react-native-executorch 0.5.1 → 0.5.3

package/common/rnexecutorch/models/vertical_ocr/VerticalOCR.h

@@ -2,18 +2,21 @@
 
 #include <executorch/extension/tensor/tensor_ptr.h>
 #include <opencv2/opencv.hpp>
+#include <string>
+#include <utility>
+#include <vector>
+
+#include "rnexecutorch/metaprogramming/ConstructorHelpers.h"
 #include <rnexecutorch/models/BaseModel.h>
 #include <rnexecutorch/models/ocr/CTCLabelConverter.h>
-#include <rnexecutorch/models/ocr/DetectorUtils.h>
-#include <rnexecutorch/models/ocr/RecognitionHandlerUtils.h>
 #include <rnexecutorch/models/ocr/Recognizer.h>
-#include <rnexecutorch/models/ocr/RecognizerUtils.h>
+#include <rnexecutorch/models/ocr/utils/DetectorUtils.h>
+#include <rnexecutorch/models/ocr/utils/RecognitionHandlerUtils.h>
+#include <rnexecutorch/models/ocr/utils/RecognizerUtils.h>
 #include <rnexecutorch/models/vertical_ocr/VerticalDetector.h>
-#include <string>
-#include <utility>
-#include <vector>
 
 namespace rnexecutorch {
+namespace models::ocr {
 
 /*
   Vertical OCR is OCR designed to handle vertical texts.
@@ -47,32 +50,36 @@ public:
                        const std::string &recognizerSource, std::string symbols,
                        bool indpendentCharacters,
                        std::shared_ptr<react::CallInvoker> callInvoker);
-  std::vector<OCRDetection> generate(std::string input);
+  std::vector<types::OCRDetection> generate(std::string input);
   std::size_t getMemoryLowerBound() const noexcept;
   void unload() noexcept;
 
 private:
   std::pair<std::string, float> _handleIndependentCharacters(
-      const ocr::DetectorBBox &box, const cv::Mat &originalImage,
-      const std::vector<ocr::DetectorBBox> &characterBoxes,
-      const ocr::PaddingInfo &paddingsBox,
-      const ocr::PaddingInfo &imagePaddings);
+      const types::DetectorBBox &box, const cv::Mat &originalImage,
+      const std::vector<types::DetectorBBox> &characterBoxes,
+      const types::PaddingInfo &paddingsBox,
+      const types::PaddingInfo &imagePaddings);
   std::pair<std::string, float>
-  _handleJointCharacters(const ocr::DetectorBBox &box,
+  _handleJointCharacters(const types::DetectorBBox &box,
                          const cv::Mat &originalImage,
-                         const std::vector<ocr::DetectorBBox> &characterBoxes,
-                         const ocr::PaddingInfo &paddingsBox,
-                         const ocr::PaddingInfo &imagePaddings);
-  OCRDetection _processSingleTextBox(ocr::DetectorBBox &box,
-                                     const cv::Mat &originalImage,
-                                     const cv::Mat &resizedLargeImage,
-                                     const ocr::PaddingInfo &imagePaddings);
+                         const std::vector<types::DetectorBBox> &characterBoxes,
+                         const types::PaddingInfo &paddingsBox,
+                         const types::PaddingInfo &imagePaddings);
+  types::OCRDetection
+  _processSingleTextBox(types::DetectorBBox &box, const cv::Mat &originalImage,
+                        const cv::Mat &resizedLargeImage,
+                        const types::PaddingInfo &imagePaddings);
   VerticalDetector detectorLarge;
   VerticalDetector detectorNarrow;
   Recognizer recognizer;
-  ocr::CTCLabelConverter converter;
+  CTCLabelConverter converter;
   bool independentCharacters;
   std::shared_ptr<react::CallInvoker> callInvoker;
 };
+} // namespace models::ocr
 
+REGISTER_CONSTRUCTOR(models::ocr::VerticalOCR, std::string, std::string,
+                     std::string, std::string, bool,
+                     std::shared_ptr<react::CallInvoker>);
 } // namespace rnexecutorch

package/{third-party/ios/ExecutorchLib/ExecutorchLib/sampler → common/runner}/sampler.cpp

@@ -184,9 +184,10 @@ template <typename T> int32_t Sampler::sample(T *logits) {
 }
 
 template int32_t Sampler::sample<float>(float *logits);
-template int32_t Sampler::sample<exec_aten::Half>(exec_aten::Half *logits);
 template int32_t
-Sampler::sample<exec_aten::BFloat16>(exec_aten::BFloat16 *logits);
+Sampler::sample<executorch::aten::Half>(executorch::aten::Half *logits);
+template int32_t
+Sampler::sample<executorch::aten::BFloat16>(executorch::aten::BFloat16 *logits);
 
 } // namespace llm
 } // namespace extension

package/{third-party/ios/ExecutorchLib/ExecutorchLib/sampler → common/runner}/sampler.h

@@ -19,18 +19,19 @@
 #endif
 
 #include <executorch/runtime/core/exec_aten/exec_aten.h>
+#include <executorch/runtime/platform/compiler.h>
 
 namespace executorch {
 namespace extension {
 namespace llm {
 // A simple llama2 sampler.
 
-template <typename T> struct ProbIndex {
+template <typename T> struct ET_EXPERIMENTAL ProbIndex {
   T prob;
   int32_t index;
 }; // struct used when sorting probabilities during top-p sampling
 
-class Sampler {
+class ET_EXPERIMENTAL Sampler {
 public:
   Sampler(int32_t vocab_size, float temperature, float topp,
           unsigned long long rng_seed);

package/lib/Error.js

@@ -0,0 +1,53 @@
+export var ETError;
+(function (ETError) {
+    // React-native-ExecuTorch errors
+    ETError[ETError["UndefinedError"] = 101] = "UndefinedError";
+    ETError[ETError["ModuleNotLoaded"] = 102] = "ModuleNotLoaded";
+    ETError[ETError["FileWriteFailed"] = 103] = "FileWriteFailed";
+    ETError[ETError["ModelGenerating"] = 104] = "ModelGenerating";
+    ETError[ETError["LanguageNotSupported"] = 105] = "LanguageNotSupported";
+    ETError[ETError["InvalidModelSource"] = 255] = "InvalidModelSource";
+    //SpeechToText errors
+    ETError[ETError["MultilingualConfiguration"] = 160] = "MultilingualConfiguration";
+    ETError[ETError["MissingDataChunk"] = 161] = "MissingDataChunk";
+    ETError[ETError["StreamingNotStarted"] = 162] = "StreamingNotStarted";
+    // ExecuTorch mapped errors
+    // Based on: https://github.com/pytorch/executorch/blob/main/runtime/core/error.h
+    // System errors
+    ETError[ETError["Ok"] = 0] = "Ok";
+    ETError[ETError["Internal"] = 1] = "Internal";
+    ETError[ETError["InvalidState"] = 2] = "InvalidState";
+    ETError[ETError["EndOfMethod"] = 3] = "EndOfMethod";
+    // Logical errors
+    ETError[ETError["NotSupported"] = 16] = "NotSupported";
+    ETError[ETError["NotImplemented"] = 17] = "NotImplemented";
+    ETError[ETError["InvalidArgument"] = 18] = "InvalidArgument";
+    ETError[ETError["InvalidType"] = 19] = "InvalidType";
+    ETError[ETError["OperatorMissing"] = 20] = "OperatorMissing";
+    // Resource errors
+    ETError[ETError["NotFound"] = 32] = "NotFound";
+    ETError[ETError["MemoryAllocationFailed"] = 33] = "MemoryAllocationFailed";
+    ETError[ETError["AccessFailed"] = 34] = "AccessFailed";
+    ETError[ETError["InvalidProgram"] = 35] = "InvalidProgram";
+    ETError[ETError["InvalidExternalData"] = 36] = "InvalidExternalData";
+    ETError[ETError["OutOfResources"] = 37] = "OutOfResources";
+    // Delegate errors
+    ETError[ETError["DelegateInvalidCompatibility"] = 48] = "DelegateInvalidCompatibility";
+    ETError[ETError["DelegateMemoryAllocationFailed"] = 49] = "DelegateMemoryAllocationFailed";
+    ETError[ETError["DelegateInvalidHandle"] = 50] = "DelegateInvalidHandle";
+})(ETError || (ETError = {}));
+export const getError = (e) => {
+    if (typeof e === 'number') {
+        if (e in ETError)
+            return ETError[e];
+        return ETError[ETError.UndefinedError];
+    }
+    // try to extract number from message (can contain false positives)
+    const error = e;
+    const errorCode = parseInt(error.message, 10);
+    const message = Number.isNaN(errorCode)
+        ? error.message
+        : ' ' + error.message.slice(`${errorCode}`.length).trimStart();
+    const ETErrorMessage = (errorCode in ETError ? ETError[errorCode] : ETError[ETError.UndefinedError]);
+    return ETErrorMessage + message;
+};

package/lib/ThreadPool.d.ts

@@ -0,0 +1,10 @@
+declare class NativeThreadPool {
+    constructor();
+    executeNative(functionName: any, args: any): Promise<any>;
+    runInference(prompt: any, options?: {}): Promise<any>;
+    processImage(imagePath: any, options?: {}): Promise<any>;
+    heavyComputation(data: any): Promise<any>;
+    getStats(): any;
+}
+declare const _default: NativeThreadPool;
+export default _default;

package/lib/ThreadPool.js

@@ -0,0 +1,28 @@
+// ThreadPool.js
+class NativeThreadPool {
+    constructor() {
+        if (!global.NativeThreadPool) {
+            throw new Error('NativeThreadPool not installed');
+        }
+        this.pool = global.NativeThreadPool;
+    }
+    // Execute any native function
+    async executeNative(functionName, args) {
+        return await this.pool.executeNative(functionName, JSON.stringify(args));
+    }
+    // Specific methods for common tasks
+    async runInference(prompt, options = {}) {
+        return await this.executeNative('runInference', { prompt, ...options });
+    }
+    async processImage(imagePath, options = {}) {
+        return await this.executeNative('processImage', { imagePath, ...options });
+    }
+    async heavyComputation(data) {
+        return await this.executeNative('heavyComputation', data);
+    }
+    // Get thread pool statistics
+    getStats() {
+        return this.pool.getStats();
+    }
+}
+export default new NativeThreadPool();

package/lib/common/Logger.d.ts

@@ -0,0 +1,8 @@
+export declare class Logger {
+    private static readonly PREFIX;
+    static log(...data: any[]): void;
+    static debug(...data: any[]): void;
+    static info(...data: any[]): void;
+    static warn(...data: any[]): void;
+    static error(...data: any[]): void;
+}

package/lib/common/Logger.js

@@ -0,0 +1,19 @@
+/* eslint-disable no-console */
+export class Logger {
+    static PREFIX = '[React Native ExecuTorch]';
+    static log(...data) {
+        console.log(Logger.PREFIX, ...data);
+    }
+    static debug(...data) {
+        console.debug(Logger.PREFIX, ...data);
+    }
+    static info(...data) {
+        console.info(Logger.PREFIX, ...data);
+    }
+    static warn(...data) {
+        console.warn(Logger.PREFIX, ...data);
+    }
+    static error(...data) {
+        console.error(Logger.PREFIX, ...data);
+    }
+}

package/lib/constants/directories.js

@@ -0,0 +1,2 @@
+import { documentDirectory } from 'expo-file-system';
+export const RNEDirectory = `${documentDirectory}react-native-executorch/`;

package/lib/constants/llmDefaults.d.ts

@@ -0,0 +1,6 @@
+import { ChatConfig, Message } from '../types/llm';
+export declare const DEFAULT_SYSTEM_PROMPT = "You are a knowledgeable, efficient, and direct AI assistant. Provide concise answers, focusing on the key information needed. Offer suggestions tactfully when appropriate to improve outcomes. Engage in productive collaboration with the user. Don't return too much text.";
+export declare const DEFAULT_STRUCTURED_OUTPUT_PROMPT: (structuredOutputSchema: string) => string;
+export declare const DEFAULT_MESSAGE_HISTORY: Message[];
+export declare const DEFAULT_CONTEXT_WINDOW_LENGTH = 5;
+export declare const DEFAULT_CHAT_CONFIG: ChatConfig;

package/lib/constants/llmDefaults.js

@@ -0,0 +1,16 @@
+export const DEFAULT_SYSTEM_PROMPT = "You are a knowledgeable, efficient, and direct AI assistant. Provide concise answers, focusing on the key information needed. Offer suggestions tactfully when appropriate to improve outcomes. Engage in productive collaboration with the user. Don't return too much text.";
+export const DEFAULT_STRUCTURED_OUTPUT_PROMPT = (structuredOutputSchema) => `The output should be formatted as a JSON instance that conforms to the JSON schema below.
+
+As an example, for the schema {"properties": {"foo": {"title": "Foo", "description": "a list of strings", "type": "array", "items": {"type": "string"}}}, "required": ["foo"]}
+the object {"foo": ["bar", "baz"]} is a well-formatted instance of the schema. The object {"properties": {"foo": ["bar", "baz"]}} is not well-formatted.
+
+Here is the output schema:
+${structuredOutputSchema}
+`;
+export const DEFAULT_MESSAGE_HISTORY = [];
+export const DEFAULT_CONTEXT_WINDOW_LENGTH = 5;
+export const DEFAULT_CHAT_CONFIG = {
+    systemPrompt: DEFAULT_SYSTEM_PROMPT,
+    initialMessageHistory: DEFAULT_MESSAGE_HISTORY,
+    contextWindowLength: DEFAULT_CONTEXT_WINDOW_LENGTH,
+};

package/lib/constants/modelUrls.d.ts

@@ -0,0 +1,223 @@
+import { AvailableModels } from '../types/stt';
+export declare const LLAMA3_2_3B: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const LLAMA3_2_3B_QLORA: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const LLAMA3_2_3B_SPINQUANT: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const LLAMA3_2_1B: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const LLAMA3_2_1B_QLORA: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const LLAMA3_2_1B_SPINQUANT: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const QWEN3_0_6B: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const QWEN3_0_6B_QUANTIZED: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const QWEN3_1_7B: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const QWEN3_1_7B_QUANTIZED: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const QWEN3_4B: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const QWEN3_4B_QUANTIZED: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const HAMMER2_1_0_5B: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const HAMMER2_1_0_5B_QUANTIZED: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const HAMMER2_1_1_5B: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const HAMMER2_1_1_5B_QUANTIZED: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const HAMMER2_1_3B: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const HAMMER2_1_3B_QUANTIZED: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const SMOLLM2_1_135M: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const SMOLLM2_1_135M_QUANTIZED: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const SMOLLM2_1_360M: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const SMOLLM2_1_360M_QUANTIZED: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const SMOLLM2_1_1_7B: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const SMOLLM2_1_1_7B_QUANTIZED: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const QWEN2_5_0_5B: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const QWEN2_5_0_5B_QUANTIZED: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const QWEN2_5_1_5B: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const QWEN2_5_1_5B_QUANTIZED: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const QWEN2_5_3B: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const QWEN2_5_3B_QUANTIZED: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const PHI_4_MINI_4B: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const PHI_4_MINI_4B_QUANTIZED: {
+    modelSource: string;
+    tokenizerSource: string;
+    tokenizerConfigSource: string;
+};
+export declare const EFFICIENTNET_V2_S: {
+    modelSource: string;
+};
+export declare const SSDLITE_320_MOBILENET_V3_LARGE: {
+    modelSource: string;
+};
+export declare const STYLE_TRANSFER_CANDY: {
+    modelSource: string;
+};
+export declare const STYLE_TRANSFER_MOSAIC: {
+    modelSource: string;
+};
+export declare const STYLE_TRANSFER_RAIN_PRINCESS: {
+    modelSource: string;
+};
+export declare const STYLE_TRANSFER_UDNIE: {
+    modelSource: string;
+};
+export declare const MOONSHINE_TINY: {
+    modelName: AvailableModels;
+    decoderSource: string;
+    encoderSource: string;
+    tokenizerSource: string;
+};
+export declare const WHISPER_TINY: {
+    modelName: AvailableModels;
+    decoderSource: string;
+    encoderSource: string;
+    tokenizerSource: string;
+};
+export declare const WHISPER_TINY_MULTILINGUAL: {
+    modelName: AvailableModels;
+    decoderSource: string;
+    encoderSource: string;
+    tokenizerSource: string;
+};
+export declare const DEEPLAB_V3_RESNET50: {
+    modelSource: string;
+};
+export declare const CLIP_VIT_BASE_PATCH32_IMAGE: {
+    modelSource: string;
+};
+export declare const ALL_MINILM_L6_V2: {
+    modelSource: string;
+    tokenizerSource: string;
+};
+export declare const ALL_MPNET_BASE_V2: {
+    modelSource: string;
+    tokenizerSource: string;
+};
+export declare const MULTI_QA_MINILM_L6_COS_V1: {
+    modelSource: string;
+    tokenizerSource: string;
+};
+export declare const MULTI_QA_MPNET_BASE_DOT_V1: {
+    modelSource: string;
+    tokenizerSource: string;
+};
+export declare const CLIP_VIT_BASE_PATCH32_TEXT: {
+    modelSource: string;
+    tokenizerSource: string;
+};