammar;
-    /** If null, a random seed will be used */
-    seed?: number | null;
-    /** text context size */
-    contextSize?: number;
-    /** prompt processing batch size */
-    batchSize?: number;
-    /** the llama_eval() call computes all logits, not just the last one */
-    logitsAll?: boolean;
-    /** embedding mode only */
-    embedding?: boolean;
-    /** number of threads to use to evaluate tokens */
-    threads?: number;
-};
-export type LlamaContextRepeatPenalty = {
-    /** Tokens to lower the predication probability of to be the next predicted token */
-    punishTokens: Uint32Array | (() => Uint32Array);
-    /**
-     * The relative amount to lower the probability of the tokens in `punishTokens` by
-     * Defaults to `1.1`.
-     * Set to `1` to disable.
-     */
-    penalty?: number;
-    /**
-     * For n time a token is in the `punishTokens` array, lower its probability by `n * frequencyPenalty`
-     * Disabled by default (`0`).
-     * Set to a value between `0` and `1` to enable.
-     */
-    frequencyPenalty?: number;
-    /**
-     * Lower the probability of all the tokens in the `punishTokens` array by `presencePenalty`
-     * Disabled by default (`0`).
-     * Set to a value between `0` and `1` to enable.
-     */
-    presencePenalty?: number;
-};
-export declare class LlamaContext {
-    private readonly _model;
-    private readonly _ctx;
-    private readonly _prependBos;
-    private _prependTokens;
-    /**
-     * @param {LlamaContextOptions} options
-     */
-    constructor({ model, prependBos, grammar, seed, contextSize, batchSize, logitsAll, embedding, threads }: LlamaContextOptions);
-    encode(text: string): Uint32Array;
-    decode(tokens: Uint32Array | Token[]): string;
-    get prependBos(): boolean;
-    /**
-     * @returns {Token | null} The BOS (Beginning Of Sequence) token.
-     */
-    getBosToken(): Token | null;
-    /**
-     * @returns {Token | null} The EOS (End Of Sequence) token.
-     */
-    getEosToken(): Token | null;
-    /**
-     * @returns {Token | null} The NL (New Line) token.
-     */
-    getNlToken(): Token | null;
-    /**
-     * @returns {string | null} The BOS (Beginning Of Sequence) token as a string.
-     */
-    getBosString(): string | null;
-    /**
-     * @returns {string | null} The EOS (End Of Sequence) token as a string.
-     */
-    getEosString(): string | null;
-    /**
-     * @returns {string | null} The NL (New Line) token as a string.
-     */
-    getNlString(): string | null;
-    getContextSize(): number;
-    printTimings(): void;
-    /**
-     * @param {Uint32Array} tokens
-     * @param {object} options
-     * @returns {AsyncGenerator<Token, void>}
-     */
-    evaluate(tokens: Uint32Array, { temperature, topK, topP, grammarEvaluationState, repeatPenalty }?: {
-        temperature?: number;
-        topK?: number;
-        topP?: number;
-        grammarEvaluationState?: LlamaGrammarEvaluationState;
-        repeatPenalty?: LlamaContextRepeatPenalty;
-    }): AsyncGenerator<Token, void>;
-}