@inferrlm/react-native-mlx 0.4.0 → 0.4.2-alpha.0

package/lib/typescript/src/llm.d.ts

@@ -0,0 +1,129 @@
+import type { GenerationStats, LLMLoadOptions, StreamEvent } from './specs/LLM.nitro';
+export type EventCallback = (event: StreamEvent) => void;
+export type Message = {
+    role: 'user' | 'assistant' | 'system';
+    content: string;
+};
+export type ToolCallInfo = {
+    name: string;
+    arguments: Record<string, unknown>;
+};
+export type ToolCallUpdate = {
+    toolCall: ToolCallInfo;
+    allToolCalls: ToolCallInfo[];
+};
+/**
+ * LLM text generation using MLX on Apple Silicon.
+ *
+ * @example
+ * ```ts
+ * import { LLM } from 'react-native-nitro-mlx'
+ *
+ * // Load a model
+ * await LLM.load('mlx-community/Qwen3-0.6B-4bit', progress => {
+ *   console.log(`Loading: ${(progress * 100).toFixed(0)}%`)
+ * })
+ *
+ * // Stream a response
+ * await LLM.stream('Hello!', token => {
+ *   process.stdout.write(token)
+ * })
+ *
+ * // Get generation stats
+ * const stats = LLM.getLastGenerationStats()
+ * console.log(`${stats.tokensPerSecond} tokens/sec`)
+ * ```
+ */
+export declare const LLM: {
+    /**
+     * Load a model into memory. Downloads the model from HuggingFace if not already cached.
+     * @param modelId - HuggingFace model ID (e.g., 'mlx-community/Qwen3-0.6B-4bit')
+     * @param options - Callback invoked with loading progress (0-1)
+     */
+    load(modelId: string, options: LLMLoadOptions): Promise<void>;
+    /**
+     * Generate a complete response for a prompt. Blocks until generation is complete.
+     * For streaming responses, use `stream()` instead.
+     * @param prompt - The input text to generate a response for
+     * @returns The complete generated text
+     */
+    generate(prompt: string): Promise<string>;
+    /**
+     * Stream a response token by token with optional tool calling support.
+     * Tools must be provided when loading the model via `load()` options.
+     * Tools are automatically executed when the model calls them.
+     * @param prompt - The input text to generate a response for
+     * @param onToken - Callback invoked for each generated token
+     * @param onToolCall - Optional callback invoked when a tool is called.
+     *   Receives the current tool call and an accumulated array of all tool calls so far.
+     * @returns The complete generated text
+     */
+    stream(prompt: string, onToken: (token: string) => void, onToolCall?: (update: ToolCallUpdate) => void): Promise<string>;
+    /**
+     * Stream with typed events for thinking blocks and tool calls.
+     * Provides granular lifecycle events for UI updates.
+     *
+     * @param prompt - The input text
+     * @param onEvent - Callback receiving typed StreamEvent objects
+     * @returns Promise resolving to final content string (thinking content stripped)
+     *
+     * @example
+     * ```ts
+     * await LLM.streamWithEvents(prompt, (event) => {
+     *   switch (event.type) {
+     *     case 'token':
+     *       appendToContent(event.token)
+     *       break
+     *     case 'thinking_start':
+     *       showThinkingIndicator()
+     *       break
+     *     case 'thinking_chunk':
+     *       appendToThinking(event.chunk)
+     *       break
+     *     case 'tool_call_start':
+     *       showToolCallCard(event.name, event.arguments)
+     *       break
+     *   }
+     * })
+     * ```
+     */
+    streamWithEvents(prompt: string, onEvent: EventCallback): Promise<string>;
+    /**
+     * Stop the current generation. Safe to call even if not generating.
+     */
+    stop(): void;
+    /**
+     * Unload the current model and release memory.
+     * Call this when you're done with the model to free up memory.
+     */
+    unload(): void;
+    /**
+     * Get statistics from the last generation.
+     * @returns Statistics including token count, tokens/sec (excluding tool execution), TTFT, total time, and tool execution time
+     */
+    getLastGenerationStats(): GenerationStats;
+    /**
+     * Get the message history if management is enabled.
+     * @returns Array of messages in the history
+     */
+    getHistory(): Message[];
+    /**
+     * Clear the message history.
+     */
+    clearHistory(): void;
+    /** Whether a model is currently loaded and ready for generation */
+    readonly isLoaded: boolean;
+    /** Whether text is currently being generated */
+    readonly isGenerating: boolean;
+    /** The ID of the currently loaded model, or empty string if none */
+    readonly modelId: string;
+    /** Enable debug logging to console */
+    debug: boolean;
+    /**
+     * System prompt used when loading the model.
+     * Set this before calling `load()`. Changes require reloading the model.
+     * @default "You are a helpful assistant."
+     */
+    systemPrompt: string;
+};
+//# sourceMappingURL=llm.d.ts.map

package/lib/typescript/src/llm.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"llm.d.ts","sourceRoot":"","sources":["../../../src/llm.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EACV,eAAe,EACf,cAAc,EAEd,WAAW,EACZ,MAAM,mBAAmB,CAAA;AAE1B,MAAM,MAAM,aAAa,GAAG,CAAC,KAAK,EAAE,WAAW,KAAK,IAAI,CAAA;AAIxD,MAAM,MAAM,OAAO,GAAG;IACpB,IAAI,EAAE,MAAM,GAAG,WAAW,GAAG,QAAQ,CAAA;IACrC,OAAO,EAAE,MAAM,CAAA;CAChB,CAAA;AAED,MAAM,MAAM,YAAY,GAAG;IACzB,IAAI,EAAE,MAAM,CAAA;IACZ,SAAS,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC,CAAA;AAED,MAAM,MAAM,cAAc,GAAG;IAC3B,QAAQ,EAAE,YAAY,CAAA;IACtB,YAAY,EAAE,YAAY,EAAE,CAAA;CAC7B,CAAA;AASD;;;;;;;;;;;;;;;;;;;;;GAqBG;AACH,eAAO,MAAM,GAAG;IACd;;;;OAIG;kBACW,MAAM,WAAW,cAAc,GAAG,OAAO,CAAC,IAAI,CAAC;IAI7D;;;;;OAKG;qBACc,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC;IAIzC;;;;;;;;;OASG;mBAEO,MAAM,WACL,CAAC,KAAK,EAAE,MAAM,KAAK,IAAI,eACnB,CAAC,MAAM,EAAE,cAAc,KAAK,IAAI,GAC5C,OAAO,CAAC,MAAM,CAAC;IAyBlB;;;;;;;;;;;;;;;;;;;;;;;;;;;OA2BG;6BACsB,MAAM,WAAW,aAAa,GAAG,OAAO,CAAC,MAAM,CAAC;IAWzE;;OAEG;YACK,IAAI;IAIZ;;;OAGG;cACO,IAAI;IAId;;;OAGG;8BACuB,eAAe;IAIzC;;;OAGG;kBACW,OAAO,EAAE;IAIvB;;OAEG;oBACa,IAAI;IAIpB,mEAAmE;uBACnD,OAAO;IAIvB,gDAAgD;2BAC5B,OAAO;IAI3B,oEAAoE;sBACrD,MAAM;IAIrB,sCAAsC;WACzB,OAAO;IAQpB;;;;OAIG;kBACiB,MAAM;CAO3B,CAAA"}

package/lib/typescript/src/modelManager.d.ts

@@ -0,0 +1,53 @@
+/**
+ * Manage MLX model downloads from HuggingFace.
+ *
+ * @example
+ * ```ts
+ * import { ModelManager } from 'react-native-nitro-mlx'
+ *
+ * // Download a model
+ * await ModelManager.download('mlx-community/Qwen3-0.6B-4bit', progress => {
+ *   console.log(`Downloading: ${(progress * 100).toFixed(0)}%`)
+ * })
+ *
+ * // Check if downloaded
+ * const isReady = await ModelManager.isDownloaded('mlx-community/Qwen3-0.6B-4bit')
+ *
+ * // List all downloaded models
+ * const models = await ModelManager.getDownloadedModels()
+ * ```
+ */
+export declare const ModelManager: {
+    /**
+     * Download a model from HuggingFace.
+     * @param modelId - HuggingFace model ID (e.g., 'mlx-community/Qwen3-0.6B-4bit')
+     * @param progressCallback - Callback invoked with download progress (0-1)
+     * @returns Path to the downloaded model directory
+     */
+    download(modelId: string, progressCallback: (progress: number) => void): Promise<string>;
+    /**
+     * Check if a model is already downloaded.
+     * @param modelId - HuggingFace model ID
+     * @returns True if the model is fully downloaded
+     */
+    isDownloaded(modelId: string): Promise<boolean>;
+    /**
+     * Get a list of all downloaded model IDs.
+     * @returns Array of model IDs that are available locally
+     */
+    getDownloadedModels(): Promise<string[]>;
+    /**
+     * Delete a downloaded model to free up disk space.
+     * @param modelId - HuggingFace model ID
+     */
+    deleteModel(modelId: string): Promise<void>;
+    /**
+     * Get the local filesystem path for a downloaded model.
+     * @param modelId - HuggingFace model ID
+     * @returns Absolute path to the model directory
+     */
+    getModelPath(modelId: string): Promise<string>;
+    /** Enable debug logging to console */
+    debug: boolean;
+};
+//# sourceMappingURL=modelManager.d.ts.map

package/lib/typescript/src/modelManager.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"modelManager.d.ts","sourceRoot":"","sources":["../../../src/modelManager.ts"],"names":[],"mappings":"AAYA;;;;;;;;;;;;;;;;;;GAkBG;AACH,eAAO,MAAM,YAAY;IACvB;;;;;OAKG;sBAEQ,MAAM,oBACG,CAAC,QAAQ,EAAE,MAAM,KAAK,IAAI,GAC3C,OAAO,CAAC,MAAM,CAAC;IAIlB;;;;OAIG;0BACmB,MAAM,GAAG,OAAO,CAAC,OAAO,CAAC;IAI/C;;;OAGG;2BACoB,OAAO,CAAC,MAAM,EAAE,CAAC;IAIxC;;;OAGG;yBACkB,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC;IAI3C;;;;OAIG;0BACmB,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC;IAI9C,sCAAsC;WACzB,OAAO;CAOrB,CAAA"}

package/lib/typescript/src/models.d.ts

@@ -0,0 +1,67 @@
+export declare enum ModelFamily {
+    Llama = "Llama",
+    Qwen = "Qwen",
+    Gemma = "Gemma",
+    Phi = "Phi",
+    SmolLM = "SmolLM",
+    OpenELM = "OpenELM",
+    PocketTTS = "PocketTTS",
+    GLMASR = "GLMASR"
+}
+export declare enum ModelProvider {
+    Meta = "Meta",
+    Alibaba = "Alibaba",
+    Google = "Google",
+    Microsoft = "Microsoft",
+    HuggingFace = "HuggingFace",
+    Apple = "Apple",
+    Kyutai = "Kyutai"
+}
+export type ModelQuantization = '4bit' | '8bit' | 'bf16';
+export type ModelType = 'llm' | 'tts' | 'stt';
+export interface ModelInfo {
+    id: MLXModel;
+    family: ModelFamily;
+    provider: ModelProvider;
+    parameters: string;
+    quantization: ModelQuantization;
+    displayName: string;
+    downloadSize: number;
+    type: ModelType;
+}
+export declare enum MLXModel {
+    Llama_3_2_1B_Instruct_4bit = "mlx-community/Llama-3.2-1B-Instruct-4bit",
+    Llama_3_2_1B_Instruct_8bit = "mlx-community/Llama-3.2-1B-Instruct-8bit",
+    Llama_3_2_3B_Instruct_4bit = "mlx-community/Llama-3.2-3B-Instruct-4bit",
+    Llama_3_2_3B_Instruct_8bit = "mlx-community/Llama-3.2-3B-Instruct-8bit",
+    Qwen2_5_0_5B_Instruct_4bit = "mlx-community/Qwen2.5-0.5B-Instruct-4bit",
+    Qwen2_5_0_5B_Instruct_8bit = "mlx-community/Qwen2.5-0.5B-Instruct-8bit",
+    Qwen2_5_1_5B_Instruct_4bit = "mlx-community/Qwen2.5-1.5B-Instruct-4bit",
+    Qwen2_5_1_5B_Instruct_8bit = "mlx-community/Qwen2.5-1.5B-Instruct-8bit",
+    Qwen2_5_3B_Instruct_4bit = "mlx-community/Qwen2.5-3B-Instruct-4bit",
+    Qwen2_5_3B_Instruct_8bit = "mlx-community/Qwen2.5-3B-Instruct-8bit",
+    Qwen3_1_7B_4bit = "mlx-community/Qwen3-1.7B-4bit",
+    Qwen3_1_7B_8bit = "mlx-community/Qwen3-1.7B-8bit",
+    Qwen3_5_0_8B_MLX_4bit = "mlx-community/Qwen3.5-0.8B-MLX-4bit",
+    Qwen3_5_0_8B_MLX_8bit = "mlx-community/Qwen3.5-0.8B-MLX-8bit",
+    Gemma_3_1B_IT_4bit = "mlx-community/gemma-3-1b-it-4bit",
+    Gemma_3_1B_IT_8bit = "mlx-community/gemma-3-1b-it-8bit",
+    Phi_3_5_Mini_Instruct_4bit = "mlx-community/Phi-3.5-mini-instruct-4bit",
+    Phi_3_5_Mini_Instruct_8bit = "mlx-community/Phi-3.5-mini-instruct-8bit",
+    Phi_4_Mini_Instruct_4bit = "mlx-community/Phi-4-mini-instruct-4bit",
+    Phi_4_Mini_Instruct_8bit = "mlx-community/Phi-4-mini-instruct-8bit",
+    SmolLM_1_7B_Instruct_4bit = "mlx-community/SmolLM-1.7B-Instruct-4bit",
+    SmolLM_1_7B_Instruct_8bit = "mlx-community/SmolLM-1.7B-Instruct-8bit",
+    SmolLM2_1_7B_Instruct_4bit = "mlx-community/SmolLM2-1.7B-Instruct-4bit",
+    SmolLM2_1_7B_Instruct_8bit = "mlx-community/SmolLM2-1.7B-Instruct-8bit",
+    OpenELM_1_1B_4bit = "mlx-community/OpenELM-1_1B-4bit",
+    OpenELM_1_1B_8bit = "mlx-community/OpenELM-1_1B-8bit",
+    OpenELM_3B_4bit = "mlx-community/OpenELM-3B-4bit",
+    OpenELM_3B_8bit = "mlx-community/OpenELM-3B-8bit",
+    PocketTTS = "mlx-community/pocket-tts",
+    PocketTTS_8bit = "mlx-community/pocket-tts-8bit",
+    PocketTTS_4bit = "mlx-community/pocket-tts-4bit",
+    GLM_ASR_Nano_4bit = "mlx-community/GLM-ASR-Nano-2512-4bit"
+}
+export declare const MLXModels: ModelInfo[];
+//# sourceMappingURL=models.d.ts.map

package/lib/typescript/src/models.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"models.d.ts","sourceRoot":"","sources":["../../../src/models.ts"],"names":[],"mappings":"AAAA,oBAAY,WAAW;IACrB,KAAK,UAAU;IACf,IAAI,SAAS;IACb,KAAK,UAAU;IACf,GAAG,QAAQ;IACX,MAAM,WAAW;IACjB,OAAO,YAAY;IACnB,SAAS,cAAc;IACvB,MAAM,WAAW;CAClB;AAED,oBAAY,aAAa;IACvB,IAAI,SAAS;IACb,OAAO,YAAY;IACnB,MAAM,WAAW;IACjB,SAAS,cAAc;IACvB,WAAW,gBAAgB;IAC3B,KAAK,UAAU;IACf,MAAM,WAAW;CAClB;AAED,MAAM,MAAM,iBAAiB,GAAG,MAAM,GAAG,MAAM,GAAG,MAAM,CAAA;AAExD,MAAM,MAAM,SAAS,GAAG,KAAK,GAAG,KAAK,GAAG,KAAK,CAAA;AAE7C,MAAM,WAAW,SAAS;IACxB,EAAE,EAAE,QAAQ,CAAA;IACZ,MAAM,EAAE,WAAW,CAAA;IACnB,QAAQ,EAAE,aAAa,CAAA;IACvB,UAAU,EAAE,MAAM,CAAA;IAClB,YAAY,EAAE,iBAAiB,CAAA;IAC/B,WAAW,EAAE,MAAM,CAAA;IACnB,YAAY,EAAE,MAAM,CAAA;IACpB,IAAI,EAAE,SAAS,CAAA;CAChB;AAED,oBAAY,QAAQ;IAElB,0BAA0B,6CAA6C;IACvE,0BAA0B,6CAA6C;IACvE,0BAA0B,6CAA6C;IACvE,0BAA0B,6CAA6C;IAGvE,0BAA0B,6CAA6C;IACvE,0BAA0B,6CAA6C;IACvE,0BAA0B,6CAA6C;IACvE,0BAA0B,6CAA6C;IACvE,wBAAwB,2CAA2C;IACnE,wBAAwB,2CAA2C;IAGnE,eAAe,kCAAkC;IACjD,eAAe,kCAAkC;IAGjD,qBAAqB,wCAAwC;IAC7D,qBAAqB,wCAAwC;IAG7D,kBAAkB,qCAAqC;IACvD,kBAAkB,qCAAqC;IAGvD,0BAA0B,6CAA6C;IACvE,0BAA0B,6CAA6C;IAGvE,wBAAwB,2CAA2C;IACnE,wBAAwB,2CAA2C;IAGnE,yBAAyB,4CAA4C;IACrE,yBAAyB,4CAA4C;IAGrE,0BAA0B,6CAA6C;IACvE,0BAA0B,6CAA6C;IAGvE,iBAAiB,oCAAoC;IACrD,iBAAiB,oCAAoC;IACrD,eAAe,kCAAkC;IACjD,eAAe,kCAAkC;IAGjD,SAAS,6BAA6B;IACtC,cAAc,kCAAkC;IAChD,cAAc,kCAAkC;IAGhD,iBAAiB,yCAAyC;CAC3D;AAED,eAAO,MAAM,SAAS,EAAE,SAAS,EAiUhC,CAAA"}

package/lib/typescript/src/specs/LLM.nitro.d.ts

@@ -0,0 +1,160 @@
+import type { AnyMap, HybridObject } from 'react-native-nitro-modules';
+/**
+ * Statistics from the last text generation.
+ */
+export interface GenerationStats {
+    tokenCount: number;
+    tokensPerSecond: number;
+    timeToFirstToken: number;
+    totalTime: number;
+    toolExecutionTime: number;
+}
+export interface GenerationStartEvent {
+    type: 'generation_start';
+    timestamp: number;
+}
+export interface TokenEvent {
+    type: 'token';
+    token: string;
+}
+export interface ThinkingStartEvent {
+    type: 'thinking_start';
+    timestamp: number;
+}
+export interface ThinkingChunkEvent {
+    type: 'thinking_chunk';
+    chunk: string;
+}
+export interface ThinkingEndEvent {
+    type: 'thinking_end';
+    content: string;
+    timestamp: number;
+}
+export interface ToolCallStartEvent {
+    type: 'tool_call_start';
+    id: string;
+    name: string;
+    arguments: string;
+}
+export interface ToolCallExecutingEvent {
+    type: 'tool_call_executing';
+    id: string;
+}
+export interface ToolCallCompletedEvent {
+    type: 'tool_call_completed';
+    id: string;
+    result: string;
+}
+export interface ToolCallFailedEvent {
+    type: 'tool_call_failed';
+    id: string;
+    error: string;
+}
+export interface GenerationEndEvent {
+    type: 'generation_end';
+    content: string;
+    stats: GenerationStats;
+}
+export type StreamEvent = GenerationStartEvent | TokenEvent | ThinkingStartEvent | ThinkingChunkEvent | ThinkingEndEvent | ToolCallStartEvent | ToolCallExecutingEvent | ToolCallCompletedEvent | ToolCallFailedEvent | GenerationEndEvent;
+export interface LLMMessage {
+    role: string;
+    content: string;
+}
+/**
+ * Parameter definition for a tool.
+ */
+export interface ToolParameter {
+    name: string;
+    type: string;
+    description: string;
+    required: boolean;
+}
+/**
+ * Tool definition that can be called by the model.
+ */
+export interface ToolDefinition {
+    name: string;
+    description: string;
+    parameters: ToolParameter[];
+    handler: (args: AnyMap) => Promise<AnyMap>;
+}
+/** Options for loading a model.
+ */
+export interface LLMLoadOptions {
+    /** Callback invoked with loading progress (0-1) */
+    onProgress?: (progress: number) => void;
+    /** Additional context to provide to the model */
+    additionalContext?: LLMMessage[];
+    /** Whether to automatically manage message history */
+    manageHistory?: boolean;
+    /** Tools available for the model to call */
+    tools?: ToolDefinition[];
+}
+/**
+ * Low-level LLM interface for text generation using MLX.
+ * @internal Use the `LLM` export from `react-native-nitro-mlx` instead.
+ */
+export interface LLM extends HybridObject<{
+    ios: 'swift';
+}> {
+    /**
+     * Load a model into memory. Downloads from HuggingFace if not already cached.
+     * @param modelId - HuggingFace model ID (e.g., 'mlx-community/Qwen3-0.6B-4bit')
+     * @param options - Callback invoked with loading progress (0-1)
+     */
+    load(modelId: string, options?: LLMLoadOptions): Promise<void>;
+    /**
+     * Generate a complete response for a prompt.
+     * @param prompt - The input text to generate a response for
+     * @returns The generated text
+     */
+    generate(prompt: string): Promise<string>;
+    /**
+     * Stream a response token by token with optional tool calling support.
+     * Tools are automatically executed when the model calls them.
+     * @param prompt - The input text to generate a response for
+     * @param onToken - Callback invoked for each generated token
+     * @param onToolCall - Optional callback invoked when a tool is called (for UI feedback)
+     * @returns The complete generated text
+     */
+    stream(prompt: string, onToken: (token: string) => void, onToolCall?: (toolName: string, args: string) => void): Promise<string>;
+    streamWithEvents(prompt: string, onEvent: (eventJson: string) => void): Promise<string>;
+    /**
+     * Stop the current generation.
+     */
+    stop(): void;
+    /**
+     * Unload the current model and release memory.
+     */
+    unload(): void;
+    /**
+     * Get statistics from the last generation.
+     * @returns Statistics including token count, speed, and timing
+     */
+    getLastGenerationStats(): GenerationStats;
+    /**
+     * Get the message history if management is enabled.
+     * @returns Array of messages in the history
+     */
+    getHistory(): LLMMessage[];
+    /**
+     * Clear the message history.
+     */
+    clearHistory(): void;
+    /** Whether a model is currently loaded */
+    readonly isLoaded: boolean;
+    /** Whether text is currently being generated */
+    readonly isGenerating: boolean;
+    /** The ID of the currently loaded model */
+    readonly modelId: string;
+    /** Enable debug logging */
+    debug: boolean;
+    /** System prompt used when loading the model */
+    systemPrompt: string;
+}
+/**
+ * Supported parameter types for tool definitions.
+ * Used for type safety in createTool().
+ */
+export type ToolParameterType = 'string' | 'number' | 'boolean' | 'array' | 'object';
+//# sourceMappingURL=LLM.nitro.d.ts.map

package/lib/typescript/src/specs/LLM.nitro.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"LLM.nitro.d.ts","sourceRoot":"","sources":["../../../../src/specs/LLM.nitro.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,4BAA4B,CAAA;AAEtE;;GAEG;AACH,MAAM,WAAW,eAAe;IAC9B,UAAU,EAAE,MAAM,CAAA;IAClB,eAAe,EAAE,MAAM,CAAA;IACvB,gBAAgB,EAAE,MAAM,CAAA;IACxB,SAAS,EAAE,MAAM,CAAA;IACjB,iBAAiB,EAAE,MAAM,CAAA;CAC1B;AAED,MAAM,WAAW,oBAAoB;IACnC,IAAI,EAAE,kBAAkB,CAAA;IACxB,SAAS,EAAE,MAAM,CAAA;CAClB;AAED,MAAM,WAAW,UAAU;IACzB,IAAI,EAAE,OAAO,CAAA;IACb,KAAK,EAAE,MAAM,CAAA;CACd;AAED,MAAM,WAAW,kBAAkB;IACjC,IAAI,EAAE,gBAAgB,CAAA;IACtB,SAAS,EAAE,MAAM,CAAA;CAClB;AAED,MAAM,WAAW,kBAAkB;IACjC,IAAI,EAAE,gBAAgB,CAAA;IACtB,KAAK,EAAE,MAAM,CAAA;CACd;AAED,MAAM,WAAW,gBAAgB;IAC/B,IAAI,EAAE,cAAc,CAAA;IACpB,OAAO,EAAE,MAAM,CAAA;IACf,SAAS,EAAE,MAAM,CAAA;CAClB;AAED,MAAM,WAAW,kBAAkB;IACjC,IAAI,EAAE,iBAAiB,CAAA;IACvB,EAAE,EAAE,MAAM,CAAA;IACV,IAAI,EAAE,MAAM,CAAA;IACZ,SAAS,EAAE,MAAM,CAAA;CAClB;AAED,MAAM,WAAW,sBAAsB;IACrC,IAAI,EAAE,qBAAqB,CAAA;IAC3B,EAAE,EAAE,MAAM,CAAA;CACX;AAED,MAAM,WAAW,sBAAsB;IACrC,IAAI,EAAE,qBAAqB,CAAA;IAC3B,EAAE,EAAE,MAAM,CAAA;IACV,MAAM,EAAE,MAAM,CAAA;CACf;AAED,MAAM,WAAW,mBAAmB;IAClC,IAAI,EAAE,kBAAkB,CAAA;IACxB,EAAE,EAAE,MAAM,CAAA;IACV,KAAK,EAAE,MAAM,CAAA;CACd;AAED,MAAM,WAAW,kBAAkB;IACjC,IAAI,EAAE,gBAAgB,CAAA;IACtB,OAAO,EAAE,MAAM,CAAA;IACf,KAAK,EAAE,eAAe,CAAA;CACvB;AAED,MAAM,MAAM,WAAW,GACnB,oBAAoB,GACpB,UAAU,GACV,kBAAkB,GAClB,kBAAkB,GAClB,gBAAgB,GAChB,kBAAkB,GAClB,sBAAsB,GACtB,sBAAsB,GACtB,mBAAmB,GACnB,kBAAkB,CAAA;AAEtB,MAAM,WAAW,UAAU;IACzB,IAAI,EAAE,MAAM,CAAA;IACZ,OAAO,EAAE,MAAM,CAAA;CAChB;AAED;;GAEG;AACH,MAAM,WAAW,aAAa;IAC5B,IAAI,EAAE,MAAM,CAAA;IACZ,IAAI,EAAE,MAAM,CAAA;IACZ,WAAW,EAAE,MAAM,CAAA;IACnB,QAAQ,EAAE,OAAO,CAAA;CAClB;AAED;;GAEG;AACH,MAAM,WAAW,cAAc;IAC7B,IAAI,EAAE,MAAM,CAAA;IACZ,WAAW,EAAE,MAAM,CAAA;IACnB,UAAU,EAAE,aAAa,EAAE,CAAA;IAC3B,OAAO,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,OAAO,CAAC,MAAM,CAAC,CAAA;CAC3C;AAED;GACG;AACH,MAAM,WAAW,cAAc;IAC7B,mDAAmD;IACnD,UAAU,CAAC,EAAE,CAAC,QAAQ,EAAE,MAAM,KAAK,IAAI,CAAA;IACvC,iDAAiD;IACjD,iBAAiB,CAAC,EAAE,UAAU,EAAE,CAAA;IAChC,sDAAsD;IACtD,aAAa,CAAC,EAAE,OAAO,CAAA;IACvB,4CAA4C;IAC5C,KAAK,CAAC,EAAE,cAAc,EAAE,CAAA;CACzB;AAED;;;GAGG;AACH,MAAM,WAAW,GAAI,SAAQ,YAAY,CAAC;IAAE,GAAG,EAAE,OAAO,CAAA;CAAE,CAAC;IACzD;;;;OAIG;IACH,IAAI,CAAC,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,cAAc,GAAG,OAAO,CAAC,IAAI,CAAC,CAAA;IAE9D;;;;OAIG;IACH,QAAQ,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAA;IAEzC;;;;;;;OAOG;IACH,MAAM,CACJ,MAAM,EAAE,MAAM,EACd,OAAO,EAAE,CAAC,KAAK,EAAE,MAAM,KAAK,IAAI,EAChC,UAAU,CAAC,EAAE,CAAC,QAAQ,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,KAAK,IAAI,GACpD,OAAO,CAAC,MAAM,CAAC,CAAA;IAElB,gBAAgB,CACd,MAAM,EAAE,MAAM,EACd,OAAO,EAAE,CAAC,SAAS,EAAE,MAAM,KAAK,IAAI,GACnC,OAAO,CAAC,MAAM,CAAC,CAAA;IAElB;;OAEG;IACH,IAAI,IAAI,IAAI,CAAA;IAEZ;;OAEG;IACH,MAAM,IAAI,IAAI,CAAA;IAEd;;;OAGG;IACH,sBAAsB,IAAI,eAAe,CAAA;IAEzC;;;OAGG;IACH,UAAU,IAAI,UAAU,EAAE,CAAA;IAE1B;;OAEG;IACH,YAAY,IAAI,IAAI,CAAA;IAEpB,0CAA0C;IAC1C,QAAQ,CAAC,QAAQ,EAAE,OAAO,CAAA;IAC1B,gDAAgD;IAChD,QAAQ,CAAC,YAAY,EAAE,OAAO,CAAA;IAC9B,2CAA2C;IAC3C,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAA;IAExB,2BAA2B;IAC3B,KAAK,EAAE,OAAO,CAAA;IACd,gDAAgD;IAChD,YAAY,EAAE,MAAM,CAAA;CACrB;AAED;;;GAGG;AACH,MAAM,MAAM,iBAAiB,GAAG,QAAQ,GAAG,QAAQ,GAAG,SAAS,GAAG,OAAO,GAAG,QAAQ,CAAA"}

package/lib/typescript/src/specs/ModelManager.nitro.d.ts

@@ -0,0 +1,41 @@
+import type { HybridObject } from 'react-native-nitro-modules';
+/**
+ * Low-level interface for managing MLX model downloads.
+ * @internal Use the `ModelManager` export from `react-native-nitro-mlx` instead.
+ */
+export interface ModelManager extends HybridObject<{
+    ios: 'swift';
+}> {
+    /**
+     * Download a model from HuggingFace.
+     * @param modelId - HuggingFace model ID (e.g., 'mlx-community/Qwen3-0.6B-4bit')
+     * @param progressCallback - Callback invoked with download progress (0-1)
+     * @returns Path to the downloaded model directory
+     */
+    download(modelId: string, progressCallback: (progress: number) => void): Promise<string>;
+    /**
+     * Check if a model is already downloaded.
+     * @param modelId - HuggingFace model ID
+     * @returns True if the model is downloaded
+     */
+    isDownloaded(modelId: string): Promise<boolean>;
+    /**
+     * Get a list of all downloaded model IDs.
+     * @returns Array of downloaded model IDs
+     */
+    getDownloadedModels(): Promise<string[]>;
+    /**
+     * Delete a downloaded model.
+     * @param modelId - HuggingFace model ID
+     */
+    deleteModel(modelId: string): Promise<void>;
+    /**
+     * Get the local filesystem path for a downloaded model.
+     * @param modelId - HuggingFace model ID
+     * @returns Path to the model directory
+     */
+    getModelPath(modelId: string): Promise<string>;
+    /** Enable debug logging */
+    debug: boolean;
+}
+//# sourceMappingURL=ModelManager.nitro.d.ts.map

package/lib/typescript/src/specs/ModelManager.nitro.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"ModelManager.nitro.d.ts","sourceRoot":"","sources":["../../../../src/specs/ModelManager.nitro.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,4BAA4B,CAAA;AAE9D;;;GAGG;AACH,MAAM,WAAW,YAAa,SAAQ,YAAY,CAAC;IAAE,GAAG,EAAE,OAAO,CAAA;CAAE,CAAC;IAClE;;;;;OAKG;IACH,QAAQ,CAAC,OAAO,EAAE,MAAM,EAAE,gBAAgB,EAAE,CAAC,QAAQ,EAAE,MAAM,KAAK,IAAI,GAAG,OAAO,CAAC,MAAM,CAAC,CAAA;IAExF;;;;OAIG;IACH,YAAY,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC,OAAO,CAAC,CAAA;IAE/C;;;OAGG;IACH,mBAAmB,IAAI,OAAO,CAAC,MAAM,EAAE,CAAC,CAAA;IAExC;;;OAGG;IACH,WAAW,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAAA;IAE3C;;;;OAIG;IACH,YAAY,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAA;IAE9C,2BAA2B;IAC3B,KAAK,EAAE,OAAO,CAAA;CACf"}

package/lib/typescript/src/specs/STT.nitro.d.ts

@@ -0,0 +1,28 @@
+import type { HybridObject } from 'react-native-nitro-modules';
+export interface STTLoadOptions {
+    onProgress?: (progress: number) => void;
+}
+export interface STTTranscriptionInfo {
+    promptTokens: number;
+    generationTokens: number;
+    tokensPerSecond: number;
+    prefillTime: number;
+    generateTime: number;
+}
+export interface STT extends HybridObject<{
+    ios: 'swift';
+}> {
+    readonly isLoaded: boolean;
+    readonly isTranscribing: boolean;
+    readonly isListening: boolean;
+    readonly modelId: string;
+    load(modelId: string, options?: STTLoadOptions): Promise<void>;
+    transcribe(audio: ArrayBuffer): Promise<string>;
+    transcribeStream(audio: ArrayBuffer, onToken: (token: string) => void): Promise<string>;
+    startListening(): Promise<void>;
+    transcribeBuffer(): Promise<string>;
+    stopListening(): Promise<string>;
+    stop(): void;
+    unload(): void;
+}
+//# sourceMappingURL=STT.nitro.d.ts.map

package/lib/typescript/src/specs/STT.nitro.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"STT.nitro.d.ts","sourceRoot":"","sources":["../../../../src/specs/STT.nitro.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,4BAA4B,CAAA;AAE9D,MAAM,WAAW,cAAc;IAC7B,UAAU,CAAC,EAAE,CAAC,QAAQ,EAAE,MAAM,KAAK,IAAI,CAAA;CACxC;AAED,MAAM,WAAW,oBAAoB;IACnC,YAAY,EAAE,MAAM,CAAA;IACpB,gBAAgB,EAAE,MAAM,CAAA;IACxB,eAAe,EAAE,MAAM,CAAA;IACvB,WAAW,EAAE,MAAM,CAAA;IACnB,YAAY,EAAE,MAAM,CAAA;CACrB;AAED,MAAM,WAAW,GAAI,SAAQ,YAAY,CAAC;IAAE,GAAG,EAAE,OAAO,CAAA;CAAE,CAAC;IACzD,QAAQ,CAAC,QAAQ,EAAE,OAAO,CAAA;IAC1B,QAAQ,CAAC,cAAc,EAAE,OAAO,CAAA;IAChC,QAAQ,CAAC,WAAW,EAAE,OAAO,CAAA;IAC7B,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAA;IAExB,IAAI,CAAC,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,cAAc,GAAG,OAAO,CAAC,IAAI,CAAC,CAAA;IAE9D,UAAU,CAAC,KAAK,EAAE,WAAW,GAAG,OAAO,CAAC,MAAM,CAAC,CAAA;IAC/C,gBAAgB,CACd,KAAK,EAAE,WAAW,EAClB,OAAO,EAAE,CAAC,KAAK,EAAE,MAAM,KAAK,IAAI,GAC/B,OAAO,CAAC,MAAM,CAAC,CAAA;IAElB,cAAc,IAAI,OAAO,CAAC,IAAI,CAAC,CAAA;IAC/B,gBAAgB,IAAI,OAAO,CAAC,MAAM,CAAC,CAAA;IACnC,aAAa,IAAI,OAAO,CAAC,MAAM,CAAC,CAAA;IAEhC,IAAI,IAAI,IAAI,CAAA;IACZ,MAAM,IAAI,IAAI,CAAA;CACf"}

package/lib/typescript/src/specs/TTS.nitro.d.ts

@@ -0,0 +1,22 @@
+import type { HybridObject } from 'react-native-nitro-modules';
+export interface TTSLoadOptions {
+    onProgress?: (progress: number) => void;
+}
+export interface TTSGenerateOptions {
+    voice?: string;
+    speed?: number;
+}
+export interface TTS extends HybridObject<{
+    ios: 'swift';
+}> {
+    readonly isLoaded: boolean;
+    readonly isGenerating: boolean;
+    readonly modelId: string;
+    readonly sampleRate: number;
+    load(modelId: string, options?: TTSLoadOptions): Promise<void>;
+    generate(text: string, options?: TTSGenerateOptions): Promise<ArrayBuffer>;
+    stream(text: string, onAudioChunk: (audio: ArrayBuffer) => void, options?: TTSGenerateOptions): Promise<void>;
+    stop(): void;
+    unload(): void;
+}
+//# sourceMappingURL=TTS.nitro.d.ts.map

package/lib/typescript/src/specs/TTS.nitro.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"TTS.nitro.d.ts","sourceRoot":"","sources":["../../../../src/specs/TTS.nitro.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,4BAA4B,CAAA;AAE9D,MAAM,WAAW,cAAc;IAC7B,UAAU,CAAC,EAAE,CAAC,QAAQ,EAAE,MAAM,KAAK,IAAI,CAAA;CACxC;AAED,MAAM,WAAW,kBAAkB;IACjC,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,KAAK,CAAC,EAAE,MAAM,CAAA;CACf;AAED,MAAM,WAAW,GAAI,SAAQ,YAAY,CAAC;IAAE,GAAG,EAAE,OAAO,CAAA;CAAE,CAAC;IACzD,QAAQ,CAAC,QAAQ,EAAE,OAAO,CAAA;IAC1B,QAAQ,CAAC,YAAY,EAAE,OAAO,CAAA;IAC9B,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAA;IACxB,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAA;IAE3B,IAAI,CAAC,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,cAAc,GAAG,OAAO,CAAC,IAAI,CAAC,CAAA;IAC9D,QAAQ,CACN,IAAI,EAAE,MAAM,EACZ,OAAO,CAAC,EAAE,kBAAkB,GAC3B,OAAO,CAAC,WAAW,CAAC,CAAA;IACvB,MAAM,CACJ,IAAI,EAAE,MAAM,EACZ,YAAY,EAAE,CAAC,KAAK,EAAE,WAAW,KAAK,IAAI,EAC1C,OAAO,CAAC,EAAE,kBAAkB,GAC3B,OAAO,CAAC,IAAI,CAAC,CAAA;IAChB,IAAI,IAAI,IAAI,CAAA;IACZ,MAAM,IAAI,IAAI,CAAA;CACf"}

package/lib/typescript/src/stt.d.ts

@@ -0,0 +1,16 @@
+import type { STTLoadOptions } from './specs/STT.nitro';
+export declare const STT: {
+    load(modelId: string, options?: STTLoadOptions): Promise<void>;
+    transcribe(audio: ArrayBuffer): Promise<string>;
+    transcribeStream(audio: ArrayBuffer, onToken: (token: string) => void): Promise<string>;
+    startListening(): Promise<void>;
+    transcribeBuffer(): Promise<string>;
+    stopListening(): Promise<string>;
+    stop(): void;
+    unload(): void;
+    readonly isLoaded: boolean;
+    readonly isTranscribing: boolean;
+    readonly isListening: boolean;
+    readonly modelId: string;
+};
+//# sourceMappingURL=stt.d.ts.map

package/lib/typescript/src/stt.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"stt.d.ts","sourceRoot":"","sources":["../../../src/stt.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAEV,cAAc,EACf,MAAM,mBAAmB,CAAA;AAW1B,eAAO,MAAM,GAAG;kBACA,MAAM,YAAY,cAAc,GAAG,OAAO,CAAC,IAAI,CAAC;sBAI5C,WAAW,GAAG,OAAO,CAAC,MAAM,CAAC;4BAKtC,WAAW,WACT,CAAC,KAAK,EAAE,MAAM,KAAK,IAAI,GAC/B,OAAO,CAAC,MAAM,CAAC;sBAIA,OAAO,CAAC,IAAI,CAAC;wBAIX,OAAO,CAAC,MAAM,CAAC;qBAIlB,OAAO,CAAC,MAAM,CAAC;YAIxB,IAAI;cAIF,IAAI;uBAIE,OAAO;6BAID,OAAO;0BAIV,OAAO;sBAIX,MAAM;CAGtB,CAAA"}

package/lib/typescript/src/tool-utils.d.ts

@@ -0,0 +1,13 @@
+import type { z } from 'zod';
+import type { ToolDefinition } from './specs/LLM.nitro';
+type ZodObjectSchema = z.ZodObject<z.core.$ZodShape>;
+type InferArgs<T extends ZodObjectSchema> = z.infer<T>;
+export interface TypeSafeToolDefinition<T extends ZodObjectSchema> {
+    name: string;
+    description: string;
+    arguments: T;
+    handler: (args: InferArgs<T>) => Promise<Record<string, unknown>>;
+}
+export declare function createTool<T extends ZodObjectSchema>(definition: TypeSafeToolDefinition<T>): ToolDefinition;
+export {};
+//# sourceMappingURL=tool-utils.d.ts.map

package/lib/typescript/src/tool-utils.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"tool-utils.d.ts","sourceRoot":"","sources":["../../../src/tool-utils.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,CAAC,EAAE,MAAM,KAAK,CAAA;AAC5B,OAAO,KAAK,EAAE,cAAc,EAAoC,MAAM,mBAAmB,CAAA;AAEzF,KAAK,eAAe,GAAG,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,CAAA;AACpD,KAAK,SAAS,CAAC,CAAC,SAAS,eAAe,IAAI,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAA;AAEtD,MAAM,WAAW,sBAAsB,CAAC,CAAC,SAAS,eAAe;IAC/D,IAAI,EAAE,MAAM,CAAA;IACZ,WAAW,EAAE,MAAM,CAAA;IACnB,SAAS,EAAE,CAAC,CAAA;IACZ,OAAO,EAAE,CAAC,IAAI,EAAE,SAAS,CAAC,CAAC,CAAC,KAAK,OAAO,CAAC,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,CAAA;CAClE;AA+CD,wBAAgB,UAAU,CAAC,CAAC,SAAS,eAAe,EAClD,UAAU,EAAE,sBAAsB,CAAC,CAAC,CAAC,GACpC,cAAc,CAYhB"}

package/lib/typescript/src/tts.d.ts

@@ -0,0 +1,13 @@
+import type { TTSLoadOptions, TTSGenerateOptions } from './specs/TTS.nitro';
+export declare const TTS: {
+    load(modelId: string, options?: TTSLoadOptions): Promise<void>;
+    generate(text: string, options?: TTSGenerateOptions): Promise<ArrayBuffer>;
+    stream(text: string, onAudioChunk: (audio: ArrayBuffer) => void, options?: TTSGenerateOptions): Promise<void>;
+    stop(): void;
+    unload(): void;
+    readonly isLoaded: boolean;
+    readonly isGenerating: boolean;
+    readonly modelId: string;
+    readonly sampleRate: number;
+};
+//# sourceMappingURL=tts.d.ts.map

package/lib/typescript/src/tts.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"tts.d.ts","sourceRoot":"","sources":["../../../src/tts.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAEV,cAAc,EACd,kBAAkB,EACnB,MAAM,mBAAmB,CAAA;AAW1B,eAAO,MAAM,GAAG;kBACA,MAAM,YAAY,cAAc,GAAG,OAAO,CAAC,IAAI,CAAC;mBAKtD,MAAM,YACF,kBAAkB,GAC3B,OAAO,CAAC,WAAW,CAAC;iBAKf,MAAM,gBACE,CAAC,KAAK,EAAE,WAAW,KAAK,IAAI,YAChC,kBAAkB,GAC3B,OAAO,CAAC,IAAI,CAAC;YAIR,IAAI;cAIF,IAAI;uBAIE,OAAO;2BAIH,OAAO;sBAIZ,MAAM;yBAIH,MAAM;CAGzB,CAAA"}

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@inferrlm/react-native-mlx",
   "description": "MLX Swift integration for React Native - InferrLM fork with enhanced features",
-  "version": "0.4.0",
+  "version": "0.4.2-alpha.0",
   "main": "./lib/module/index.js",
   "module": "./lib/module/index.js",
   "types": "./lib/typescript/src/index.d.ts",

package/src/models.ts

@@ -53,6 +53,10 @@ export enum MLXModel {
   Qwen3_1_7B_4bit = 'mlx-community/Qwen3-1.7B-4bit',
   Qwen3_1_7B_8bit = 'mlx-community/Qwen3-1.7B-8bit',
 
+  // Qwen 3.5 - 0.8B variant
+  Qwen3_5_0_8B_MLX_4bit = 'mlx-community/Qwen3.5-0.8B-MLX-4bit',
+  Qwen3_5_0_8B_MLX_8bit = 'mlx-community/Qwen3.5-0.8B-MLX-8bit',
+
   // Gemma 3 (Google) - 1B variant
   Gemma_3_1B_IT_4bit = 'mlx-community/gemma-3-1b-it-4bit',
   Gemma_3_1B_IT_8bit = 'mlx-community/gemma-3-1b-it-8bit',
@@ -209,6 +213,26 @@ export const MLXModels: ModelInfo[] = [
     downloadSize: 1839729195,
     type: 'llm',
   },
+  {
+    id: MLXModel.Qwen3_5_0_8B_MLX_4bit,
+    family: ModelFamily.Qwen,
+    provider: ModelProvider.Alibaba,
+    parameters: '0.8B',
+    quantization: '4bit',
+    displayName: 'Qwen 3.5 0.8B (4-bit)',
+    downloadSize: 550000000,
+    type: 'llm',
+  },
+  {
+    id: MLXModel.Qwen3_5_0_8B_MLX_8bit,
+    family: ModelFamily.Qwen,
+    provider: ModelProvider.Alibaba,
+    parameters: '0.8B',
+    quantization: '8bit',
+    displayName: 'Qwen 3.5 0.8B (8-bit)',
+    downloadSize: 950000000,
+    type: 'llm',
+  },
   {
     id: MLXModel.Gemma_3_1B_IT_4bit,
     family: ModelFamily.Gemma,