mirror of
https://github.com/ggerganov/whisper.cpp.git
synced 2025-06-16 13:58:09 +00:00
whisper : reorganize source code + improve CMake (#2256)
* scripts : update sync [no ci] * files : reorganize [no ci] * sync : llama.cpp * cmake : link math library * cmake : build normal ggml library * files : move headers to include * objc : fix path to ggml-metal.h * ci : fix WHISPER_CUDA -> GGML_CUDA * scripts : sync LICENSE [no ci]
This commit is contained in:
Georgi Gerganov
GitHub
146
src/coreml/whisper-decoder-impl.h
Normal file
146
src/coreml/whisper-decoder-impl.h
Normal file
@ -0,0 +1,146 @@
|
||||
//
|
||||
// whisper-decoder-impl.h
|
||||
//
|
||||
// This file was automatically generated and should not be edited.
|
||||
//
|
||||
|
||||
#import <Foundation/Foundation.h>
|
||||
#import <CoreML/CoreML.h>
|
||||
#include <stdint.h>
|
||||
#include <os/log.h>
|
||||
|
||||
NS_ASSUME_NONNULL_BEGIN
|
||||
|
||||
|
||||
/// Model Prediction Input Type
|
||||
API_AVAILABLE(macos(12.0), ios(15.0), watchos(8.0), tvos(15.0)) __attribute__((visibility("hidden")))
|
||||
@interface whisper_decoder_implInput : NSObject<MLFeatureProvider>
|
||||
|
||||
/// token_data as 1 by 1 matrix of 32-bit integers
|
||||
@property (readwrite, nonatomic, strong) MLMultiArray * token_data;
|
||||
|
||||
/// audio_data as 1 × 384 × 1 × 1500 4-dimensional array of floats
|
||||
@property (readwrite, nonatomic, strong) MLMultiArray * audio_data;
|
||||
- (instancetype)init NS_UNAVAILABLE;
|
||||
- (instancetype)initWithToken_data:(MLMultiArray *)token_data audio_data:(MLMultiArray *)audio_data NS_DESIGNATED_INITIALIZER;
|
||||
|
||||
@end
|
||||
|
||||
|
||||
/// Model Prediction Output Type
|
||||
API_AVAILABLE(macos(12.0), ios(15.0), watchos(8.0), tvos(15.0)) __attribute__((visibility("hidden")))
|
||||
@interface whisper_decoder_implOutput : NSObject<MLFeatureProvider>
|
||||
|
||||
/// var_1346 as multidimensional array of floats
|
||||
@property (readwrite, nonatomic, strong) MLMultiArray * var_1346;
|
||||
- (instancetype)init NS_UNAVAILABLE;
|
||||
- (instancetype)initWithVar_1346:(MLMultiArray *)var_1346 NS_DESIGNATED_INITIALIZER;
|
||||
|
||||
@end
|
||||
|
||||
|
||||
/// Class for model loading and prediction
|
||||
API_AVAILABLE(macos(12.0), ios(15.0), watchos(8.0), tvos(15.0)) __attribute__((visibility("hidden")))
|
||||
@interface whisper_decoder_impl : NSObject
|
||||
@property (readonly, nonatomic, nullable) MLModel * model;
|
||||
|
||||
/**
|
||||
URL of the underlying .mlmodelc directory.
|
||||
*/
|
||||
+ (nullable NSURL *)URLOfModelInThisBundle;
|
||||
|
||||
/**
|
||||
Initialize whisper_decoder_impl instance from an existing MLModel object.
|
||||
|
||||
Usually the application does not use this initializer unless it makes a subclass of whisper_decoder_impl.
|
||||
Such application may want to use `-[MLModel initWithContentsOfURL:configuration:error:]` and `+URLOfModelInThisBundle` to create a MLModel object to pass-in.
|
||||
*/
|
||||
- (instancetype)initWithMLModel:(MLModel *)model NS_DESIGNATED_INITIALIZER;
|
||||
|
||||
/**
|
||||
Initialize whisper_decoder_impl instance with the model in this bundle.
|
||||
*/
|
||||
- (nullable instancetype)init;
|
||||
|
||||
/**
|
||||
Initialize whisper_decoder_impl instance with the model in this bundle.
|
||||
|
||||
@param configuration The model configuration object
|
||||
@param error If an error occurs, upon return contains an NSError object that describes the problem. If you are not interested in possible errors, pass in NULL.
|
||||
*/
|
||||
- (nullable instancetype)initWithConfiguration:(MLModelConfiguration *)configuration error:(NSError * _Nullable __autoreleasing * _Nullable)error;
|
||||
|
||||
/**
|
||||
Initialize whisper_decoder_impl instance from the model URL.
|
||||
|
||||
@param modelURL URL to the .mlmodelc directory for whisper_decoder_impl.
|
||||
@param error If an error occurs, upon return contains an NSError object that describes the problem. If you are not interested in possible errors, pass in NULL.
|
||||
*/
|
||||
- (nullable instancetype)initWithContentsOfURL:(NSURL *)modelURL error:(NSError * _Nullable __autoreleasing * _Nullable)error;
|
||||
|
||||
/**
|
||||
Initialize whisper_decoder_impl instance from the model URL.
|
||||
|
||||
@param modelURL URL to the .mlmodelc directory for whisper_decoder_impl.
|
||||
@param configuration The model configuration object
|
||||
@param error If an error occurs, upon return contains an NSError object that describes the problem. If you are not interested in possible errors, pass in NULL.
|
||||
*/
|
||||
- (nullable instancetype)initWithContentsOfURL:(NSURL *)modelURL configuration:(MLModelConfiguration *)configuration error:(NSError * _Nullable __autoreleasing * _Nullable)error;
|
||||
|
||||
/**
|
||||
Construct whisper_decoder_impl instance asynchronously with configuration.
|
||||
Model loading may take time when the model content is not immediately available (e.g. encrypted model). Use this factory method especially when the caller is on the main thread.
|
||||
|
||||
@param configuration The model configuration
|
||||
@param handler When the model load completes successfully or unsuccessfully, the completion handler is invoked with a valid whisper_decoder_impl instance or NSError object.
|
||||
*/
|
||||
+ (void)loadWithConfiguration:(MLModelConfiguration *)configuration completionHandler:(void (^)(whisper_decoder_impl * _Nullable model, NSError * _Nullable error))handler;
|
||||
|
||||
/**
|
||||
Construct whisper_decoder_impl instance asynchronously with URL of .mlmodelc directory and optional configuration.
|
||||
|
||||
Model loading may take time when the model content is not immediately available (e.g. encrypted model). Use this factory method especially when the caller is on the main thread.
|
||||
|
||||
@param modelURL The model URL.
|
||||
@param configuration The model configuration
|
||||
@param handler When the model load completes successfully or unsuccessfully, the completion handler is invoked with a valid whisper_decoder_impl instance or NSError object.
|
||||
*/
|
||||
+ (void)loadContentsOfURL:(NSURL *)modelURL configuration:(MLModelConfiguration *)configuration completionHandler:(void (^)(whisper_decoder_impl * _Nullable model, NSError * _Nullable error))handler;
|
||||
|
||||
/**
|
||||
Make a prediction using the standard interface
|
||||
@param input an instance of whisper_decoder_implInput to predict from
|
||||
@param error If an error occurs, upon return contains an NSError object that describes the problem. If you are not interested in possible errors, pass in NULL.
|
||||
@return the prediction as whisper_decoder_implOutput
|
||||
*/
|
||||
- (nullable whisper_decoder_implOutput *)predictionFromFeatures:(whisper_decoder_implInput *)input error:(NSError * _Nullable __autoreleasing * _Nullable)error;
|
||||
|
||||
/**
|
||||
Make a prediction using the standard interface
|
||||
@param input an instance of whisper_decoder_implInput to predict from
|
||||
@param options prediction options
|
||||
@param error If an error occurs, upon return contains an NSError object that describes the problem. If you are not interested in possible errors, pass in NULL.
|
||||
@return the prediction as whisper_decoder_implOutput
|
||||
*/
|
||||
- (nullable whisper_decoder_implOutput *)predictionFromFeatures:(whisper_decoder_implInput *)input options:(MLPredictionOptions *)options error:(NSError * _Nullable __autoreleasing * _Nullable)error;
|
||||
|
||||
/**
|
||||
Make a prediction using the convenience interface
|
||||
@param token_data as 1 by 1 matrix of 32-bit integers:
|
||||
@param audio_data as 1 × 384 × 1 × 1500 4-dimensional array of floats:
|
||||
@param error If an error occurs, upon return contains an NSError object that describes the problem. If you are not interested in possible errors, pass in NULL.
|
||||
@return the prediction as whisper_decoder_implOutput
|
||||
*/
|
||||
- (nullable whisper_decoder_implOutput *)predictionFromToken_data:(MLMultiArray *)token_data audio_data:(MLMultiArray *)audio_data error:(NSError * _Nullable __autoreleasing * _Nullable)error;
|
||||
|
||||
/**
|
||||
Batch prediction
|
||||
@param inputArray array of whisper_decoder_implInput instances to obtain predictions from
|
||||
@param options prediction options
|
||||
@param error If an error occurs, upon return contains an NSError object that describes the problem. If you are not interested in possible errors, pass in NULL.
|
||||
@return the predictions as NSArray<whisper_decoder_implOutput *>
|
||||
*/
|
||||
- (nullable NSArray<whisper_decoder_implOutput *> *)predictionsFromInputs:(NSArray<whisper_decoder_implInput*> *)inputArray options:(MLPredictionOptions *)options error:(NSError * _Nullable __autoreleasing * _Nullable)error;
|
||||
@end
|
||||
|
||||
NS_ASSUME_NONNULL_END
|
||||
201
src/coreml/whisper-decoder-impl.m
Normal file
201
src/coreml/whisper-decoder-impl.m
Normal file
@ -0,0 +1,201 @@
|
||||
//
|
||||
// whisper-decoder-impl.m
|
||||
//
|
||||
// This file was automatically generated and should not be edited.
|
||||
//
|
||||
|
||||
#if !__has_feature(objc_arc)
|
||||
#error This file must be compiled with automatic reference counting enabled (-fobjc-arc)
|
||||
#endif
|
||||
|
||||
#import "whisper-decoder-impl.h"
|
||||
|
||||
@implementation whisper_decoder_implInput
|
||||
|
||||
- (instancetype)initWithToken_data:(MLMultiArray *)token_data audio_data:(MLMultiArray *)audio_data {
|
||||
self = [super init];
|
||||
if (self) {
|
||||
_token_data = token_data;
|
||||
_audio_data = audio_data;
|
||||
}
|
||||
return self;
|
||||
}
|
||||
|
||||
- (NSSet<NSString *> *)featureNames {
|
||||
return [NSSet setWithArray:@[@"token_data", @"audio_data"]];
|
||||
}
|
||||
|
||||
- (nullable MLFeatureValue *)featureValueForName:(NSString *)featureName {
|
||||
if ([featureName isEqualToString:@"token_data"]) {
|
||||
return [MLFeatureValue featureValueWithMultiArray:self.token_data];
|
||||
}
|
||||
if ([featureName isEqualToString:@"audio_data"]) {
|
||||
return [MLFeatureValue featureValueWithMultiArray:self.audio_data];
|
||||
}
|
||||
return nil;
|
||||
}
|
||||
|
||||
@end
|
||||
|
||||
@implementation whisper_decoder_implOutput
|
||||
|
||||
- (instancetype)initWithVar_1346:(MLMultiArray *)var_1346 {
|
||||
self = [super init];
|
||||
if (self) {
|
||||
_var_1346 = var_1346;
|
||||
}
|
||||
return self;
|
||||
}
|
||||
|
||||
- (NSSet<NSString *> *)featureNames {
|
||||
return [NSSet setWithArray:@[@"var_1346"]];
|
||||
}
|
||||
|
||||
- (nullable MLFeatureValue *)featureValueForName:(NSString *)featureName {
|
||||
if ([featureName isEqualToString:@"var_1346"]) {
|
||||
return [MLFeatureValue featureValueWithMultiArray:self.var_1346];
|
||||
}
|
||||
return nil;
|
||||
}
|
||||
|
||||
@end
|
||||
|
||||
@implementation whisper_decoder_impl
|
||||
|
||||
|
||||
/**
|
||||
URL of the underlying .mlmodelc directory.
|
||||
*/
|
||||
+ (nullable NSURL *)URLOfModelInThisBundle {
|
||||
NSString *assetPath = [[NSBundle bundleForClass:[self class]] pathForResource:@"whisper_decoder_impl" ofType:@"mlmodelc"];
|
||||
if (nil == assetPath) { os_log_error(OS_LOG_DEFAULT, "Could not load whisper-decoder-impl.mlmodelc in the bundle resource"); return nil; }
|
||||
return [NSURL fileURLWithPath:assetPath];
|
||||
}
|
||||
|
||||
|
||||
/**
|
||||
Initialize whisper_decoder_impl instance from an existing MLModel object.
|
||||
|
||||
Usually the application does not use this initializer unless it makes a subclass of whisper_decoder_impl.
|
||||
Such application may want to use `-[MLModel initWithContentsOfURL:configuration:error:]` and `+URLOfModelInThisBundle` to create a MLModel object to pass-in.
|
||||
*/
|
||||
- (instancetype)initWithMLModel:(MLModel *)model {
|
||||
self = [super init];
|
||||
if (!self) { return nil; }
|
||||
_model = model;
|
||||
if (_model == nil) { return nil; }
|
||||
return self;
|
||||
}
|
||||
|
||||
|
||||
/**
|
||||
Initialize whisper_decoder_impl instance with the model in this bundle.
|
||||
*/
|
||||
- (nullable instancetype)init {
|
||||
return [self initWithContentsOfURL:(NSURL * _Nonnull)self.class.URLOfModelInThisBundle error:nil];
|
||||
}
|
||||
|
||||
|
||||
/**
|
||||
Initialize whisper_decoder_impl instance with the model in this bundle.
|
||||
|
||||
@param configuration The model configuration object
|
||||
@param error If an error occurs, upon return contains an NSError object that describes the problem. If you are not interested in possible errors, pass in NULL.
|
||||
*/
|
||||
- (nullable instancetype)initWithConfiguration:(MLModelConfiguration *)configuration error:(NSError * _Nullable __autoreleasing * _Nullable)error {
|
||||
return [self initWithContentsOfURL:(NSURL * _Nonnull)self.class.URLOfModelInThisBundle configuration:configuration error:error];
|
||||
}
|
||||
|
||||
|
||||
/**
|
||||
Initialize whisper_decoder_impl instance from the model URL.
|
||||
|
||||
@param modelURL URL to the .mlmodelc directory for whisper_decoder_impl.
|
||||
@param error If an error occurs, upon return contains an NSError object that describes the problem. If you are not interested in possible errors, pass in NULL.
|
||||
*/
|
||||
- (nullable instancetype)initWithContentsOfURL:(NSURL *)modelURL error:(NSError * _Nullable __autoreleasing * _Nullable)error {
|
||||
MLModel *model = [MLModel modelWithContentsOfURL:modelURL error:error];
|
||||
if (model == nil) { return nil; }
|
||||
return [self initWithMLModel:model];
|
||||
}
|
||||
|
||||
|
||||
/**
|
||||
Initialize whisper_decoder_impl instance from the model URL.
|
||||
|
||||
@param modelURL URL to the .mlmodelc directory for whisper_decoder_impl.
|
||||
@param configuration The model configuration object
|
||||
@param error If an error occurs, upon return contains an NSError object that describes the problem. If you are not interested in possible errors, pass in NULL.
|
||||
*/
|
||||
- (nullable instancetype)initWithContentsOfURL:(NSURL *)modelURL configuration:(MLModelConfiguration *)configuration error:(NSError * _Nullable __autoreleasing * _Nullable)error {
|
||||
MLModel *model = [MLModel modelWithContentsOfURL:modelURL configuration:configuration error:error];
|
||||
if (model == nil) { return nil; }
|
||||
return [self initWithMLModel:model];
|
||||
}
|
||||
|
||||
|
||||
/**
|
||||
Construct whisper_decoder_impl instance asynchronously with configuration.
|
||||
Model loading may take time when the model content is not immediately available (e.g. encrypted model). Use this factory method especially when the caller is on the main thread.
|
||||
|
||||
@param configuration The model configuration
|
||||
@param handler When the model load completes successfully or unsuccessfully, the completion handler is invoked with a valid whisper_decoder_impl instance or NSError object.
|
||||
*/
|
||||
+ (void)loadWithConfiguration:(MLModelConfiguration *)configuration completionHandler:(void (^)(whisper_decoder_impl * _Nullable model, NSError * _Nullable error))handler {
|
||||
[self loadContentsOfURL:(NSURL * _Nonnull)[self URLOfModelInThisBundle]
|
||||
configuration:configuration
|
||||
completionHandler:handler];
|
||||
}
|
||||
|
||||
|
||||
/**
|
||||
Construct whisper_decoder_impl instance asynchronously with URL of .mlmodelc directory and optional configuration.
|
||||
|
||||
Model loading may take time when the model content is not immediately available (e.g. encrypted model). Use this factory method especially when the caller is on the main thread.
|
||||
|
||||
@param modelURL The model URL.
|
||||
@param configuration The model configuration
|
||||
@param handler When the model load completes successfully or unsuccessfully, the completion handler is invoked with a valid whisper_decoder_impl instance or NSError object.
|
||||
*/
|
||||
+ (void)loadContentsOfURL:(NSURL *)modelURL configuration:(MLModelConfiguration *)configuration completionHandler:(void (^)(whisper_decoder_impl * _Nullable model, NSError * _Nullable error))handler {
|
||||
[MLModel loadContentsOfURL:modelURL
|
||||
configuration:configuration
|
||||
completionHandler:^(MLModel *model, NSError *error) {
|
||||
if (model != nil) {
|
||||
whisper_decoder_impl *typedModel = [[whisper_decoder_impl alloc] initWithMLModel:model];
|
||||
handler(typedModel, nil);
|
||||
} else {
|
||||
handler(nil, error);
|
||||
}
|
||||
}];
|
||||
}
|
||||
|
||||
- (nullable whisper_decoder_implOutput *)predictionFromFeatures:(whisper_decoder_implInput *)input error:(NSError * _Nullable __autoreleasing * _Nullable)error {
|
||||
return [self predictionFromFeatures:input options:[[MLPredictionOptions alloc] init] error:error];
|
||||
}
|
||||
|
||||
- (nullable whisper_decoder_implOutput *)predictionFromFeatures:(whisper_decoder_implInput *)input options:(MLPredictionOptions *)options error:(NSError * _Nullable __autoreleasing * _Nullable)error {
|
||||
id<MLFeatureProvider> outFeatures = [self.model predictionFromFeatures:input options:options error:error];
|
||||
if (!outFeatures) { return nil; }
|
||||
return [[whisper_decoder_implOutput alloc] initWithVar_1346:(MLMultiArray *)[outFeatures featureValueForName:@"var_1346"].multiArrayValue];
|
||||
}
|
||||
|
||||
- (nullable whisper_decoder_implOutput *)predictionFromToken_data:(MLMultiArray *)token_data audio_data:(MLMultiArray *)audio_data error:(NSError * _Nullable __autoreleasing * _Nullable)error {
|
||||
whisper_decoder_implInput *input_ = [[whisper_decoder_implInput alloc] initWithToken_data:token_data audio_data:audio_data];
|
||||
return [self predictionFromFeatures:input_ error:error];
|
||||
}
|
||||
|
||||
- (nullable NSArray<whisper_decoder_implOutput *> *)predictionsFromInputs:(NSArray<whisper_decoder_implInput*> *)inputArray options:(MLPredictionOptions *)options error:(NSError * _Nullable __autoreleasing * _Nullable)error {
|
||||
id<MLBatchProvider> inBatch = [[MLArrayBatchProvider alloc] initWithFeatureProviderArray:inputArray];
|
||||
id<MLBatchProvider> outBatch = [self.model predictionsFromBatch:inBatch options:options error:error];
|
||||
if (!outBatch) { return nil; }
|
||||
NSMutableArray<whisper_decoder_implOutput*> *results = [NSMutableArray arrayWithCapacity:(NSUInteger)outBatch.count];
|
||||
for (NSInteger i = 0; i < outBatch.count; i++) {
|
||||
id<MLFeatureProvider> resultProvider = [outBatch featuresAtIndex:i];
|
||||
whisper_decoder_implOutput * result = [[whisper_decoder_implOutput alloc] initWithVar_1346:(MLMultiArray *)[resultProvider featureValueForName:@"var_1346"].multiArrayValue];
|
||||
[results addObject:result];
|
||||
}
|
||||
return results;
|
||||
}
|
||||
|
||||
@end
|
||||
142
src/coreml/whisper-encoder-impl.h
Normal file
142
src/coreml/whisper-encoder-impl.h
Normal file
@ -0,0 +1,142 @@
|
||||
//
|
||||
// whisper-encoder-impl.h
|
||||
//
|
||||
// This file was automatically generated and should not be edited.
|
||||
//
|
||||
|
||||
#import <Foundation/Foundation.h>
|
||||
#import <CoreML/CoreML.h>
|
||||
#include <stdint.h>
|
||||
#include <os/log.h>
|
||||
|
||||
NS_ASSUME_NONNULL_BEGIN
|
||||
|
||||
|
||||
/// Model Prediction Input Type
|
||||
API_AVAILABLE(macos(12.0), ios(15.0), watchos(8.0), tvos(15.0)) __attribute__((visibility("hidden")))
|
||||
@interface whisper_encoder_implInput : NSObject<MLFeatureProvider>
|
||||
|
||||
/// logmel_data as 1 × 80 × 3000 3-dimensional array of floats
|
||||
@property (readwrite, nonatomic, strong) MLMultiArray * logmel_data;
|
||||
- (instancetype)init NS_UNAVAILABLE;
|
||||
- (instancetype)initWithLogmel_data:(MLMultiArray *)logmel_data NS_DESIGNATED_INITIALIZER;
|
||||
|
||||
@end
|
||||
|
||||
|
||||
/// Model Prediction Output Type
|
||||
API_AVAILABLE(macos(12.0), ios(15.0), watchos(8.0), tvos(15.0)) __attribute__((visibility("hidden")))
|
||||
@interface whisper_encoder_implOutput : NSObject<MLFeatureProvider>
|
||||
|
||||
/// output as multidimensional array of floats
|
||||
@property (readwrite, nonatomic, strong) MLMultiArray * output;
|
||||
- (instancetype)init NS_UNAVAILABLE;
|
||||
- (instancetype)initWithOutput:(MLMultiArray *)output NS_DESIGNATED_INITIALIZER;
|
||||
|
||||
@end
|
||||
|
||||
|
||||
/// Class for model loading and prediction
|
||||
API_AVAILABLE(macos(12.0), ios(15.0), watchos(8.0), tvos(15.0)) __attribute__((visibility("hidden")))
|
||||
@interface whisper_encoder_impl : NSObject
|
||||
@property (readonly, nonatomic, nullable) MLModel * model;
|
||||
|
||||
/**
|
||||
URL of the underlying .mlmodelc directory.
|
||||
*/
|
||||
+ (nullable NSURL *)URLOfModelInThisBundle;
|
||||
|
||||
/**
|
||||
Initialize whisper_encoder_impl instance from an existing MLModel object.
|
||||
|
||||
Usually the application does not use this initializer unless it makes a subclass of whisper_encoder_impl.
|
||||
Such application may want to use `-[MLModel initWithContentsOfURL:configuration:error:]` and `+URLOfModelInThisBundle` to create a MLModel object to pass-in.
|
||||
*/
|
||||
- (instancetype)initWithMLModel:(MLModel *)model NS_DESIGNATED_INITIALIZER;
|
||||
|
||||
/**
|
||||
Initialize whisper_encoder_impl instance with the model in this bundle.
|
||||
*/
|
||||
- (nullable instancetype)init;
|
||||
|
||||
/**
|
||||
Initialize whisper_encoder_impl instance with the model in this bundle.
|
||||
|
||||
@param configuration The model configuration object
|
||||
@param error If an error occurs, upon return contains an NSError object that describes the problem. If you are not interested in possible errors, pass in NULL.
|
||||
*/
|
||||
- (nullable instancetype)initWithConfiguration:(MLModelConfiguration *)configuration error:(NSError * _Nullable __autoreleasing * _Nullable)error;
|
||||
|
||||
/**
|
||||
Initialize whisper_encoder_impl instance from the model URL.
|
||||
|
||||
@param modelURL URL to the .mlmodelc directory for whisper_encoder_impl.
|
||||
@param error If an error occurs, upon return contains an NSError object that describes the problem. If you are not interested in possible errors, pass in NULL.
|
||||
*/
|
||||
- (nullable instancetype)initWithContentsOfURL:(NSURL *)modelURL error:(NSError * _Nullable __autoreleasing * _Nullable)error;
|
||||
|
||||
/**
|
||||
Initialize whisper_encoder_impl instance from the model URL.
|
||||
|
||||
@param modelURL URL to the .mlmodelc directory for whisper_encoder_impl.
|
||||
@param configuration The model configuration object
|
||||
@param error If an error occurs, upon return contains an NSError object that describes the problem. If you are not interested in possible errors, pass in NULL.
|
||||
*/
|
||||
- (nullable instancetype)initWithContentsOfURL:(NSURL *)modelURL configuration:(MLModelConfiguration *)configuration error:(NSError * _Nullable __autoreleasing * _Nullable)error;
|
||||
|
||||
/**
|
||||
Construct whisper_encoder_impl instance asynchronously with configuration.
|
||||
Model loading may take time when the model content is not immediately available (e.g. encrypted model). Use this factory method especially when the caller is on the main thread.
|
||||
|
||||
@param configuration The model configuration
|
||||
@param handler When the model load completes successfully or unsuccessfully, the completion handler is invoked with a valid whisper_encoder_impl instance or NSError object.
|
||||
*/
|
||||
+ (void)loadWithConfiguration:(MLModelConfiguration *)configuration completionHandler:(void (^)(whisper_encoder_impl * _Nullable model, NSError * _Nullable error))handler;
|
||||
|
||||
/**
|
||||
Construct whisper_encoder_impl instance asynchronously with URL of .mlmodelc directory and optional configuration.
|
||||
|
||||
Model loading may take time when the model content is not immediately available (e.g. encrypted model). Use this factory method especially when the caller is on the main thread.
|
||||
|
||||
@param modelURL The model URL.
|
||||
@param configuration The model configuration
|
||||
@param handler When the model load completes successfully or unsuccessfully, the completion handler is invoked with a valid whisper_encoder_impl instance or NSError object.
|
||||
*/
|
||||
+ (void)loadContentsOfURL:(NSURL *)modelURL configuration:(MLModelConfiguration *)configuration completionHandler:(void (^)(whisper_encoder_impl * _Nullable model, NSError * _Nullable error))handler;
|
||||
|
||||
/**
|
||||
Make a prediction using the standard interface
|
||||
@param input an instance of whisper_encoder_implInput to predict from
|
||||
@param error If an error occurs, upon return contains an NSError object that describes the problem. If you are not interested in possible errors, pass in NULL.
|
||||
@return the prediction as whisper_encoder_implOutput
|
||||
*/
|
||||
- (nullable whisper_encoder_implOutput *)predictionFromFeatures:(whisper_encoder_implInput *)input error:(NSError * _Nullable __autoreleasing * _Nullable)error;
|
||||
|
||||
/**
|
||||
Make a prediction using the standard interface
|
||||
@param input an instance of whisper_encoder_implInput to predict from
|
||||
@param options prediction options
|
||||
@param error If an error occurs, upon return contains an NSError object that describes the problem. If you are not interested in possible errors, pass in NULL.
|
||||
@return the prediction as whisper_encoder_implOutput
|
||||
*/
|
||||
- (nullable whisper_encoder_implOutput *)predictionFromFeatures:(whisper_encoder_implInput *)input options:(MLPredictionOptions *)options error:(NSError * _Nullable __autoreleasing * _Nullable)error;
|
||||
|
||||
/**
|
||||
Make a prediction using the convenience interface
|
||||
@param logmel_data as 1 × n_mel × 3000 3-dimensional array of floats:
|
||||
@param error If an error occurs, upon return contains an NSError object that describes the problem. If you are not interested in possible errors, pass in NULL.
|
||||
@return the prediction as whisper_encoder_implOutput
|
||||
*/
|
||||
- (nullable whisper_encoder_implOutput *)predictionFromLogmel_data:(MLMultiArray *)logmel_data error:(NSError * _Nullable __autoreleasing * _Nullable)error;
|
||||
|
||||
/**
|
||||
Batch prediction
|
||||
@param inputArray array of whisper_encoder_implInput instances to obtain predictions from
|
||||
@param options prediction options
|
||||
@param error If an error occurs, upon return contains an NSError object that describes the problem. If you are not interested in possible errors, pass in NULL.
|
||||
@return the predictions as NSArray<whisper_encoder_implOutput *>
|
||||
*/
|
||||
- (nullable NSArray<whisper_encoder_implOutput *> *)predictionsFromInputs:(NSArray<whisper_encoder_implInput*> *)inputArray options:(MLPredictionOptions *)options error:(NSError * _Nullable __autoreleasing * _Nullable)error;
|
||||
@end
|
||||
|
||||
NS_ASSUME_NONNULL_END
|
||||
197
src/coreml/whisper-encoder-impl.m
Normal file
197
src/coreml/whisper-encoder-impl.m
Normal file
@ -0,0 +1,197 @@
|
||||
//
|
||||
// whisper-encoder-impl.m
|
||||
//
|
||||
// This file was automatically generated and should not be edited.
|
||||
//
|
||||
|
||||
#if !__has_feature(objc_arc)
|
||||
#error This file must be compiled with automatic reference counting enabled (-fobjc-arc)
|
||||
#endif
|
||||
|
||||
#import "whisper-encoder-impl.h"
|
||||
|
||||
@implementation whisper_encoder_implInput
|
||||
|
||||
- (instancetype)initWithLogmel_data:(MLMultiArray *)logmel_data {
|
||||
self = [super init];
|
||||
if (self) {
|
||||
_logmel_data = logmel_data;
|
||||
}
|
||||
return self;
|
||||
}
|
||||
|
||||
- (NSSet<NSString *> *)featureNames {
|
||||
return [NSSet setWithArray:@[@"logmel_data"]];
|
||||
}
|
||||
|
||||
- (nullable MLFeatureValue *)featureValueForName:(NSString *)featureName {
|
||||
if ([featureName isEqualToString:@"logmel_data"]) {
|
||||
return [MLFeatureValue featureValueWithMultiArray:self.logmel_data];
|
||||
}
|
||||
return nil;
|
||||
}
|
||||
|
||||
@end
|
||||
|
||||
@implementation whisper_encoder_implOutput
|
||||
|
||||
- (instancetype)initWithOutput:(MLMultiArray *)output {
|
||||
self = [super init];
|
||||
if (self) {
|
||||
_output = output;
|
||||
}
|
||||
return self;
|
||||
}
|
||||
|
||||
- (NSSet<NSString *> *)featureNames {
|
||||
return [NSSet setWithArray:@[@"output"]];
|
||||
}
|
||||
|
||||
- (nullable MLFeatureValue *)featureValueForName:(NSString *)featureName {
|
||||
if ([featureName isEqualToString:@"output"]) {
|
||||
return [MLFeatureValue featureValueWithMultiArray:self.output];
|
||||
}
|
||||
return nil;
|
||||
}
|
||||
|
||||
@end
|
||||
|
||||
@implementation whisper_encoder_impl
|
||||
|
||||
|
||||
/**
|
||||
URL of the underlying .mlmodelc directory.
|
||||
*/
|
||||
+ (nullable NSURL *)URLOfModelInThisBundle {
|
||||
NSString *assetPath = [[NSBundle bundleForClass:[self class]] pathForResource:@"whisper_encoder_impl" ofType:@"mlmodelc"];
|
||||
if (nil == assetPath) { os_log_error(OS_LOG_DEFAULT, "Could not load whisper-encoder-impl.mlmodelc in the bundle resource"); return nil; }
|
||||
return [NSURL fileURLWithPath:assetPath];
|
||||
}
|
||||
|
||||
|
||||
/**
|
||||
Initialize whisper_encoder_impl instance from an existing MLModel object.
|
||||
|
||||
Usually the application does not use this initializer unless it makes a subclass of whisper_encoder_impl.
|
||||
Such application may want to use `-[MLModel initWithContentsOfURL:configuration:error:]` and `+URLOfModelInThisBundle` to create a MLModel object to pass-in.
|
||||
*/
|
||||
- (instancetype)initWithMLModel:(MLModel *)model {
|
||||
self = [super init];
|
||||
if (!self) { return nil; }
|
||||
_model = model;
|
||||
if (_model == nil) { return nil; }
|
||||
return self;
|
||||
}
|
||||
|
||||
|
||||
/**
|
||||
Initialize whisper_encoder_impl instance with the model in this bundle.
|
||||
*/
|
||||
- (nullable instancetype)init {
|
||||
return [self initWithContentsOfURL:(NSURL * _Nonnull)self.class.URLOfModelInThisBundle error:nil];
|
||||
}
|
||||
|
||||
|
||||
/**
|
||||
Initialize whisper_encoder_impl instance with the model in this bundle.
|
||||
|
||||
@param configuration The model configuration object
|
||||
@param error If an error occurs, upon return contains an NSError object that describes the problem. If you are not interested in possible errors, pass in NULL.
|
||||
*/
|
||||
- (nullable instancetype)initWithConfiguration:(MLModelConfiguration *)configuration error:(NSError * _Nullable __autoreleasing * _Nullable)error {
|
||||
return [self initWithContentsOfURL:(NSURL * _Nonnull)self.class.URLOfModelInThisBundle configuration:configuration error:error];
|
||||
}
|
||||
|
||||
|
||||
/**
|
||||
Initialize whisper_encoder_impl instance from the model URL.
|
||||
|
||||
@param modelURL URL to the .mlmodelc directory for whisper_encoder_impl.
|
||||
@param error If an error occurs, upon return contains an NSError object that describes the problem. If you are not interested in possible errors, pass in NULL.
|
||||
*/
|
||||
- (nullable instancetype)initWithContentsOfURL:(NSURL *)modelURL error:(NSError * _Nullable __autoreleasing * _Nullable)error {
|
||||
MLModel *model = [MLModel modelWithContentsOfURL:modelURL error:error];
|
||||
if (model == nil) { return nil; }
|
||||
return [self initWithMLModel:model];
|
||||
}
|
||||
|
||||
|
||||
/**
|
||||
Initialize whisper_encoder_impl instance from the model URL.
|
||||
|
||||
@param modelURL URL to the .mlmodelc directory for whisper_encoder_impl.
|
||||
@param configuration The model configuration object
|
||||
@param error If an error occurs, upon return contains an NSError object that describes the problem. If you are not interested in possible errors, pass in NULL.
|
||||
*/
|
||||
- (nullable instancetype)initWithContentsOfURL:(NSURL *)modelURL configuration:(MLModelConfiguration *)configuration error:(NSError * _Nullable __autoreleasing * _Nullable)error {
|
||||
MLModel *model = [MLModel modelWithContentsOfURL:modelURL configuration:configuration error:error];
|
||||
if (model == nil) { return nil; }
|
||||
return [self initWithMLModel:model];
|
||||
}
|
||||
|
||||
|
||||
/**
|
||||
Construct whisper_encoder_impl instance asynchronously with configuration.
|
||||
Model loading may take time when the model content is not immediately available (e.g. encrypted model). Use this factory method especially when the caller is on the main thread.
|
||||
|
||||
@param configuration The model configuration
|
||||
@param handler When the model load completes successfully or unsuccessfully, the completion handler is invoked with a valid whisper_encoder_impl instance or NSError object.
|
||||
*/
|
||||
+ (void)loadWithConfiguration:(MLModelConfiguration *)configuration completionHandler:(void (^)(whisper_encoder_impl * _Nullable model, NSError * _Nullable error))handler {
|
||||
[self loadContentsOfURL:(NSURL * _Nonnull)[self URLOfModelInThisBundle]
|
||||
configuration:configuration
|
||||
completionHandler:handler];
|
||||
}
|
||||
|
||||
|
||||
/**
|
||||
Construct whisper_encoder_impl instance asynchronously with URL of .mlmodelc directory and optional configuration.
|
||||
|
||||
Model loading may take time when the model content is not immediately available (e.g. encrypted model). Use this factory method especially when the caller is on the main thread.
|
||||
|
||||
@param modelURL The model URL.
|
||||
@param configuration The model configuration
|
||||
@param handler When the model load completes successfully or unsuccessfully, the completion handler is invoked with a valid whisper_encoder_impl instance or NSError object.
|
||||
*/
|
||||
+ (void)loadContentsOfURL:(NSURL *)modelURL configuration:(MLModelConfiguration *)configuration completionHandler:(void (^)(whisper_encoder_impl * _Nullable model, NSError * _Nullable error))handler {
|
||||
[MLModel loadContentsOfURL:modelURL
|
||||
configuration:configuration
|
||||
completionHandler:^(MLModel *model, NSError *error) {
|
||||
if (model != nil) {
|
||||
whisper_encoder_impl *typedModel = [[whisper_encoder_impl alloc] initWithMLModel:model];
|
||||
handler(typedModel, nil);
|
||||
} else {
|
||||
handler(nil, error);
|
||||
}
|
||||
}];
|
||||
}
|
||||
|
||||
- (nullable whisper_encoder_implOutput *)predictionFromFeatures:(whisper_encoder_implInput *)input error:(NSError * _Nullable __autoreleasing * _Nullable)error {
|
||||
return [self predictionFromFeatures:input options:[[MLPredictionOptions alloc] init] error:error];
|
||||
}
|
||||
|
||||
- (nullable whisper_encoder_implOutput *)predictionFromFeatures:(whisper_encoder_implInput *)input options:(MLPredictionOptions *)options error:(NSError * _Nullable __autoreleasing * _Nullable)error {
|
||||
id<MLFeatureProvider> outFeatures = [self.model predictionFromFeatures:input options:options error:error];
|
||||
if (!outFeatures) { return nil; }
|
||||
return [[whisper_encoder_implOutput alloc] initWithOutput:(MLMultiArray *)[outFeatures featureValueForName:@"output"].multiArrayValue];
|
||||
}
|
||||
|
||||
- (nullable whisper_encoder_implOutput *)predictionFromLogmel_data:(MLMultiArray *)logmel_data error:(NSError * _Nullable __autoreleasing * _Nullable)error {
|
||||
whisper_encoder_implInput *input_ = [[whisper_encoder_implInput alloc] initWithLogmel_data:logmel_data];
|
||||
return [self predictionFromFeatures:input_ error:error];
|
||||
}
|
||||
|
||||
- (nullable NSArray<whisper_encoder_implOutput *> *)predictionsFromInputs:(NSArray<whisper_encoder_implInput*> *)inputArray options:(MLPredictionOptions *)options error:(NSError * _Nullable __autoreleasing * _Nullable)error {
|
||||
id<MLBatchProvider> inBatch = [[MLArrayBatchProvider alloc] initWithFeatureProviderArray:inputArray];
|
||||
id<MLBatchProvider> outBatch = [self.model predictionsFromBatch:inBatch options:options error:error];
|
||||
if (!outBatch) { return nil; }
|
||||
NSMutableArray<whisper_encoder_implOutput*> *results = [NSMutableArray arrayWithCapacity:(NSUInteger)outBatch.count];
|
||||
for (NSInteger i = 0; i < outBatch.count; i++) {
|
||||
id<MLFeatureProvider> resultProvider = [outBatch featuresAtIndex:i];
|
||||
whisper_encoder_implOutput * result = [[whisper_encoder_implOutput alloc] initWithOutput:(MLMultiArray *)[resultProvider featureValueForName:@"output"].multiArrayValue];
|
||||
[results addObject:result];
|
||||
}
|
||||
return results;
|
||||
}
|
||||
|
||||
@end
|
||||
26
src/coreml/whisper-encoder.h
Normal file
26
src/coreml/whisper-encoder.h
Normal file
@ -0,0 +1,26 @@
|
||||
// Wrapper of the Core ML Whisper Encoder model
|
||||
//
|
||||
// Code is derived from the work of Github user @wangchou
|
||||
// ref: https://github.com/wangchou/callCoreMLFromCpp
|
||||
|
||||
#include <stdint.h>
|
||||
|
||||
#if __cplusplus
|
||||
extern "C" {
|
||||
#endif
|
||||
|
||||
struct whisper_coreml_context;
|
||||
|
||||
struct whisper_coreml_context * whisper_coreml_init(const char * path_model);
|
||||
void whisper_coreml_free(struct whisper_coreml_context * ctx);
|
||||
|
||||
void whisper_coreml_encode(
|
||||
const whisper_coreml_context * ctx,
|
||||
int64_t n_ctx,
|
||||
int64_t n_mel,
|
||||
float * mel,
|
||||
float * out);
|
||||
|
||||
#if __cplusplus
|
||||
}
|
||||
#endif
|
||||
73
src/coreml/whisper-encoder.mm
Normal file
73
src/coreml/whisper-encoder.mm
Normal file
@ -0,0 +1,73 @@
|
||||
#if !__has_feature(objc_arc)
|
||||
#error This file must be compiled with automatic reference counting enabled (-fobjc-arc)
|
||||
#endif
|
||||
|
||||
#import "whisper-encoder.h"
|
||||
#import "whisper-encoder-impl.h"
|
||||
|
||||
#import <CoreML/CoreML.h>
|
||||
|
||||
#include <stdlib.h>
|
||||
|
||||
#if __cplusplus
|
||||
extern "C" {
|
||||
#endif
|
||||
|
||||
struct whisper_coreml_context {
|
||||
const void * data;
|
||||
};
|
||||
|
||||
struct whisper_coreml_context * whisper_coreml_init(const char * path_model) {
|
||||
NSString * path_model_str = [[NSString alloc] initWithUTF8String:path_model];
|
||||
|
||||
NSURL * url_model = [NSURL fileURLWithPath: path_model_str];
|
||||
|
||||
// select which device to run the Core ML model on
|
||||
MLModelConfiguration *config = [[MLModelConfiguration alloc] init];
|
||||
// config.computeUnits = MLComputeUnitsCPUAndGPU;
|
||||
//config.computeUnits = MLComputeUnitsCPUAndNeuralEngine;
|
||||
config.computeUnits = MLComputeUnitsAll;
|
||||
|
||||
const void * data = CFBridgingRetain([[whisper_encoder_impl alloc] initWithContentsOfURL:url_model configuration:config error:nil]);
|
||||
|
||||
if (data == NULL) {
|
||||
return NULL;
|
||||
}
|
||||
|
||||
whisper_coreml_context * ctx = new whisper_coreml_context;
|
||||
|
||||
ctx->data = data;
|
||||
|
||||
return ctx;
|
||||
}
|
||||
|
||||
void whisper_coreml_free(struct whisper_coreml_context * ctx) {
|
||||
CFRelease(ctx->data);
|
||||
delete ctx;
|
||||
}
|
||||
|
||||
void whisper_coreml_encode(
|
||||
const whisper_coreml_context * ctx,
|
||||
int64_t n_ctx,
|
||||
int64_t n_mel,
|
||||
float * mel,
|
||||
float * out) {
|
||||
MLMultiArray * inMultiArray = [
|
||||
[MLMultiArray alloc] initWithDataPointer: mel
|
||||
shape: @[@1, @(n_mel), @(n_ctx)]
|
||||
dataType: MLMultiArrayDataTypeFloat32
|
||||
strides: @[@(n_ctx*n_mel), @(n_ctx), @1]
|
||||
deallocator: nil
|
||||
error: nil
|
||||
];
|
||||
|
||||
@autoreleasepool {
|
||||
whisper_encoder_implOutput * outCoreML = [(__bridge id) ctx->data predictionFromLogmel_data:inMultiArray error:nil];
|
||||
|
||||
memcpy(out, outCoreML.output.dataPointer, outCoreML.output.count * sizeof(float));
|
||||
}
|
||||
}
|
||||
|
||||
#if __cplusplus
|
||||
}
|
||||
#endif
|
||||
Reference in New Issue
Block a user