react-native-executorch 0.5.2 → 0.5.4

package/android/src/main/java/com/swmansion/rnexecutorch/LLM.kt

@@ -1,63 +0,0 @@
-package com.swmansion.rnexecutorch
-
-import android.util.Log
-import com.facebook.react.bridge.Promise
-import com.facebook.react.bridge.ReactApplicationContext
-import org.pytorch.executorch.extension.llm.LlmCallback
-import org.pytorch.executorch.extension.llm.LlmModule
-
-class LLM(
-  reactContext: ReactApplicationContext,
-) : NativeLLMSpec(reactContext),
-  LlmCallback {
-  private var llmModule: LlmModule? = null
-
-  override fun getName(): String = NAME
-
-  override fun initialize() {
-    super.initialize()
-  }
-
-  override fun onResult(result: String) {
-    emitOnToken(result)
-  }
-
-  override fun onStats(tps: Float) {
-    Log.d("rn_executorch", "TPS: $tps")
-  }
-
-  override fun loadLLM(
-    modelSource: String,
-    tokenizerSource: String,
-    promise: Promise,
-  ) {
-    try {
-      llmModule = LlmModule(modelSource, tokenizerSource, 0.7f)
-      promise.resolve("Model loaded successfully")
-    } catch (e: Exception) {
-      promise.reject("Model loading failed", e.message)
-    }
-  }
-
-  override fun forward(
-    input: String,
-    promise: Promise,
-  ) {
-    Thread {
-      llmModule!!.generate(input, this)
-      promise.resolve("Inference completed successfully")
-    }.start()
-  }
-
-  override fun interrupt() {
-    llmModule!!.stop()
-  }
-
-  override fun releaseResources() {
-    llmModule = null
-  }
-
-  companion object {
-    const val NAME = "LLM"
-  }
-}

package/common/rnexecutorch/models/EncoderDecoderBase.cpp

@@ -1,21 +0,0 @@
-#include <rnexecutorch/models/EncoderDecoderBase.h>
-
-namespace rnexecutorch {
-
-EncoderDecoderBase::EncoderDecoderBase(
-    const std::string &encoderPath, const std::string &decoderPath,
-    std::shared_ptr<react::CallInvoker> callInvoker)
-    : callInvoker(callInvoker),
-      encoder_(std::make_unique<BaseModel>(encoderPath, callInvoker)),
-      decoder_(std::make_unique<BaseModel>(decoderPath, callInvoker)) {};
-
-size_t EncoderDecoderBase::getMemoryLowerBound() const noexcept {
-  return encoder_->getMemoryLowerBound() + decoder_->getMemoryLowerBound();
-}
-
-void EncoderDecoderBase::unload() noexcept {
-  encoder_.reset(nullptr);
-  decoder_.reset(nullptr);
-}
-
-} // namespace rnexecutorch

package/common/rnexecutorch/models/EncoderDecoderBase.h

@@ -1,31 +0,0 @@
-#pragma once
-
-#include <ReactCommon/CallInvoker.h>
-#include <memory>
-#include <rnexecutorch/models/BaseModel.h>
-#include <string>
-
-namespace rnexecutorch {
-
-using namespace facebook;
-using executorch::aten::Tensor;
-using executorch::runtime::EValue;
-
-class EncoderDecoderBase {
-public:
-  explicit EncoderDecoderBase(const std::string &encoderPath,
-                              const std::string &decoderPath,
-                              std::shared_ptr<react::CallInvoker> callInvoker);
-  size_t getMemoryLowerBound() const noexcept;
-  void unload() noexcept;
-
-protected:
-  std::shared_ptr<react::CallInvoker> callInvoker;
-  std::unique_ptr<BaseModel> encoder_;
-  std::unique_ptr<BaseModel> decoder_;
-
-private:
-  size_t memorySizeLowerBound;
-};
-
-} // namespace rnexecutorch

package/common/rnexecutorch/models/speech_to_text/SpeechToTextStrategy.h

@@ -1,27 +0,0 @@
-#pragma once
-
-#include "executorch/extension/tensor/tensor_ptr.h"
-#include <rnexecutorch/host_objects/JSTensorViewOut.h>
-#include <span>
-#include <vector>
-
-namespace rnexecutorch {
-
-using TensorPtr = ::executorch::extension::TensorPtr;
-
-class SpeechToTextStrategy {
-public:
-  virtual ~SpeechToTextStrategy() = default;
-
-  virtual TensorPtr prepareAudioInput(std::span<float> waveform) = 0;
-
-  virtual TensorPtr
-  prepareTokenInput(const std::vector<int64_t> &prevTokens) = 0;
-
-  virtual std::string getDecoderMethod() const = 0;
-
-  virtual std::shared_ptr<OwningArrayBuffer> extractOutputToken(
-      const executorch::aten::Tensor &decoderOutputTensor) const = 0;
-};
-
-} // namespace rnexecutorch

package/common/rnexecutorch/models/speech_to_text/WhisperStrategy.cpp

@@ -1,50 +0,0 @@
-#include "executorch/extension/tensor/tensor_ptr.h"
-#include "rnexecutorch/data_processing/dsp.h"
-#include <rnexecutorch/models/speech_to_text/WhisperStrategy.h>
-
-namespace rnexecutorch {
-
-using namespace ::executorch::extension;
-using namespace ::executorch::aten;
-
-TensorPtr WhisperStrategy::prepareAudioInput(std::span<float> waveform) {
-  constexpr auto fftWindowSize = 512;
-  constexpr auto stftHopLength = 160;
-  constexpr auto innerDim = 256;
-  preprocessedData =
-      dsp::stftFromWaveform(waveform, fftWindowSize, stftHopLength);
-  const auto numFrames = preprocessedData.size() / innerDim;
-  std::vector<int32_t> inputShape = {static_cast<int32_t>(numFrames), innerDim};
-  return make_tensor_ptr(std::move(inputShape), std::move(preprocessedData));
-}
-
-TensorPtr
-WhisperStrategy::prepareTokenInput(const std::vector<int64_t> &prevTokens) {
-  tokens32.clear();
-  tokens32.reserve(prevTokens.size());
-  for (auto token : prevTokens) {
-    tokens32.push_back(static_cast<int32_t>(token));
-  }
-  auto tensorSizes = {1, static_cast<int32_t>(tokens32.size())};
-  return make_tensor_ptr(std::move(tensorSizes), std::move(tokens32));
-}
-
-std::shared_ptr<OwningArrayBuffer> WhisperStrategy::extractOutputToken(
-    const executorch::aten::Tensor &decoderOutputTensor) const {
-  const auto innerDim = decoderOutputTensor.size(1);
-  const auto dictSize = decoderOutputTensor.size(2);
-  auto outputNumel = decoderOutputTensor.numel();
-  auto dataPtr =
-      static_cast<const float *>(decoderOutputTensor.const_data_ptr()) +
-      (innerDim - 1) * dictSize;
-
-  std::span<const float> modelOutput(dataPtr, outputNumel / innerDim);
-  auto createBuffer = [](const auto &data, size_t size) {
-    auto buffer = std::make_shared<OwningArrayBuffer>(size);
-    std::memcpy(buffer->data(), data, size);
-    return buffer;
-  };
-  return createBuffer(modelOutput.data(), modelOutput.size_bytes());
-}
-
-} // namespace rnexecutorch

package/common/rnexecutorch/models/speech_to_text/WhisperStrategy.h

@@ -1,25 +0,0 @@
-#pragma once
-
-#include "SpeechToTextStrategy.h"
-#include <span>
-#include <vector>
-
-namespace rnexecutorch {
-
-class WhisperStrategy final : public SpeechToTextStrategy {
-public:
-  TensorPtr prepareAudioInput(std::span<float> waveform) override;
-
-  TensorPtr prepareTokenInput(const std::vector<int64_t> &prevTokens) override;
-
-  std::string getDecoderMethod() const override { return "forward"; }
-
-  std::shared_ptr<OwningArrayBuffer> extractOutputToken(
-      const executorch::aten::Tensor &decoderOutputTensor) const override;
-
-private:
-  std::vector<float> preprocessedData;
-  std::vector<int32_t> tokens32;
-};
-
-} // namespace rnexecutorch

package/ios/ExecutorchLib.xcframework/Info.plist

@@ -1,43 +0,0 @@
-<?xml version="1.0" encoding="UTF-8"?>
-<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
-<plist version="1.0">
-<dict>
-	<key>AvailableLibraries</key>
-	<array>
-		<dict>
-			<key>BinaryPath</key>
-			<string>ExecutorchLib.framework/ExecutorchLib</string>
-			<key>LibraryIdentifier</key>
-			<string>ios-arm64</string>
-			<key>LibraryPath</key>
-			<string>ExecutorchLib.framework</string>
-			<key>SupportedArchitectures</key>
-			<array>
-				<string>arm64</string>
-			</array>
-			<key>SupportedPlatform</key>
-			<string>ios</string>
-		</dict>
-		<dict>
-			<key>BinaryPath</key>
-			<string>ExecutorchLib.framework/ExecutorchLib</string>
-			<key>LibraryIdentifier</key>
-			<string>ios-arm64-simulator</string>
-			<key>LibraryPath</key>
-			<string>ExecutorchLib.framework</string>
-			<key>SupportedArchitectures</key>
-			<array>
-				<string>arm64</string>
-			</array>
-			<key>SupportedPlatform</key>
-			<string>ios</string>
-			<key>SupportedPlatformVariant</key>
-			<string>simulator</string>
-		</dict>
-	</array>
-	<key>CFBundlePackageType</key>
-	<string>XFWK</string>
-	<key>XCFrameworkFormatVersion</key>
-	<string>1.0</string>
-</dict>
-</plist>

package/ios/ExecutorchLib.xcframework/ios-arm64/ExecutorchLib.framework/Headers/ETModel.h

@@ -1,27 +0,0 @@
-#ifndef ETModel_hpp
-#define ETModel_hpp
-
-#import <UIKit/UIKit.h>
-
-@interface ETModel : NSObject
-
-- (NSNumber *)loadModel:(NSString *)filePath;
-- (NSNumber *)loadMethod:(NSString *)methodName;
-- (NSNumber *)loadForward;
-- (NSArray *)execute:(NSString *)methodName
-              inputs:(NSArray *)inputs
-              shapes:(NSArray *)shapes
-          inputTypes:(NSArray *)inputTypes;
-- (NSArray *)forward:(NSArray *)inputs
-              shapes:(NSArray *)shapes
-          inputTypes:(NSArray *)inputTypes;
-- (NSNumber *)getNumberOfInputs;
-- (NSNumber *)getInputType:(NSNumber *)index;
-- (NSArray *)getInputShape:(NSNumber *)index;
-- (NSNumber *)getNumberOfOutputs;
-- (NSNumber *)getOutputType:(NSNumber *)index;
-- (NSArray *)getOutputShape:(NSNumber *)index;
-
-@end
-
-#endif // ETModel_hpp

package/ios/ExecutorchLib.xcframework/ios-arm64/ExecutorchLib.framework/Headers/HuggingFaceTokenizer.h

@@ -1,14 +0,0 @@
-#import <Foundation/Foundation.h>
-
-@interface HuggingFaceTokenizer : NSObject
-
-- (instancetype)initWithTokenizerPath:(NSString *)tokenizerPath;
-- (NSArray<NSNumber *> *)encode:(NSString *)text;
-- (NSString *)decode:(NSArray<NSNumber *> *)tokenIds;
-- (NSString *)decode:(NSArray<NSNumber *> *)tokenIds
-    skipSpecialTokens:(BOOL)skipSpecialTokens;
-- (NSUInteger)getVocabSize;
-- (NSString *)idToToken:(NSInteger)tokenId;
-- (NSInteger)tokenToId:(NSString *)token;
-
-@end

package/ios/ExecutorchLib.xcframework/ios-arm64/ExecutorchLib.framework/Headers/LLaMARunner.h

@@ -1,32 +0,0 @@
-/*
- * Copyright (c) Meta Platforms, Inc. and affiliates.
- * All rights reserved.
- *
- * This source code is licensed under the BSD-style license found in the
- * LICENSE file in the root directory of this source tree.
- */
-
-#import <UIKit/UIKit.h>
-
-NS_ASSUME_NONNULL_BEGIN
-
-FOUNDATION_EXPORT NSErrorDomain const LLaMARunnerErrorDomain;
-
-NS_SWIFT_NAME(Runner)
-@interface LLaMARunner : NSObject
-
-- (instancetype)initWithModelPath:(NSString *)filePath
-                    tokenizerPath:(NSString *)tokenizerPath;
-- (BOOL)isLoaded;
-- (BOOL)loadWithError:(NSError **)error;
-- (BOOL)generate:(NSString *)prompt
-    withTokenCallback:(nullable void (^)(NSString *))callback
-                error:(NSError **)error;
-- (void)stop;
-
-+ (instancetype)new NS_UNAVAILABLE;
-- (instancetype)init NS_UNAVAILABLE;
-
-@end
-
-NS_ASSUME_NONNULL_END

package/ios/ExecutorchLib.xcframework/ios-arm64-simulator/ExecutorchLib.framework/Headers/ETModel.h

@@ -1,27 +0,0 @@
-#ifndef ETModel_hpp
-#define ETModel_hpp
-
-#import <UIKit/UIKit.h>
-
-@interface ETModel : NSObject
-
-- (NSNumber *)loadModel:(NSString *)filePath;
-- (NSNumber *)loadMethod:(NSString *)methodName;
-- (NSNumber *)loadForward;
-- (NSArray *)execute:(NSString *)methodName
-              inputs:(NSArray *)inputs
-              shapes:(NSArray *)shapes
-          inputTypes:(NSArray *)inputTypes;
-- (NSArray *)forward:(NSArray *)inputs
-              shapes:(NSArray *)shapes
-          inputTypes:(NSArray *)inputTypes;
-- (NSNumber *)getNumberOfInputs;
-- (NSNumber *)getInputType:(NSNumber *)index;
-- (NSArray *)getInputShape:(NSNumber *)index;
-- (NSNumber *)getNumberOfOutputs;
-- (NSNumber *)getOutputType:(NSNumber *)index;
-- (NSArray *)getOutputShape:(NSNumber *)index;
-
-@end
-
-#endif // ETModel_hpp

package/ios/ExecutorchLib.xcframework/ios-arm64-simulator/ExecutorchLib.framework/Headers/HuggingFaceTokenizer.h

@@ -1,14 +0,0 @@
-#import <Foundation/Foundation.h>
-
-@interface HuggingFaceTokenizer : NSObject
-
-- (instancetype)initWithTokenizerPath:(NSString *)tokenizerPath;
-- (NSArray<NSNumber *> *)encode:(NSString *)text;
-- (NSString *)decode:(NSArray<NSNumber *> *)tokenIds;
-- (NSString *)decode:(NSArray<NSNumber *> *)tokenIds
-    skipSpecialTokens:(BOOL)skipSpecialTokens;
-- (NSUInteger)getVocabSize;
-- (NSString *)idToToken:(NSInteger)tokenId;
-- (NSInteger)tokenToId:(NSString *)token;
-
-@end

package/ios/ExecutorchLib.xcframework/ios-arm64-simulator/ExecutorchLib.framework/Headers/LLaMARunner.h

@@ -1,32 +0,0 @@
-/*
- * Copyright (c) Meta Platforms, Inc. and affiliates.
- * All rights reserved.
- *
- * This source code is licensed under the BSD-style license found in the
- * LICENSE file in the root directory of this source tree.
- */
-
-#import <UIKit/UIKit.h>
-
-NS_ASSUME_NONNULL_BEGIN
-
-FOUNDATION_EXPORT NSErrorDomain const LLaMARunnerErrorDomain;
-
-NS_SWIFT_NAME(Runner)
-@interface LLaMARunner : NSObject
-
-- (instancetype)initWithModelPath:(NSString *)filePath
-                    tokenizerPath:(NSString *)tokenizerPath;
-- (BOOL)isLoaded;
-- (BOOL)loadWithError:(NSError **)error;
-- (BOOL)generate:(NSString *)prompt
-    withTokenCallback:(nullable void (^)(NSString *))callback
-                error:(NSError **)error;
-- (void)stop;
-
-+ (instancetype)new NS_UNAVAILABLE;
-- (instancetype)init NS_UNAVAILABLE;
-
-@end
-
-NS_ASSUME_NONNULL_END

package/ios/RnExecutorch/LLM.h

@@ -1,5 +0,0 @@
-#import <RnExecutorchSpec/RnExecutorchSpec.h>
-
-@interface LLM : NativeLLMSpecBase <NativeLLMSpec>
-
-@end

package/ios/RnExecutorch/LLM.mm

@@ -1,78 +0,0 @@
-#import "LLM.h"
-#import <ExecutorchLib/LLaMARunner.h>
-
-@implementation LLM {
-  LLaMARunner *runner;
-}
-
-- (instancetype)init {
-  self = [super init];
-
-  return self;
-}
-
-RCT_EXPORT_MODULE()
-
-- (void)onResult:(NSString *)token prompt:(NSString *)prompt {
-  if ([token isEqualToString:prompt]) {
-    return;
-  }
-
-  dispatch_async(dispatch_get_main_queue(), ^{
-    [self emitOnToken:token];
-  });
-}
-
-- (void)loadLLM:(NSString *)modelSource
-    tokenizerSource:(NSString *)tokenizerSource
-            resolve:(RCTPromiseResolveBlock)resolve
-             reject:(RCTPromiseRejectBlock)reject {
-  @try {
-    self->runner = [[LLaMARunner alloc] initWithModelPath:modelSource
-                                            tokenizerPath:tokenizerSource];
-
-    resolve(@"Model and tokenizer loaded successfully");
-    return;
-  } @catch (NSException *exception) {
-    [self releaseResources];
-    reject(@"Model or tokenizer loading failed", exception.reason, nil);
-    return;
-  }
-}
-
-- (void)forward:(NSString *)input
-        resolve:(RCTPromiseResolveBlock)resolve
-         reject:(RCTPromiseRejectBlock)reject {
-
-  dispatch_async(
-      dispatch_get_global_queue(DISPATCH_QUEUE_PRIORITY_DEFAULT, 0), ^{
-        NSError *error = nil;
-        [self->runner generate:input
-             withTokenCallback:^(NSString *token) {
-               [self onResult:token prompt:input];
-             }
-                         error:&error];
-
-        if (error) {
-          reject(@"error_in_generation", error.localizedDescription, nil);
-          return;
-        }
-        resolve(@"Inference completed successfully");
-        return;
-      });
-}
-
-- (void)interrupt {
-  [self->runner stop];
-}
-
-- (void)releaseResources {
-  self->runner = nil;
-}
-
-- (std::shared_ptr<facebook::react::TurboModule>)getTurboModule:
-    (const facebook::react::ObjCTurboModule::InitParams &)params {
-  return std::make_shared<facebook::react::NativeLLMSpecJSI>(params);
-}
-
-@end

package/lib/Error.d.ts

@@ -1,30 +0,0 @@
-export declare enum ETError {
-    UndefinedError = 101,
-    ModuleNotLoaded = 102,
-    FileWriteFailed = 103,
-    ModelGenerating = 104,
-    LanguageNotSupported = 105,
-    InvalidModelSource = 255,
-    MultilingualConfiguration = 160,
-    MissingDataChunk = 161,
-    StreamingNotStarted = 162,
-    Ok = 0,
-    Internal = 1,
-    InvalidState = 2,
-    EndOfMethod = 3,
-    NotSupported = 16,
-    NotImplemented = 17,
-    InvalidArgument = 18,
-    InvalidType = 19,
-    OperatorMissing = 20,
-    NotFound = 32,
-    MemoryAllocationFailed = 33,
-    AccessFailed = 34,
-    InvalidProgram = 35,
-    InvalidExternalData = 36,
-    OutOfResources = 37,
-    DelegateInvalidCompatibility = 48,
-    DelegateMemoryAllocationFailed = 49,
-    DelegateInvalidHandle = 50
-}
-export declare const getError: (e: unknown | ETError | Error) => string;

package/lib/constants/directories.d.ts

@@ -1 +0,0 @@
-export declare const RNEDirectory: string;

package/lib/constants/ocr/symbols.d.ts

@@ -1,75 +0,0 @@
-export declare const alphabets: {
-    cyrillic: string;
-    english: string;
-    latin: string;
-    japanese: string;
-    zhSim: string;
-    korean: string;
-    telugu: string;
-    kannada: string;
-};
-export declare const symbols: {
-    abq: string;
-    ady: string;
-    af: string;
-    ava: string;
-    az: string;
-    be: string;
-    bg: string;
-    bs: string;
-    chSim: string;
-    che: string;
-    cs: string;
-    cy: string;
-    da: string;
-    dar: string;
-    de: string;
-    en: string;
-    es: string;
-    et: string;
-    fr: string;
-    ga: string;
-    hr: string;
-    hu: string;
-    id: string;
-    inh: string;
-    ic: string;
-    it: string;
-    ja: string;
-    kbd: string;
-    kn: string;
-    ko: string;
-    ku: string;
-    la: string;
-    lbe: string;
-    lez: string;
-    lt: string;
-    lv: string;
-    mi: string;
-    mn: string;
-    ms: string;
-    mt: string;
-    nl: string;
-    no: string;
-    oc: string;
-    pi: string;
-    pl: string;
-    pt: string;
-    ro: string;
-    ru: string;
-    rsCyrillic: string;
-    rsLatin: string;
-    sk: string;
-    sl: string;
-    sq: string;
-    sv: string;
-    sw: string;
-    tab: string;
-    te: string;
-    tjk: string;
-    tl: string;
-    tr: string;
-    uk: string;
-    uz: string;
-    vi: string;
-};

package/lib/controllers/OCRController.d.ts

@@ -1,23 +0,0 @@
-import { ResourceSource } from '../types/common';
-import { OCRLanguage } from '../types/ocr';
-export declare class OCRController {
-    private nativeModule;
-    isReady: boolean;
-    isGenerating: boolean;
-    error: string | null;
-    private isReadyCallback;
-    private isGeneratingCallback;
-    private errorCallback;
-    constructor({ isReadyCallback, isGeneratingCallback, errorCallback, }?: {
-        isReadyCallback?: ((_isReady: boolean) => void) | undefined;
-        isGeneratingCallback?: ((_isGenerating: boolean) => void) | undefined;
-        errorCallback?: ((_error: string) => void) | undefined;
-    });
-    load: (detectorSource: ResourceSource, recognizerSources: {
-        recognizerLarge: ResourceSource;
-        recognizerMedium: ResourceSource;
-        recognizerSmall: ResourceSource;
-    }, language: OCRLanguage, onDownloadProgressCallback?: (downloadProgress: number) => void) => Promise<void>;
-    forward: (input: string) => Promise<any>;
-    delete(): void;
-}

package/lib/controllers/VerticalOCRController.d.ts

@@ -1,25 +0,0 @@
-import { ResourceSource } from '../types/common';
-import { OCRLanguage } from '../types/ocr';
-export declare class VerticalOCRController {
-    private ocrNativeModule;
-    isReady: boolean;
-    isGenerating: boolean;
-    error: string | null;
-    private isReadyCallback;
-    private isGeneratingCallback;
-    private errorCallback;
-    constructor({ isReadyCallback, isGeneratingCallback, errorCallback, }?: {
-        isReadyCallback?: ((_isReady: boolean) => void) | undefined;
-        isGeneratingCallback?: ((_isGenerating: boolean) => void) | undefined;
-        errorCallback?: ((_error: string) => void) | undefined;
-    });
-    load: (detectorSources: {
-        detectorLarge: ResourceSource;
-        detectorNarrow: ResourceSource;
-    }, recognizerSources: {
-        recognizerLarge: ResourceSource;
-        recognizerSmall: ResourceSource;
-    }, language: OCRLanguage, independentCharacters: boolean, onDownloadProgressCallback: (downloadProgress: number) => void) => Promise<void>;
-    forward: (input: string) => Promise<any>;
-    delete(): void;
-}