ggml : try to fix the abort mechanism

.github/workflows/build.yml

@@ -88,7 +88,7 @@ jobs:
             -w /workspace ${{ env.ubuntu_image }} /bin/sh -c '
             apt update
             apt install -y build-essential cmake libsdl2-dev
-            cmake . -DWHISPER_SDL2=ON -DCMAKE_BUILD_TYPE=${{ matrix.build }}
+            cmake . -DWHISPER_SUPPORT_SDL2=ON -DCMAKE_BUILD_TYPE=${{ matrix.build }}
             make
             ctest -L gh --output-on-failure'
 
@@ -115,7 +115,7 @@ jobs:
             -w /workspace ${{ env.ubuntu_image }} /bin/sh -c '
             apt update
             apt install -y build-essential cmake libsdl2-dev
-            cmake . -DWHISPER_SDL2=ON -DCMAKE_BUILD_TYPE=${{ matrix.build }} -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_C_COMPILER=clang
+            cmake . -DWHISPER_SUPPORT_SDL2=ON -DCMAKE_BUILD_TYPE=${{ matrix.build }} -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_C_COMPILER=clang
             make
             ctest -L gh --output-on-failure'
 
@@ -182,7 +182,7 @@ jobs:
         run: >
           cmake -S . -B ./build -A ${{ matrix.arch }}
           -DCMAKE_BUILD_TYPE=${{ matrix.build }}
-          -DWHISPER_SDL2=${{ matrix.sdl2 }}
+          -DWHISPER_SUPPORT_SDL2=${{ matrix.sdl2 }}
 
       - name: Build
         run: |
@@ -217,10 +217,10 @@ jobs:
         sdl2: [ON]
         include:
           - arch: Win32
-            obzip: https://github.com/OpenMathLib/OpenBLAS/releases/download/v0.3.24/OpenBLAS-0.3.24-x86.zip
+            obzip: https://github.com/xianyi/OpenBLAS/releases/download/v0.3.21/OpenBLAS-0.3.21-x86.zip
             s2arc: x86
           - arch: x64
-            obzip: https://github.com/OpenMathLib/OpenBLAS/releases/download/v0.3.24/OpenBLAS-0.3.24-x64.zip
+            obzip: https://github.com/xianyi/OpenBLAS/releases/download/v0.3.21/OpenBLAS-0.3.21-x64.zip
             s2arc: x64
           - sdl2: ON
             s2ver: 2.26.0
@@ -239,7 +239,7 @@ jobs:
           7z x blas.zip -oblas -y
           copy blas/include/cblas.h .
           copy blas/include/openblas_config.h .
-          echo "OPENBLAS_PATH=$env:GITHUB_WORKSPACE/blas" >> $env:GITHUB_ENV
+          echo "blasdir=$env:GITHUB_WORKSPACE/blas" >> $env:GITHUB_ENV
 
       - name: Fetch SDL2 and set SDL2_DIR
         if: matrix.sdl2 == 'ON'
@@ -252,9 +252,9 @@ jobs:
         run: >
           cmake -S . -B ./build -A ${{ matrix.arch }}
           -DCMAKE_BUILD_TYPE=${{ matrix.build }}
-          -DWHISPER_OPENBLAS=${{ matrix.blas }}
-          -DCMAKE_LIBRARY_PATH="$env:OPENBLAS_PATH/lib"
-          -DWHISPER_SDL2=${{ matrix.sdl2 }}
+          -DWHISPER_SUPPORT_OPENBLAS=${{ matrix.blas }}
+          -DCMAKE_LIBRARY_PATH="$env:blasdir/lib"
+          -DWHISPER_SUPPORT_SDL2=${{ matrix.sdl2 }}
 
       - name: Build
         run: |
@@ -263,7 +263,7 @@ jobs:
 
       - name: Copy libopenblas.dll
         if: matrix.blas == 'ON'
-        run: copy "$env:OPENBLAS_PATH/bin/libopenblas.dll" build/bin/${{ matrix.build }}
+        run: copy "$env:blasdir/bin/libopenblas.dll" build/bin/${{ matrix.build }}
 
       - name: Copy SDL2.dll
         if: matrix.sdl2 == 'ON'

.gitignore

@@ -18,11 +18,6 @@ build-no-accel/
 build-sanitize-addr/
 build-sanitize-thread/
 
-# SPM
-.build/
-.swiftpm
-*.metallib
-
 /main
 /stream
 /command

CMakeLists.txt

@@ -1,6 +1,6 @@
 cmake_minimum_required (VERSION 3.5)
 
-project(whisper.cpp VERSION 1.4.3)
+project(whisper.cpp VERSION 1.4.2)
 
 # Add path to modules
 list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/cmake/")

Makefile

@@ -417,10 +417,9 @@ samples:
 .PHONY: medium.en
 .PHONY: medium
 .PHONY: large-v1
-.PHONY: large-v2
 .PHONY: large
 
-tiny.en tiny base.en base small.en small medium.en medium large-v1 large-v2 large: main
+tiny.en tiny base.en base small.en small medium.en medium large-v1 large: main
 	bash ./models/download-ggml-model.sh $@
 	@echo ""
 	@echo "==============================================="

Package.swift

@@ -1,77 +0,0 @@
-// swift-tools-version:5.5
-
-import PackageDescription
-
-#if arch(arm) || arch(arm64)
-let platforms: [SupportedPlatform]? = [
-    .macOS(.v12),
-    .iOS(.v14),
-    .watchOS(.v4),
-    .tvOS(.v14)
-]
-let exclude: [String] = []
-let resources: [Resource] = [
-    .process("ggml-metal.metal")
-]
-let additionalSources: [String] = ["ggml-metal.m"]
-let additionalSettings: [CSetting] = [
-    .unsafeFlags(["-fno-objc-arc"]),
-    .define("GGML_USE_METAL")
-]
-#else
-let platforms: [SupportedPlatform]? = nil
-let exclude: [String] = ["ggml-metal.metal"]
-let resources: [Resource] = []
-let additionalSources: [String] = []
-let additionalSettings: [CSetting] = []
-#endif
-
-let package = Package(
-    name: "whisper",
-    platforms: platforms,
-    products: [
-        .library(name: "whisper", targets: ["whisper"]),
-    ],
-    targets: [
-        .target(
-            name: "whisper",
-            path: ".",
-            exclude: exclude + [
-               "bindings",
-               "cmake",
-               "coreml",
-               "examples",
-               "extra",
-               "models",
-               "samples",
-               "tests",
-               "CMakeLists.txt",
-               "ggml-cuda.cu",
-               "ggml-cuda.h",
-               "Makefile"
-            ],
-            sources: [
-                "ggml.c",
-                "whisper.cpp",
-                "ggml-alloc.c",
-                "ggml-backend.c",
-                "ggml-quants.c"
-            ] + additionalSources,
-            resources: resources,
-            publicHeadersPath: "spm-headers",
-            cSettings: [
-                .unsafeFlags(["-Wno-shorten-64-to-32", "-O3", "-DNDEBUG"]),
-                .define("GGML_USE_ACCELERATE")
-                // NOTE: NEW_LAPACK will required iOS version 16.4+
-                // We should consider add this in the future when we drop support for iOS 14
-                // (ref: ref: https://developer.apple.com/documentation/accelerate/1513264-cblas_sgemm?language=objc)
-                // .define("ACCELERATE_NEW_LAPACK"),
-                // .define("ACCELERATE_LAPACK_ILP64")
-            ] + additionalSettings,
-            linkerSettings: [
-                .linkedFramework("Accelerate")
-            ]
-        )
-    ],
-    cxxLanguageStandard: .cxx11
-)

README.md

@@ -6,7 +6,7 @@
 [![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](https://opensource.org/licenses/MIT)
 [![npm](https://img.shields.io/npm/v/whisper.cpp.svg)](https://www.npmjs.com/package/whisper.cpp/)
 
-Beta: [v1.4.3](https://github.com/ggerganov/whisper.cpp/releases/tag/v1.4.3) / Stable: [v1.2.1](https://github.com/ggerganov/whisper.cpp/releases/tag/v1.2.1) / [Roadmap | F.A.Q.](https://github.com/ggerganov/whisper.cpp/discussions/126)
+Beta: [v1.4.2](https://github.com/ggerganov/whisper.cpp/releases/tag/v1.4.2) / Stable: [v1.2.1](https://github.com/ggerganov/whisper.cpp/releases/tag/v1.2.1) / [Roadmap | F.A.Q.](https://github.com/ggerganov/whisper.cpp/discussions/126)
 
 High-performance inference of [OpenAI's Whisper](https://github.com/openai/whisper) automatic speech recognition (ASR) model:
 
@@ -234,7 +234,6 @@ make small
 make medium.en
 make medium
 make large-v1
-make large-v2
 make large
 ```
 
@@ -246,7 +245,7 @@ make large
 | base   | 142 MB | ~210 MB | `465707469ff3a37a2b9b8d8f89f2f99de7299dac` |
 | small  | 466 MB | ~600 MB | `55356645c2b361a969dfd0ef2c5a50d530afd8d5` |
 | medium | 1.5 GB | ~1.7 GB | `fd9727b6e1217c2f614f9b698455c4ffd82463b4` |
-| large  | 2.9 GB | ~3.3 GB | `ad82bf6a9043ceed055076d0fd39f5f186ff8062` |
+| large  | 2.9 GB | ~3.3 GB | `0f4c8e34f21cf1a914c59d8b3ce882345ad349d6` |
 
 ## Quantization
 

bindings/go/examples/go-model-download/main.go

@@ -24,7 +24,7 @@ const (
 
 var (
 	// The models which will be downloaded, if no model is specified as an argument
-	modelNames = []string{"ggml-tiny.en", "ggml-tiny", "ggml-base.en", "ggml-base", "ggml-small.en", "ggml-small", "ggml-medium.en", "ggml-medium", "ggml-large-v1", "ggml-large-v2", "ggml-large"}
+	modelNames = []string{"ggml-tiny.en", "ggml-tiny", "ggml-base.en", "ggml-base", "ggml-small.en", "ggml-small", "ggml-medium.en", "ggml-medium", "ggml-large-v1", "ggml-large"}
 )
 
 var (

bindings/go/whisper.go

@@ -83,6 +83,7 @@ const (
 	SampleRate = C.WHISPER_SAMPLE_RATE                 // Expected sample rate, samples per second
 	SampleBits = uint16(unsafe.Sizeof(C.float(0))) * 8 // Sample size in bits
 	NumFFT     = C.WHISPER_N_FFT
+	NumMEL     = C.WHISPER_N_MEL
 	HopLength  = C.WHISPER_HOP_LENGTH
 	ChunkSize  = C.WHISPER_CHUNK_SIZE
 )
@@ -102,7 +103,7 @@ var (
 func Whisper_init(path string) *Context {
 	cPath := C.CString(path)
 	defer C.free(unsafe.Pointer(cPath))
-	if ctx := C.whisper_init_from_file_with_params(cPath, C.whisper_context_default_params()); ctx != nil {
+	if ctx := C.whisper_init_from_file(cPath); ctx != nil {
 		return (*Context)(ctx)
 	} else {
 		return nil

bindings/java/src/main/java/io/github/ggerganov/whispercpp/WhisperContext.java

@@ -4,7 +4,6 @@ import com.sun.jna.Structure;
 import com.sun.jna.ptr.PointerByReference;
 import io.github.ggerganov.whispercpp.ggml.GgmlType;
 import io.github.ggerganov.whispercpp.WhisperModel;
-import io.github.ggerganov.whispercpp.params.WhisperContextParams;
 
 import java.util.List;
 
@@ -24,9 +23,8 @@ public class WhisperContext extends Structure {
     public PointerByReference vocab;
     public PointerByReference state;
 
-    /** populated by whisper_init_from_file_with_params() */
+    /** populated by whisper_init_from_file() */
     String path_model;
-    WhisperContextParams params;
 
 //    public static class ByReference extends WhisperContext implements Structure.ByReference {
 //    }

bindings/java/src/main/java/io/github/ggerganov/whispercpp/WhisperCpp.java

@@ -2,7 +2,6 @@ package io.github.ggerganov.whispercpp;
 
 import com.sun.jna.Native;
 import com.sun.jna.Pointer;
-import io.github.ggerganov.whispercpp.params.WhisperContextParams;
 import io.github.ggerganov.whispercpp.params.WhisperFullParams;
 import io.github.ggerganov.whispercpp.params.WhisperSamplingStrategy;
 
@@ -16,9 +15,8 @@ import java.io.IOException;
 public class WhisperCpp implements AutoCloseable {
     private WhisperCppJnaLibrary lib = WhisperCppJnaLibrary.instance;
     private Pointer ctx = null;
-    private Pointer paramsPointer = null;
-    private Pointer greedyParamsPointer = null;
-    private Pointer beamParamsPointer = null;
+    private Pointer greedyPointer = null;
+    private Pointer beamPointer = null;
 
     public File modelDir() {
         String modelDirPath = System.getenv("XDG_CACHE_HOME");
@@ -33,18 +31,6 @@ public class WhisperCpp implements AutoCloseable {
      * @param modelPath - absolute path, or just the name (eg: "base", "base-en" or "base.en")
      */
     public void initContext(String modelPath) throws FileNotFoundException {
-        initContextImpl(modelPath, getContextDefaultParams());
-    }
-
-    /**
-     * @param modelPath - absolute path, or just the name (eg: "base", "base-en" or "base.en")
-     * @param params - params to use when initialising the context
-     */
-    public void initContext(String modelPath, WhisperContextParams params) throws FileNotFoundException {
-        initContextImpl(modelPath, params);
-    }
-
-    private void initContextImpl(String modelPath, WhisperContextParams params) throws FileNotFoundException {
         if (ctx != null) {
             lib.whisper_free(ctx);
         }
@@ -57,26 +43,13 @@ public class WhisperCpp implements AutoCloseable {
             modelPath = new File(modelDir(), modelPath).getAbsolutePath();
         }
 
-        ctx = lib.whisper_init_from_file_with_params(modelPath, params);
+        ctx = lib.whisper_init_from_file(modelPath);
 
         if (ctx == null) {
             throw new FileNotFoundException(modelPath);
         }
     }
 
-    /**
-     * Provides default params which can be used with `whisper_init_from_file_with_params()` etc.
-     * Because this function allocates memory for the params, the caller must call either:
-     * - call `whisper_free_context_params()`
-     * - `Native.free(Pointer.nativeValue(pointer));`
-     */
-    public WhisperContextParams getContextDefaultParams() {
-        paramsPointer = lib.whisper_context_default_params_by_ref();
-        WhisperContextParams params = new WhisperContextParams(paramsPointer);
-        params.read();
-        return params;
-    }
-    
     /**
      * Provides default params which can be used with `whisper_full()` etc.
      * Because this function allocates memory for the params, the caller must call either:
@@ -90,15 +63,15 @@ public class WhisperCpp implements AutoCloseable {
 
         // whisper_full_default_params_by_ref allocates memory which we need to delete, so only create max 1 pointer for each strategy.
         if (strategy == WhisperSamplingStrategy.WHISPER_SAMPLING_GREEDY) {
-            if (greedyParamsPointer == null) {
-                greedyParamsPointer = lib.whisper_full_default_params_by_ref(strategy.ordinal());
+            if (greedyPointer == null) {
+                greedyPointer = lib.whisper_full_default_params_by_ref(strategy.ordinal());
             }
-            pointer = greedyParamsPointer;
+            pointer = greedyPointer;
         } else {
-            if (beamParamsPointer == null) {
-                beamParamsPointer = lib.whisper_full_default_params_by_ref(strategy.ordinal());
+            if (beamPointer == null) {
+                beamPointer = lib.whisper_full_default_params_by_ref(strategy.ordinal());
             }
-            pointer = beamParamsPointer;
+            pointer = beamPointer;
         }
 
         WhisperFullParams params = new WhisperFullParams(pointer);
@@ -120,17 +93,13 @@ public class WhisperCpp implements AutoCloseable {
     }
 
     private void freeParams() {
-        if (paramsPointer != null) {
-            Native.free(Pointer.nativeValue(paramsPointer));
-            paramsPointer = null;
+        if (greedyPointer != null) {
+            Native.free(Pointer.nativeValue(greedyPointer));
+            greedyPointer = null;
         }
-        if (greedyParamsPointer != null) {
-            Native.free(Pointer.nativeValue(greedyParamsPointer));
-            greedyParamsPointer = null;
-        }
-        if (beamParamsPointer != null) {
-            Native.free(Pointer.nativeValue(beamParamsPointer));
-            beamParamsPointer = null;
+        if (beamPointer != null) {
+            Native.free(Pointer.nativeValue(beamPointer));
+            beamPointer = null;
         }
     }
 

bindings/java/src/main/java/io/github/ggerganov/whispercpp/WhisperCppJnaLibrary.java

@@ -5,7 +5,6 @@ import com.sun.jna.Native;
 import com.sun.jna.Pointer;
 import io.github.ggerganov.whispercpp.model.WhisperModelLoader;
 import io.github.ggerganov.whispercpp.model.WhisperTokenData;
-import io.github.ggerganov.whispercpp.params.WhisperContextParams;
 import io.github.ggerganov.whispercpp.params.WhisperFullParams;
 
 public interface WhisperCppJnaLibrary extends Library {
@@ -14,31 +13,12 @@ public interface WhisperCppJnaLibrary extends Library {
     String whisper_print_system_info();
 
     /**
-     * DEPRECATED. Allocate (almost) all memory needed for the model by loading from a file.
+     * Allocate (almost) all memory needed for the model by loading from a file.
      *
      * @param path_model Path to the model file
      * @return Whisper context on success, null on failure
      */
     Pointer whisper_init_from_file(String path_model);
-    
-    /**
-     * Provides default params which can be used with `whisper_init_from_file_with_params()` etc.
-     * Because this function allocates memory for the params, the caller must call either:
-     * - call `whisper_free_context_params()`
-     * - `Native.free(Pointer.nativeValue(pointer));`
-     */
-    Pointer whisper_context_default_params_by_ref();
-
-    void whisper_free_context_params(Pointer params);
-
-    /**
-     * Allocate (almost) all memory needed for the model by loading from a file.
-     *
-     * @param path_model Path to the model file
-     * @param params     Pointer to whisper_context_params
-     * @return Whisper context on success, null on failure
-     */
-    Pointer whisper_init_from_file_with_params(String path_model, WhisperContextParams params);
 
     /**
      * Allocate (almost) all memory needed for the model by loading from a buffer.

bindings/java/src/main/java/io/github/ggerganov/whispercpp/params/WhisperContextParams.java

@@ -1,31 +0,0 @@
-package io.github.ggerganov.whispercpp.params;
-
-import com.sun.jna.*;
-
-import java.util.Arrays;
-import java.util.List;
-
-/**
- * Parameters for the whisper_init_from_file_with_params() function.
- * If you change the order or add new parameters, make sure to update the default values in whisper.cpp:
- * whisper_context_default_params()
- */
-public class WhisperContextParams extends Structure {
-
-    public WhisperContextParams(Pointer p) {
-        super(p);
-    }
-
-    /** Use GPU for inference Number (default = true) */
-    public CBool use_gpu;
-
-    /** Use GPU for inference Number (default = true) */
-    public void useGpu(boolean enable) {
-        use_gpu = enable ? CBool.TRUE : CBool.FALSE;
-    }
-
-    @Override
-    protected List<String> getFieldOrder() {
-        return Arrays.asList("use_gpu");
-    }
-}

bindings/javascript/emscripten.cpp

@@ -20,7 +20,7 @@ struct whisper_context * g_context;
 EMSCRIPTEN_BINDINGS(whisper) {
     emscripten::function("init", emscripten::optional_override([](const std::string & path_model) {
         if (g_context == nullptr) {
-            g_context = whisper_init_from_file_with_params(path_model.c_str(), whisper_context_default_params());
+            g_context = whisper_init_from_file(path_model.c_str());
             if (g_context != nullptr) {
                 return true;
             } else {

bindings/javascript/package.json

@@ -1,6 +1,6 @@
 {
   "name": "whisper.cpp",
-  "version": "1.4.3",
+  "version": "1.4.2",
   "description": "Whisper speech recognition",
   "main": "whisper.js",
   "scripts": {

bindings/ruby/ext/ruby_whisper.cpp

@@ -87,7 +87,7 @@ static VALUE ruby_whisper_initialize(int argc, VALUE *argv, VALUE self) {
   if (!rb_respond_to(whisper_model_file_path, rb_intern("to_s"))) {
     rb_raise(rb_eRuntimeError, "Expected file path to model to initialize Whisper::Context");
   }
-  rw->context = whisper_init_from_file_with_params(StringValueCStr(whisper_model_file_path), whisper_context_default_params());
+  rw->context = whisper_init_from_file(StringValueCStr(whisper_model_file_path));
   if (rw->context == nullptr) {
     rb_raise(rb_eRuntimeError, "error: failed to initialize whisper context");
   }

examples/addon.node/__test__/whisper.spec.js

@@ -11,7 +11,6 @@ const whisperParamsMock = {
   language: "en",
   model: path.join(__dirname, "../../../models/ggml-base.en.bin"),
   fname_inp: path.join(__dirname, "../../../samples/jfk.wav"),
-  use_gpu: true,
 };
 
 describe("Run whisper.node", () => {

examples/addon.node/addon.cpp

@@ -36,7 +36,6 @@ struct whisper_params {
     bool print_colors   = false;
     bool print_progress = false;
     bool no_timestamps  = false;
-    bool use_gpu        = true;
 
     std::string language = "en";
     std::string prompt;
@@ -154,9 +153,7 @@ int run(whisper_params &params, std::vector<std::vector<std::string>> &result) {
 
     // whisper init
 
-    struct whisper_context_params cparams;
-    cparams.use_gpu = params.use_gpu;
-    struct whisper_context * ctx = whisper_init_from_file_with_params(params.model.c_str(), cparams);
+    struct whisper_context * ctx = whisper_init_from_file(params.model.c_str());
 
     if (ctx == nullptr) {
         fprintf(stderr, "error: failed to initialize whisper context\n");
@@ -318,12 +315,10 @@ Napi::Value whisper(const Napi::CallbackInfo& info) {
   std::string language = whisper_params.Get("language").As<Napi::String>();
   std::string model = whisper_params.Get("model").As<Napi::String>();
   std::string input = whisper_params.Get("fname_inp").As<Napi::String>();
-  bool use_gpu = whisper_params.Get("use_gpu").As<Napi::Boolean>();
 
   params.language = language;
   params.model = model;
   params.fname_inp.emplace_back(input);
-  params.use_gpu = use_gpu;
 
   Napi::Function callback = info[1].As<Napi::Function>();
   Worker* worker = new Worker(callback, params);

examples/addon.node/index.js

@@ -11,7 +11,6 @@ const whisperParams = {
   language: "en",
   model: path.join(__dirname, "../../models/ggml-base.en.bin"),
   fname_inp: "../../samples/jfk.wav",
-  use_gpu: true,
 };
 
 const arguments = process.argv.slice(2);

examples/bench.wasm/emscripten.cpp

@@ -23,9 +23,7 @@ void bench_main(size_t index) {
 
     fprintf(stderr, "%s: running benchmark with %d threads - please wait...\n", __func__, n_threads);
 
-    const int n_mels = whisper_model_n_mels(ctx);
-
-    if (int ret = whisper_set_mel(ctx, nullptr, 0, n_mels)) {
+    if (int ret = whisper_set_mel(ctx, nullptr, 0, WHISPER_N_MEL)) {
         fprintf(stderr, "error: failed to set mel: %d\n", ret);
         return;
     }
@@ -59,7 +57,7 @@ EMSCRIPTEN_BINDINGS(bench) {
     emscripten::function("init", emscripten::optional_override([](const std::string & path_model) {
         for (size_t i = 0; i < g_contexts.size(); ++i) {
             if (g_contexts[i] == nullptr) {
-                g_contexts[i] = whisper_init_from_file_with_params(path_model.c_str(), whisper_context_default_params());
+                g_contexts[i] = whisper_init_from_file(path_model.c_str());
                 if (g_contexts[i] != nullptr) {
                     if (g_worker.joinable()) {
                         g_worker.join();

examples/bench/bench.cpp

@@ -11,8 +11,6 @@ struct whisper_params {
     int32_t what = 0; // what to benchmark: 0 - whisper ecoder, 1 - memcpy, 2 - ggml_mul_mat
 
     std::string model = "models/ggml-base.en.bin";
-
-    bool use_gpu = true;
 };
 
 void whisper_print_usage(int argc, char ** argv, const whisper_params & params);
@@ -25,10 +23,9 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params) {
             whisper_print_usage(argc, argv, params);
             exit(0);
         }
-        else if (arg == "-t"  || arg == "--threads") { params.n_threads = std::stoi(argv[++i]); }
-        else if (arg == "-m"  || arg == "--model")   { params.model     = argv[++i]; }
-        else if (arg == "-w"  || arg == "--what")    { params.what      = atoi(argv[++i]); }
-        else if (arg == "-ng" || arg == "--no-gpu")  { params.use_gpu   = false; }
+        else if (arg == "-t" || arg == "--threads") { params.n_threads = std::stoi(argv[++i]); }
+        else if (arg == "-m" || arg == "--model")   { params.model     = argv[++i]; }
+        else if (arg == "-w" || arg == "--what")    { params.what     = atoi(argv[++i]); }
         else {
             fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
             whisper_print_usage(argc, argv, params);
@@ -48,7 +45,6 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
     fprintf(stderr, "  -t N,     --threads N   [%-7d] number of threads to use during computation\n", params.n_threads);
     fprintf(stderr, "  -m FNAME, --model FNAME [%-7s] model path\n",                                  params.model.c_str());
     fprintf(stderr, "  -w N,     --what N      [%-7d] what to benchmark:\n",                          params.what);
-    fprintf(stderr, "  -ng,      --no-gpu      [%-7s] disable GPU\n",                                 params.use_gpu ? "false" : "true");
     fprintf(stderr, "                           %-7s  0 - whisper\n",                                 "");
     fprintf(stderr, "                           %-7s  1 - memcpy\n",                                  "");
     fprintf(stderr, "                           %-7s  2 - ggml_mul_mat\n",                            "");
@@ -58,10 +54,7 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
 int whisper_bench_full(const whisper_params & params) {
     // whisper init
 
-    struct whisper_context_params cparams;
-    cparams.use_gpu = params.use_gpu;
-
-    struct whisper_context * ctx = whisper_init_from_file_with_params(params.model.c_str(), cparams);
+    struct whisper_context * ctx = whisper_init_from_file(params.model.c_str());
 
     {
         fprintf(stderr, "\n");
@@ -73,9 +66,7 @@ int whisper_bench_full(const whisper_params & params) {
         return 2;
     }
 
-    const int n_mels = whisper_model_n_mels(ctx);
-
-    if (int ret = whisper_set_mel(ctx, nullptr, 0, n_mels)) {
+    if (int ret = whisper_set_mel(ctx, nullptr, 0, WHISPER_N_MEL)) {
         fprintf(stderr, "error: failed to set mel: %d\n", ret);
         return 3;
     }

examples/command.wasm/emscripten.cpp

@@ -243,7 +243,7 @@ EMSCRIPTEN_BINDINGS(command) {
     emscripten::function("init", emscripten::optional_override([](const std::string & path_model) {
         for (size_t i = 0; i < g_contexts.size(); ++i) {
             if (g_contexts[i] == nullptr) {
-                g_contexts[i] = whisper_init_from_file_with_params(path_model.c_str(), whisper_context_default_params());
+                g_contexts[i] = whisper_init_from_file(path_model.c_str());
                 if (g_contexts[i] != nullptr) {
                     g_running = true;
                     if (g_worker.joinable()) {

examples/command/command.cpp

@@ -38,7 +38,6 @@ struct whisper_params {
     bool print_special = false;
     bool print_energy  = false;
     bool no_timestamps = true;
-    bool use_gpu       = true;
 
     std::string language  = "en";
     std::string model     = "models/ggml-base.en.bin";
@@ -69,7 +68,6 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params) {
         else if (arg == "-tr"  || arg == "--translate")     { params.translate     = true; }
         else if (arg == "-ps"  || arg == "--print-special") { params.print_special = true; }
         else if (arg == "-pe"  || arg == "--print-energy")  { params.print_energy  = true; }
-        else if (arg == "-ng"  || arg == "--no-gpu")        { params.use_gpu       = false; }
         else if (arg == "-l"   || arg == "--language")      { params.language      = argv[++i]; }
         else if (arg == "-m"   || arg == "--model")         { params.model         = argv[++i]; }
         else if (arg == "-f"   || arg == "--file")          { params.fname_out     = argv[++i]; }
@@ -103,7 +101,6 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
     fprintf(stderr, "  -tr,        --translate      [%-7s] translate from source language to english\n",   params.translate ? "true" : "false");
     fprintf(stderr, "  -ps,        --print-special  [%-7s] print special tokens\n",                        params.print_special ? "true" : "false");
     fprintf(stderr, "  -pe,        --print-energy   [%-7s] print sound energy (for debugging)\n",          params.print_energy ? "true" : "false");
-    fprintf(stderr, "  -ng,        --no-gpu         [%-7s] disable GPU\n",                                 params.use_gpu ? "false" : "true");
     fprintf(stderr, "  -l LANG,    --language LANG  [%-7s] spoken language\n",                             params.language.c_str());
     fprintf(stderr, "  -m FNAME,   --model FNAME    [%-7s] model path\n",                                  params.model.c_str());
     fprintf(stderr, "  -f FNAME,   --file FNAME     [%-7s] text output file name\n",                       params.fname_out.c_str());
@@ -613,10 +610,7 @@ int main(int argc, char ** argv) {
 
     // whisper init
 
-    struct whisper_context_params cparams;
-    cparams.use_gpu = params.use_gpu;
-
-    struct whisper_context * ctx = whisper_init_from_file_with_params(params.model.c_str(), cparams);
+    struct whisper_context * ctx = whisper_init_from_file(params.model.c_str());
 
     // print some info about the processing
     {

examples/livestream.sh

@@ -48,7 +48,7 @@ if [ -n "$3" ]; then
 fi
 
 # Whisper models
-models=( "tiny.en" "tiny" "base.en" "base" "small.en" "small" "medium.en" "medium" "large-v1" "large-v2" "large" )
+models=( "tiny.en" "tiny" "base.en" "base" "small.en" "small" "medium.en" "medium" "large-v1" "large" )
 
 # list available models
 function list_models {

examples/lsp/lsp.cpp

@@ -30,7 +30,6 @@ struct whisper_params {
     bool translate     = false;
     bool print_special = false;
     bool print_energy  = false;
-    bool use_gpu       = true;
 
     std::string language  = "en";
     std::string model     = "models/ggml-base.en.bin";
@@ -73,7 +72,6 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params) {
         else if (arg == "-tr"  || arg == "--translate")     { params.translate     = true; }
         else if (arg == "-ps"  || arg == "--print-special") { params.print_special = true; }
         else if (arg == "-pe"  || arg == "--print-energy")  { params.print_energy  = true; }
-        else if (arg == "-ng"  || arg == "--no-gpu")        { params.use_gpu       = false; }
         else if (arg == "-l"   || arg == "--language")      { params.language      = argv[++i]; }
         else if (arg == "-m"   || arg == "--model")         { params.model         = argv[++i]; }
         else {
@@ -104,7 +102,6 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
     fprintf(stderr, "  -tr,        --translate      [%-7s] translate from source language to english\n",   params.translate ? "true" : "false");
     fprintf(stderr, "  -ps,        --print-special  [%-7s] print special tokens\n",                        params.print_special ? "true" : "false");
     fprintf(stderr, "  -pe,        --print-energy   [%-7s] print sound energy (for debugging)\n",          params.print_energy ? "true" : "false");
-    fprintf(stderr, "  -ng,        --no-gpu         [%-7s] disable GPU\n",                                 params.use_gpu ? "false" : "true");
     fprintf(stderr, "  -l LANG,    --language LANG  [%-7s] spoken language\n",                             params.language.c_str());
     fprintf(stderr, "  -m FNAME,   --model FNAME    [%-7s] model path\n",                                  params.model.c_str());
     fprintf(stderr, "\n");
@@ -435,9 +432,7 @@ int main(int argc, char ** argv) {
     }
 
     // whisper init
-    struct whisper_context_params cparams;
-    cparams.use_gpu = params.use_gpu;
-    struct whisper_context * ctx = whisper_init_from_file_with_params(params.model.c_str(), cparams);
+    struct whisper_context * ctx = whisper_init_from_file(params.model.c_str());
     // init audio
 
     audio_async audio(30*1000);

examples/main/main.cpp

@@ -90,7 +90,6 @@ struct whisper_params {
     bool print_progress  = false;
     bool no_timestamps   = false;
     bool log_score       = false;
-    bool use_gpu         = true;
 
     std::string language  = "en";
     std::string prompt;
@@ -166,7 +165,6 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params) {
         else if (arg == "-f"    || arg == "--file")            { params.fname_inp.emplace_back(argv[++i]); }
         else if (arg == "-oved" || arg == "--ov-e-device")     { params.openvino_encode_device = argv[++i]; }
         else if (arg == "-ls"   || arg == "--log-score")       { params.log_score = true; }
-        else if (arg == "-ng"   || arg == "--no-gpu")          { params.use_gpu = false; }
         else {
             fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
             whisper_print_usage(argc, argv, params);
@@ -223,7 +221,6 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
     fprintf(stderr, "  -f FNAME,  --file FNAME        [%-7s] input WAV file path\n",                            "");
     fprintf(stderr, "  -oved D,   --ov-e-device DNAME [%-7s] the OpenVINO device used for encode inference\n",  params.openvino_encode_device.c_str());
     fprintf(stderr, "  -ls,       --log-score         [%-7s] log best decoder scores of tokens\n",              params.log_score?"true":"false");
-    fprintf(stderr, "  -ng,       --no-gpu            [%-7s] disable GPU\n",                                    params.use_gpu ? "false" : "true");
     fprintf(stderr, "\n");
 }
 
@@ -880,10 +877,7 @@ int main(int argc, char ** argv) {
 
     // whisper init
 
-    struct whisper_context_params cparams;
-    cparams.use_gpu = params.use_gpu;
-
-    struct whisper_context * ctx = whisper_init_from_file_with_params(params.model.c_str(), cparams);
+    struct whisper_context * ctx = whisper_init_from_file(params.model.c_str());
 
     if (ctx == nullptr) {
         fprintf(stderr, "error: failed to initialize whisper context\n");

examples/stream.wasm/emscripten.cpp

@@ -132,7 +132,7 @@ EMSCRIPTEN_BINDINGS(stream) {
     emscripten::function("init", emscripten::optional_override([](const std::string & path_model) {
         for (size_t i = 0; i < g_contexts.size(); ++i) {
             if (g_contexts[i] == nullptr) {
-                g_contexts[i] = whisper_init_from_file_with_params(path_model.c_str(), whisper_context_default_params());
+                g_contexts[i] = whisper_init_from_file(path_model.c_str());
                 if (g_contexts[i] != nullptr) {
                     g_running = true;
                     if (g_worker.joinable()) {

examples/stream/stream.cpp

@@ -48,12 +48,11 @@ struct whisper_params {
     bool no_context    = true;
     bool no_timestamps = false;
     bool tinydiarize   = false;
-    bool save_audio    = false; // save audio to wav file
-    bool use_gpu       = true;
 
     std::string language  = "en";
     std::string model     = "models/ggml-base.en.bin";
     std::string fname_out;
+    bool save_audio = false; // save audio to wav file
 };
 
 void whisper_print_usage(int argc, char ** argv, const whisper_params & params);
@@ -66,26 +65,25 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params) {
             whisper_print_usage(argc, argv, params);
             exit(0);
         }
-        else if (arg == "-t"    || arg == "--threads")       { params.n_threads     = std::stoi(argv[++i]); }
-        else if (                  arg == "--step")          { params.step_ms       = std::stoi(argv[++i]); }
-        else if (                  arg == "--length")        { params.length_ms     = std::stoi(argv[++i]); }
-        else if (                  arg == "--keep")          { params.keep_ms       = std::stoi(argv[++i]); }
-        else if (arg == "-c"    || arg == "--capture")       { params.capture_id    = std::stoi(argv[++i]); }
-        else if (arg == "-mt"   || arg == "--max-tokens")    { params.max_tokens    = std::stoi(argv[++i]); }
-        else if (arg == "-ac"   || arg == "--audio-ctx")     { params.audio_ctx     = std::stoi(argv[++i]); }
-        else if (arg == "-vth"  || arg == "--vad-thold")     { params.vad_thold     = std::stof(argv[++i]); }
-        else if (arg == "-fth"  || arg == "--freq-thold")    { params.freq_thold    = std::stof(argv[++i]); }
-        else if (arg == "-su"   || arg == "--speed-up")      { params.speed_up      = true; }
-        else if (arg == "-tr"   || arg == "--translate")     { params.translate     = true; }
-        else if (arg == "-nf"   || arg == "--no-fallback")   { params.no_fallback   = true; }
-        else if (arg == "-ps"   || arg == "--print-special") { params.print_special = true; }
-        else if (arg == "-kc"   || arg == "--keep-context")  { params.no_context    = false; }
-        else if (arg == "-l"    || arg == "--language")      { params.language      = argv[++i]; }
-        else if (arg == "-m"    || arg == "--model")         { params.model         = argv[++i]; }
-        else if (arg == "-f"    || arg == "--file")          { params.fname_out     = argv[++i]; }
-        else if (arg == "-tdrz" || arg == "--tinydiarize")   { params.tinydiarize   = true; }
-        else if (arg == "-sa"   || arg == "--save-audio")    { params.save_audio    = true; }
-        else if (arg == "-ng"   || arg == "--no-gpu")        { params.use_gpu       = false; }
+        else if (arg == "-t"   || arg == "--threads")       { params.n_threads     = std::stoi(argv[++i]); }
+        else if (                 arg == "--step")          { params.step_ms       = std::stoi(argv[++i]); }
+        else if (                 arg == "--length")        { params.length_ms     = std::stoi(argv[++i]); }
+        else if (                 arg == "--keep")          { params.keep_ms       = std::stoi(argv[++i]); }
+        else if (arg == "-c"   || arg == "--capture")       { params.capture_id    = std::stoi(argv[++i]); }
+        else if (arg == "-mt"  || arg == "--max-tokens")    { params.max_tokens    = std::stoi(argv[++i]); }
+        else if (arg == "-ac"  || arg == "--audio-ctx")     { params.audio_ctx     = std::stoi(argv[++i]); }
+        else if (arg == "-vth" || arg == "--vad-thold")     { params.vad_thold     = std::stof(argv[++i]); }
+        else if (arg == "-fth" || arg == "--freq-thold")    { params.freq_thold    = std::stof(argv[++i]); }
+        else if (arg == "-su"  || arg == "--speed-up")      { params.speed_up      = true; }
+        else if (arg == "-tr"  || arg == "--translate")     { params.translate     = true; }
+        else if (arg == "-nf"  || arg == "--no-fallback")   { params.no_fallback   = true; }
+        else if (arg == "-ps"  || arg == "--print-special") { params.print_special = true; }
+        else if (arg == "-kc"  || arg == "--keep-context")  { params.no_context    = false; }
+        else if (arg == "-l"   || arg == "--language")      { params.language      = argv[++i]; }
+        else if (arg == "-m"   || arg == "--model")         { params.model         = argv[++i]; }
+        else if (arg == "-f"   || arg == "--file")          { params.fname_out     = argv[++i]; }
+        else if (arg == "-tdrz" || arg == "--tinydiarize")  { params.tinydiarize   = true; }
+        else if (arg == "-sa"  || arg == "--save-audio")    { params.save_audio    = true; }
 
         else {
             fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
@@ -120,9 +118,8 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
     fprintf(stderr, "  -l LANG,  --language LANG [%-7s] spoken language\n",                                params.language.c_str());
     fprintf(stderr, "  -m FNAME, --model FNAME   [%-7s] model path\n",                                     params.model.c_str());
     fprintf(stderr, "  -f FNAME, --file FNAME    [%-7s] text output file name\n",                          params.fname_out.c_str());
-    fprintf(stderr, "  -tdrz,    --tinydiarize   [%-7s] enable tinydiarize (requires a tdrz model)\n",     params.tinydiarize ? "true" : "false");
+    fprintf(stderr, "  -tdrz,     --tinydiarize  [%-7s] enable tinydiarize (requires a tdrz model)\n",     params.tinydiarize ? "true" : "false");
     fprintf(stderr, "  -sa,      --save-audio    [%-7s] save the recorded audio to a file\n",              params.save_audio ? "true" : "false");
-    fprintf(stderr, "  -ng,      --no-gpu        [%-7s] disable GPU inference\n",                          params.use_gpu ? "false" : "true");
     fprintf(stderr, "\n");
 }
 
@@ -166,10 +163,7 @@ int main(int argc, char ** argv) {
         exit(0);
     }
 
-    struct whisper_context_params cparams;
-    cparams.use_gpu = params.use_gpu;
-
-    struct whisper_context * ctx = whisper_init_from_file_with_params(params.model.c_str(), cparams);
+    struct whisper_context * ctx = whisper_init_from_file(params.model.c_str());
 
     std::vector<float> pcmf32    (n_samples_30s, 0.0f);
     std::vector<float> pcmf32_old;
@@ -430,4 +424,4 @@ int main(int argc, char ** argv) {
     whisper_free(ctx);
 
     return 0;
-}
+}

examples/talk-llama/CMakeLists.txt

@@ -14,8 +14,6 @@ if (WHISPER_SDL2)
         ../common-sdl.cpp
         ../../ggml.c
         ../../ggml-alloc.c
-        ../../ggml-backend.c
-        ../../ggml-quants.c
         ../../whisper.cpp)
 
     target_include_directories(${TARGET} PRIVATE ${SDL2_INCLUDE_DIRS} ../../)

examples/talk-llama/talk-llama.cpp

@@ -63,7 +63,6 @@ struct whisper_params {
     bool print_energy   = false;
     bool no_timestamps  = true;
     bool verbose_prompt = false;
-    bool use_gpu        = true;
 
     std::string person      = "Georgi";
     std::string language    = "en";
@@ -85,26 +84,25 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params) {
             whisper_print_usage(argc, argv, params);
             exit(0);
         }
-        else if (arg == "-t"   || arg == "--threads")        { params.n_threads      = std::stoi(argv[++i]); }
-        else if (arg == "-vms" || arg == "--voice-ms")       { params.voice_ms       = std::stoi(argv[++i]); }
-        else if (arg == "-c"   || arg == "--capture")        { params.capture_id     = std::stoi(argv[++i]); }
-        else if (arg == "-mt"  || arg == "--max-tokens")     { params.max_tokens     = std::stoi(argv[++i]); }
-        else if (arg == "-ac"  || arg == "--audio-ctx")      { params.audio_ctx      = std::stoi(argv[++i]); }
-        else if (arg == "-vth" || arg == "--vad-thold")      { params.vad_thold      = std::stof(argv[++i]); }
-        else if (arg == "-fth" || arg == "--freq-thold")     { params.freq_thold     = std::stof(argv[++i]); }
-        else if (arg == "-su"  || arg == "--speed-up")       { params.speed_up       = true; }
-        else if (arg == "-tr"  || arg == "--translate")      { params.translate      = true; }
-        else if (arg == "-ps"  || arg == "--print-special")  { params.print_special  = true; }
-        else if (arg == "-pe"  || arg == "--print-energy")   { params.print_energy   = true; }
-        else if (arg == "-vp"  || arg == "--verbose-prompt") { params.verbose_prompt = true; }
-        else if (arg == "-ng"  || arg == "--no-gpu")         { params.use_gpu        = false; }
-        else if (arg == "-p"   || arg == "--person")         { params.person         = argv[++i]; }
-        else if (arg == "--session")                         { params.path_session   = argv[++i];}
-        else if (arg == "-l"   || arg == "--language")       { params.language       = argv[++i]; }
-        else if (arg == "-mw"  || arg == "--model-whisper")  { params.model_wsp      = argv[++i]; }
-        else if (arg == "-ml"  || arg == "--model-llama")    { params.model_llama    = argv[++i]; }
-        else if (arg == "-s"   || arg == "--speak")          { params.speak          = argv[++i]; }
-        else if (arg == "--prompt-file")                     {
+        else if (arg == "-t"   || arg == "--threads")       { params.n_threads     = std::stoi(argv[++i]); }
+        else if (arg == "-vms" || arg == "--voice-ms")      { params.voice_ms      = std::stoi(argv[++i]); }
+        else if (arg == "-c"   || arg == "--capture")       { params.capture_id    = std::stoi(argv[++i]); }
+        else if (arg == "-mt"  || arg == "--max-tokens")    { params.max_tokens    = std::stoi(argv[++i]); }
+        else if (arg == "-ac"  || arg == "--audio-ctx")     { params.audio_ctx     = std::stoi(argv[++i]); }
+        else if (arg == "-vth" || arg == "--vad-thold")     { params.vad_thold     = std::stof(argv[++i]); }
+        else if (arg == "-fth" || arg == "--freq-thold")    { params.freq_thold    = std::stof(argv[++i]); }
+        else if (arg == "-su"  || arg == "--speed-up")      { params.speed_up      = true; }
+        else if (arg == "-tr"  || arg == "--translate")     { params.translate     = true; }
+        else if (arg == "-ps"  || arg == "--print-special") { params.print_special = true; }
+        else if (arg == "-pe"  || arg == "--print-energy")  { params.print_energy  = true; }
+        else if (arg == "--verbose-prompt")                 { params.verbose_prompt = true; }
+        else if (arg == "-p"   || arg == "--person")        { params.person        = argv[++i]; }
+        else if (arg == "--session")                        { params.path_session  = argv[++i];}
+        else if (arg == "-l"   || arg == "--language")      { params.language      = argv[++i]; }
+        else if (arg == "-mw"  || arg == "--model-whisper") { params.model_wsp     = argv[++i]; }
+        else if (arg == "-ml"  || arg == "--model-llama")   { params.model_llama   = argv[++i]; }
+        else if (arg == "-s"   || arg == "--speak")         { params.speak         = argv[++i]; }
+        else if (arg == "--prompt-file")                    {
             std::ifstream file(argv[++i]);
             std::copy(std::istreambuf_iterator<char>(file), std::istreambuf_iterator<char>(), back_inserter(params.prompt));
             if (params.prompt.back() == '\n') {
@@ -112,7 +110,6 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params) {
             }
         }
         else if (arg == "-f"   || arg == "--file")          { params.fname_out     = argv[++i]; }
-        else if (arg == "-ng"  || arg == "--no-gpu")        { params.use_gpu       = false; }
         else {
             fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
             whisper_print_usage(argc, argv, params);
@@ -128,28 +125,27 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
     fprintf(stderr, "usage: %s [options]\n", argv[0]);
     fprintf(stderr, "\n");
     fprintf(stderr, "options:\n");
-    fprintf(stderr, "  -h,       --help           [default] show this help message and exit\n");
-    fprintf(stderr, "  -t N,     --threads N      [%-7d] number of threads to use during computation\n", params.n_threads);
-    fprintf(stderr, "  -vms N,   --voice-ms N     [%-7d] voice duration in milliseconds\n",              params.voice_ms);
-    fprintf(stderr, "  -c ID,    --capture ID     [%-7d] capture device ID\n",                           params.capture_id);
-    fprintf(stderr, "  -mt N,    --max-tokens N   [%-7d] maximum number of tokens per audio chunk\n",    params.max_tokens);
-    fprintf(stderr, "  -ac N,    --audio-ctx N    [%-7d] audio context size (0 - all)\n",                params.audio_ctx);
-    fprintf(stderr, "  -vth N,   --vad-thold N    [%-7.2f] voice activity detection threshold\n",        params.vad_thold);
-    fprintf(stderr, "  -fth N,   --freq-thold N   [%-7.2f] high-pass frequency cutoff\n",                params.freq_thold);
-    fprintf(stderr, "  -su,      --speed-up       [%-7s] speed up audio by x2 (reduced accuracy)\n",     params.speed_up ? "true" : "false");
-    fprintf(stderr, "  -tr,      --translate      [%-7s] translate from source language to english\n",   params.translate ? "true" : "false");
-    fprintf(stderr, "  -ps,      --print-special  [%-7s] print special tokens\n",                        params.print_special ? "true" : "false");
-    fprintf(stderr, "  -pe,      --print-energy   [%-7s] print sound energy (for debugging)\n",          params.print_energy ? "true" : "false");
-    fprintf(stderr, "  -vp,      --verbose-prompt [%-7s] print prompt at start\n",                       params.verbose_prompt ? "true" : "false");
-    fprintf(stderr, "  -ng,      --no-gpu         [%-7s] disable GPU\n",                                 params.use_gpu ? "false" : "true");
-    fprintf(stderr, "  -p NAME,  --person NAME    [%-7s] person name (for prompt selection)\n",          params.person.c_str());
-    fprintf(stderr, "  -l LANG,  --language LANG  [%-7s] spoken language\n",                             params.language.c_str());
-    fprintf(stderr, "  -mw FILE, --model-whisper  [%-7s] whisper model file\n",                          params.model_wsp.c_str());
-    fprintf(stderr, "  -ml FILE, --model-llama    [%-7s] llama model file\n",                            params.model_llama.c_str());
-    fprintf(stderr, "  -s FILE,  --speak TEXT     [%-7s] command for TTS\n",                             params.speak.c_str());
-    fprintf(stderr, "  --prompt-file FNAME        [%-7s] file with custom prompt to start dialog\n",     "");
-    fprintf(stderr, "  --session FNAME                   file to cache model state in (may be large!) (default: none)\n");
-    fprintf(stderr, "  -f FNAME, --file FNAME     [%-7s] text output file name\n",                       params.fname_out.c_str());
+    fprintf(stderr, "  -h,       --help          [default] show this help message and exit\n");
+    fprintf(stderr, "  -t N,     --threads N     [%-7d] number of threads to use during computation\n", params.n_threads);
+    fprintf(stderr, "  -vms N,   --voice-ms N    [%-7d] voice duration in milliseconds\n",              params.voice_ms);
+    fprintf(stderr, "  -c ID,    --capture ID    [%-7d] capture device ID\n",                           params.capture_id);
+    fprintf(stderr, "  -mt N,    --max-tokens N  [%-7d] maximum number of tokens per audio chunk\n",    params.max_tokens);
+    fprintf(stderr, "  -ac N,    --audio-ctx N   [%-7d] audio context size (0 - all)\n",                params.audio_ctx);
+    fprintf(stderr, "  -vth N,   --vad-thold N   [%-7.2f] voice activity detection threshold\n",        params.vad_thold);
+    fprintf(stderr, "  -fth N,   --freq-thold N  [%-7.2f] high-pass frequency cutoff\n",                params.freq_thold);
+    fprintf(stderr, "  -su,      --speed-up      [%-7s] speed up audio by x2 (reduced accuracy)\n",     params.speed_up ? "true" : "false");
+    fprintf(stderr, "  -tr,      --translate     [%-7s] translate from source language to english\n",   params.translate ? "true" : "false");
+    fprintf(stderr, "  -ps,      --print-special [%-7s] print special tokens\n",                        params.print_special ? "true" : "false");
+    fprintf(stderr, "  -pe,      --print-energy  [%-7s] print sound energy (for debugging)\n",          params.print_energy ? "true" : "false");
+    fprintf(stderr, "  -p NAME,  --person NAME   [%-7s] person name (for prompt selection)\n",          params.person.c_str());
+    fprintf(stderr, "  -l LANG,  --language LANG [%-7s] spoken language\n",                             params.language.c_str());
+    fprintf(stderr, "  -mw FILE, --model-whisper [%-7s] whisper model file\n",                          params.model_wsp.c_str());
+    fprintf(stderr, "  -ml FILE, --model-llama   [%-7s] llama model file\n",                            params.model_llama.c_str());
+    fprintf(stderr, "  -s FILE,  --speak TEXT    [%-7s] command for TTS\n",                             params.speak.c_str());
+    fprintf(stderr, "  --prompt-file FNAME       [%-7s] file with custom prompt to start dialog\n",     "");
+    fprintf(stderr, "  --session FNAME       file to cache model state in (may be large!) (default: none)\n");
+    fprintf(stderr, "  --verbose-prompt          [%-7s] print prompt at start\n",                       params.verbose_prompt ? "true" : "false");
+    fprintf(stderr, "  -f FNAME, --file FNAME    [%-7s] text output file name\n",                       params.fname_out.c_str());
     fprintf(stderr, "\n");
 }
 
@@ -256,19 +252,13 @@ int main(int argc, char ** argv) {
 
     // whisper init
 
-    struct whisper_context_params cparams;
-    cparams.use_gpu = params.use_gpu;
-
-    struct whisper_context * ctx_wsp = whisper_init_from_file_with_params(params.model_wsp.c_str(), cparams);
+    struct whisper_context * ctx_wsp = whisper_init_from_file(params.model_wsp.c_str());
 
     // llama init
 
     llama_backend_init(true);
 
     auto lmparams = llama_model_default_params();
-    if (!params.use_gpu) {
-        lmparams.n_gpu_layers = 0;
-    }
 
     struct llama_model * model_llama = llama_load_model_from_file(params.model_llama.c_str(), lmparams);
 

examples/talk.wasm/emscripten.cpp

@@ -271,7 +271,7 @@ EMSCRIPTEN_BINDINGS(talk) {
     emscripten::function("init", emscripten::optional_override([](const std::string & path_model) {
         for (size_t i = 0; i < g_contexts.size(); ++i) {
             if (g_contexts[i] == nullptr) {
-                g_contexts[i] = whisper_init_from_file_with_params(path_model.c_str(), whisper_context_default_params());
+                g_contexts[i] = whisper_init_from_file(path_model.c_str());
                 if (g_contexts[i] != nullptr) {
                     g_running = true;
                     if (g_worker.joinable()) {

examples/talk/talk.cpp

@@ -31,7 +31,6 @@ struct whisper_params {
     bool print_special = false;
     bool print_energy  = false;
     bool no_timestamps = true;
-    bool use_gpu       = true;
 
     std::string person    = "Santa";
     std::string language  = "en";
@@ -62,7 +61,6 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params) {
         else if (arg == "-tr"  || arg == "--translate")     { params.translate     = true; }
         else if (arg == "-ps"  || arg == "--print-special") { params.print_special = true; }
         else if (arg == "-pe"  || arg == "--print-energy")  { params.print_energy  = true; }
-        else if (arg == "-ng"  || arg == "--no-gpu")        { params.use_gpu       = false; }
         else if (arg == "-p"   || arg == "--person")        { params.person        = argv[++i]; }
         else if (arg == "-l"   || arg == "--language")      { params.language      = argv[++i]; }
         else if (arg == "-mw"  || arg == "--model-whisper") { params.model_wsp     = argv[++i]; }
@@ -96,7 +94,6 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
     fprintf(stderr, "  -tr,      --translate     [%-7s] translate from source language to english\n",   params.translate ? "true" : "false");
     fprintf(stderr, "  -ps,      --print-special [%-7s] print special tokens\n",                        params.print_special ? "true" : "false");
     fprintf(stderr, "  -pe,      --print-energy  [%-7s] print sound energy (for debugging)\n",          params.print_energy ? "true" : "false");
-    fprintf(stderr, "  -ng,      --no-gpu        [%-7s] disable GPU\n",                                 params.use_gpu ? "false" : "true");
     fprintf(stderr, "  -p NAME,  --person NAME   [%-7s] person name (for prompt selection)\n",          params.person.c_str());
     fprintf(stderr, "  -l LANG,  --language LANG [%-7s] spoken language\n",                             params.language.c_str());
     fprintf(stderr, "  -mw FILE, --model-whisper [%-7s] whisper model file\n",                          params.model_wsp.c_str());
@@ -184,10 +181,8 @@ int main(int argc, char ** argv) {
     }
 
     // whisper init
-    struct whisper_context_params cparams;
-    cparams.use_gpu = params.use_gpu;
 
-    struct whisper_context * ctx_wsp = whisper_init_from_file_with_params(params.model_wsp.c_str(), cparams);
+    struct whisper_context * ctx_wsp = whisper_init_from_file(params.model_wsp.c_str());
 
     // gpt init
 

examples/twitch.sh

@@ -21,7 +21,7 @@ help()
     echo "Usage: ./twitch.sh -s [step] -m [model] -t [threads] [url]"
     echo "options:"
     echo "-s       Step in seconds (default is $step)."
-    echo "-m       Choose model, options are: 'tiny.en' 'tiny' 'base.en' 'base' 'small.en' 'small' 'medium.en' 'medium' 'large-v1' 'large-v2' 'large' (default is '$model')."
+    echo "-m       Choose model, options are: 'tiny.en' 'tiny' 'base.en' 'base' 'small.en' 'small' 'medium.en' 'medium' 'large-v1' 'large' (default is '$model')."
     echo "-t       Number of threads to use."
     echo "-h       Print this help page."
     echo

examples/whisper.android/app/build.gradle

@@ -18,7 +18,9 @@ android {
         vectorDrawables {
             useSupportLibrary true
         }
-
+        ndk {
+            abiFilters 'arm64-v8a', 'armeabi-v7a', 'x86', 'x86_64'
+        }
     }
 
     buildTypes {
@@ -41,10 +43,20 @@ android {
     composeOptions {
         kotlinCompilerExtensionVersion '1.5.0'
     }
+    ndkVersion "25.2.9519653"
+    externalNativeBuild {
+        cmake {
+            path = file("src/main/jni/whisper/CMakeLists.txt")
+        }
+    }
+    packagingOptions {
+        resources {
+            excludes += '/META-INF/{AL2.0,LGPL2.1}'
+        }
+    }
 }
 
 dependencies {
-    implementation project(':lib')
     implementation 'androidx.activity:activity-compose:1.7.2'
     implementation 'androidx.compose.material:material-icons-core:1.5.0'
     implementation 'androidx.compose.material3:material3:1.1.1'

examples/whisper.android/app/src/main/java/com/whispercppdemo/ui/main/MainScreenViewModel.kt

@@ -15,7 +15,7 @@ import androidx.lifecycle.viewmodel.initializer
 import androidx.lifecycle.viewmodel.viewModelFactory
 import com.whispercppdemo.media.decodeWaveFile
 import com.whispercppdemo.recorder.Recorder
-import com.whispercpp.whisper.WhisperContext
+import com.whispercppdemo.whisper.WhisperContext
 import kotlinx.coroutines.Dispatchers
 import kotlinx.coroutines.launch
 import kotlinx.coroutines.runBlocking
@@ -35,7 +35,7 @@ class MainScreenViewModel(private val application: Application) : ViewModel() {
     private val modelsPath = File(application.filesDir, "models")
     private val samplesPath = File(application.filesDir, "samples")
     private var recorder: Recorder = Recorder()
-    private var whisperContext: com.whispercpp.whisper.WhisperContext? = null
+    private var whisperContext: WhisperContext? = null
     private var mediaPlayer: MediaPlayer? = null
     private var recordedFile: File? = null
 
@@ -47,7 +47,7 @@ class MainScreenViewModel(private val application: Application) : ViewModel() {
     }
 
     private suspend fun printSystemInfo() {
-        printMessage(String.format("System Info: %s\n", com.whispercpp.whisper.WhisperContext.getSystemInfo()))
+        printMessage(String.format("System Info: %s\n", WhisperContext.getSystemInfo()))
     }
 
     private suspend fun loadData() {
@@ -78,7 +78,7 @@ class MainScreenViewModel(private val application: Application) : ViewModel() {
         printMessage("Loading model...\n")
         val models = application.assets.list("models/")
         if (models != null) {
-            whisperContext = com.whispercpp.whisper.WhisperContext.createContextFromAsset(application.assets, "models/" + models[0])
+            whisperContext = WhisperContext.createContextFromAsset(application.assets, "models/" + models[0])
             printMessage("Loaded model ${models[0]}.\n")
         }
 

examples/whisper.android/app/src/main/java/com/whispercppdemo/whisper/LibWhisper.kt

@@ -1,4 +1,4 @@
-package com.whispercpp.whisper
+package com.whispercppdemo.whisper
 
 import android.content.res.AssetManager
 import android.os.Build

examples/whisper.android/app/src/main/java/com/whispercppdemo/whisper/WhisperCpuConfig.kt

@@ -1,4 +1,4 @@
-package com.whispercpp.whisper
+package com.whispercppdemo.whisper
 
 import android.util.Log
 import java.io.BufferedReader

examples/whisper.android/app/src/main/jni/whisper/jni.c

@@ -127,11 +127,11 @@ static struct whisper_context *whisper_init_from_asset(
             .close = &asset_close
     };
 
-    return whisper_init_with_params(&loader, whisper_context_default_params());
+    return whisper_init(&loader);
 }
 
 JNIEXPORT jlong JNICALL
-Java_com_whispercpp_whisper_WhisperLib_00024Companion_initContextFromAsset(
+Java_com_whispercppdemo_whisper_WhisperLib_00024Companion_initContextFromAsset(
         JNIEnv *env, jobject thiz, jobject assetManager, jstring asset_path_str) {
     UNUSED(thiz);
     struct whisper_context *context = NULL;
@@ -142,18 +142,18 @@ Java_com_whispercpp_whisper_WhisperLib_00024Companion_initContextFromAsset(
 }
 
 JNIEXPORT jlong JNICALL
-Java_com_whispercpp_whisper_WhisperLib_00024Companion_initContext(
+Java_com_whispercppdemo_whisper_WhisperLib_00024Companion_initContext(
         JNIEnv *env, jobject thiz, jstring model_path_str) {
     UNUSED(thiz);
     struct whisper_context *context = NULL;
     const char *model_path_chars = (*env)->GetStringUTFChars(env, model_path_str, NULL);
-    context = whisper_init_from_file_with_params(model_path_chars, whisper_context_default_params());
+    context = whisper_init_from_file(model_path_chars);
     (*env)->ReleaseStringUTFChars(env, model_path_str, model_path_chars);
     return (jlong) context;
 }
 
 JNIEXPORT void JNICALL
-Java_com_whispercpp_whisper_WhisperLib_00024Companion_freeContext(
+Java_com_whispercppdemo_whisper_WhisperLib_00024Companion_freeContext(
         JNIEnv *env, jobject thiz, jlong context_ptr) {
     UNUSED(env);
     UNUSED(thiz);
@@ -162,7 +162,7 @@ Java_com_whispercpp_whisper_WhisperLib_00024Companion_freeContext(
 }
 
 JNIEXPORT void JNICALL
-Java_com_whispercpp_whisper_WhisperLib_00024Companion_fullTranscribe(
+Java_com_whispercppdemo_whisper_WhisperLib_00024Companion_fullTranscribe(
         JNIEnv *env, jobject thiz, jlong context_ptr, jint num_threads, jfloatArray audio_data) {
     UNUSED(thiz);
     struct whisper_context *context = (struct whisper_context *) context_ptr;
@@ -194,7 +194,7 @@ Java_com_whispercpp_whisper_WhisperLib_00024Companion_fullTranscribe(
 }
 
 JNIEXPORT jint JNICALL
-Java_com_whispercpp_whisper_WhisperLib_00024Companion_getTextSegmentCount(
+Java_com_whispercppdemo_whisper_WhisperLib_00024Companion_getTextSegmentCount(
         JNIEnv *env, jobject thiz, jlong context_ptr) {
     UNUSED(env);
     UNUSED(thiz);
@@ -203,7 +203,7 @@ Java_com_whispercpp_whisper_WhisperLib_00024Companion_getTextSegmentCount(
 }
 
 JNIEXPORT jstring JNICALL
-Java_com_whispercpp_whisper_WhisperLib_00024Companion_getTextSegment(
+Java_com_whispercppdemo_whisper_WhisperLib_00024Companion_getTextSegment(
         JNIEnv *env, jobject thiz, jlong context_ptr, jint index) {
     UNUSED(thiz);
     struct whisper_context *context = (struct whisper_context *) context_ptr;
@@ -213,7 +213,7 @@ Java_com_whispercpp_whisper_WhisperLib_00024Companion_getTextSegment(
 }
 
 JNIEXPORT jstring JNICALL
-Java_com_whispercpp_whisper_WhisperLib_00024Companion_getSystemInfo(
+Java_com_whispercppdemo_whisper_WhisperLib_00024Companion_getSystemInfo(
         JNIEnv *env, jobject thiz
 ) {
     UNUSED(thiz);
@@ -223,7 +223,7 @@ Java_com_whispercpp_whisper_WhisperLib_00024Companion_getSystemInfo(
 }
 
 JNIEXPORT jstring JNICALL
-Java_com_whispercpp_whisper_WhisperLib_00024Companion_benchMemcpy(JNIEnv *env, jobject thiz,
+Java_com_whispercppdemo_whisper_WhisperLib_00024Companion_benchMemcpy(JNIEnv *env, jobject thiz,
                                                                       jint n_threads) {
     UNUSED(thiz);
     const char *bench_ggml_memcpy = whisper_bench_memcpy_str(n_threads);
@@ -231,7 +231,7 @@ Java_com_whispercpp_whisper_WhisperLib_00024Companion_benchMemcpy(JNIEnv *env, j
 }
 
 JNIEXPORT jstring JNICALL
-Java_com_whispercpp_whisper_WhisperLib_00024Companion_benchGgmlMulMat(JNIEnv *env, jobject thiz,
+Java_com_whispercppdemo_whisper_WhisperLib_00024Companion_benchGgmlMulMat(JNIEnv *env, jobject thiz,
                                                                           jint n_threads) {
     UNUSED(thiz);
     const char *bench_ggml_mul_mat = whisper_bench_ggml_mul_mat_str(n_threads);

examples/whisper.android/lib/.gitignore

@@ -1 +0,0 @@
-/build

examples/whisper.android/lib/build.gradle

@@ -1,51 +0,0 @@
-plugins {
-    id 'com.android.library'
-    id 'org.jetbrains.kotlin.android'
-}
-
-android {
-    namespace 'com.whispercpp'
-    compileSdk 34
-
-    defaultConfig {
-        minSdk 26
-        targetSdk 34
-        versionCode 1
-        versionName "1.0"
-
-        ndk {
-            abiFilters 'arm64-v8a', 'armeabi-v7a', 'x86', 'x86_64'
-        }
-    }
-
-    buildTypes {
-        release {
-            minifyEnabled false
-        }
-    }
-    compileOptions {
-        sourceCompatibility JavaVersion.VERSION_1_8
-        targetCompatibility JavaVersion.VERSION_1_8
-    }
-    kotlinOptions {
-        jvmTarget = '1.8'
-    }
-
-    ndkVersion "25.2.9519653"
-    externalNativeBuild {
-        cmake {
-            path = file("src/main/jni/whisper/CMakeLists.txt")
-        }
-    }
-    packagingOptions {
-        resources {
-            excludes += '/META-INF/{AL2.0,LGPL2.1}'
-        }
-    }
-}
-
-dependencies {
-    implementation 'androidx.core:core-ktx:1.9.0'
-    implementation 'androidx.appcompat:appcompat:1.6.1'
-    implementation 'com.google.android.material:material:1.8.0'
-}

examples/whisper.android/lib/src/main/AndroidManifest.xml

@@ -1,4 +0,0 @@
-<?xml version="1.0" encoding="utf-8"?>
-<manifest xmlns:android="http://schemas.android.com/apk/res/android">
-
-</manifest>

examples/whisper.android/settings.gradle

@@ -14,4 +14,3 @@ dependencyResolutionManagement {
 }
 rootProject.name = "WhisperCppDemo"
 include ':app'
-include ':lib'

examples/whisper.objc/whisper.objc.xcodeproj/project.pbxproj

@@ -17,8 +17,8 @@
 		18627C8629052BE000BD2A04 /* Assets.xcassets in Resources */ = {isa = PBXBuildFile; fileRef = 18627C8529052BE000BD2A04 /* Assets.xcassets */; };
 		18627C8929052BE000BD2A04 /* LaunchScreen.storyboard in Resources */ = {isa = PBXBuildFile; fileRef = 18627C8729052BE000BD2A04 /* LaunchScreen.storyboard */; };
 		18627C8C29052BE000BD2A04 /* main.m in Sources */ = {isa = PBXBuildFile; fileRef = 18627C8B29052BE000BD2A04 /* main.m */; };
-		18627C9429052C4900BD2A04 /* whisper.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 18627C9329052C4900BD2A04 /* whisper.cpp */; settings = {COMPILER_FLAGS = "-DWHISPER_USE_COREML -DWHISPER_COREML_ALLOW_FALLBACK -DGGML_USE_METAL"; }; };
-		18627C9629052C5800BD2A04 /* ggml.c in Sources */ = {isa = PBXBuildFile; fileRef = 18627C9529052C5800BD2A04 /* ggml.c */; settings = {COMPILER_FLAGS = "-DGGML_USE_ACCELERATE -DGGML_USE_METAL"; }; };
+		18627C9429052C4900BD2A04 /* whisper.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 18627C9329052C4900BD2A04 /* whisper.cpp */; settings = {COMPILER_FLAGS = "-DWHISPER_USE_COREML"; }; };
+		18627C9629052C5800BD2A04 /* ggml.c in Sources */ = {isa = PBXBuildFile; fileRef = 18627C9529052C5800BD2A04 /* ggml.c */; settings = {COMPILER_FLAGS = "-DGGML_USE_ACCELERATE"; }; };
 		18627C9B29052CFF00BD2A04 /* ggml-base.en.bin in Resources */ = {isa = PBXBuildFile; fileRef = 18627C9A29052CFF00BD2A04 /* ggml-base.en.bin */; };
 		18ABE15A2AF556340044A204 /* ggml-backend.c in Sources */ = {isa = PBXBuildFile; fileRef = 18ABE1572AF556340044A204 /* ggml-backend.c */; };
 		18ABE15B2AF556340044A204 /* ggml-quants.c in Sources */ = {isa = PBXBuildFile; fileRef = 18ABE1592AF556340044A204 /* ggml-quants.c */; };

examples/whisper.objc/whisper.objc/ViewController.m

@@ -61,13 +61,7 @@ void AudioInputCallback(void * inUserData,
         NSLog(@"Loading model from %@", modelPath);
 
         // create ggml context
-
-        struct whisper_context_params cparams = whisper_context_default_params();
-#if TARGET_OS_SIMULATOR
-        cparams.use_gpu = false;
-        NSLog(@"Running on simulator, using CPU");
-#endif
-        stateInp.ctx = whisper_init_from_file_with_params([modelPath UTF8String], cparams);
+        stateInp.ctx = whisper_init_from_file([modelPath UTF8String]);
 
         // check if the model was loaded successfully
         if (stateInp.ctx == NULL) {

examples/whisper.swiftui/whisper.cpp.swift/LibWhisper.swift

@@ -1,5 +1,4 @@
 import Foundation
-import whisper
 
 enum WhisperError: Error {
     case couldNotInitializeContext
@@ -56,12 +55,7 @@ actor WhisperContext {
     }
     
     static func createContext(path: String) throws -> WhisperContext {
-        var params = whisper_context_default_params()
-#if targetEnvironment(simulator)
-        params.use_gpu = false
-        print("Running on the simulator, using CPU")
-#endif
-        let context = whisper_init_from_file_with_params(path, params)
+        let context = whisper_init_from_file(path)
         if let context {
             return WhisperContext(context: context)
         } else {

examples/whisper.swiftui/whisper.cpp.swift/WhisperCppDemo-Bridging-Header.h

@@ -0,0 +1,4 @@
+//
+//  Use this file to import your target's public headers that you would like to expose to Swift.
+//
+#import "whisper.h"

examples/whisper.swiftui/whisper.swiftui.xcodeproj/project.pbxproj

@@ -15,9 +15,14 @@
 		0AAC5D9B29539CCF003032C3 /* WhisperCppDemoApp.swift in Sources */ = {isa = PBXBuildFile; fileRef = 0AAC5D9A29539CCF003032C3 /* WhisperCppDemoApp.swift */; };
 		0AAC5D9D29539CCF003032C3 /* ContentView.swift in Sources */ = {isa = PBXBuildFile; fileRef = 0AAC5D9C29539CCF003032C3 /* ContentView.swift */; };
 		0AAC5D9F29539CD0003032C3 /* Assets.xcassets in Resources */ = {isa = PBXBuildFile; fileRef = 0AAC5D9E29539CD0003032C3 /* Assets.xcassets */; };
+		0AAC5DA329539CD0003032C3 /* Preview Assets.xcassets in Resources */ = {isa = PBXBuildFile; fileRef = 0AAC5DA229539CD0003032C3 /* Preview Assets.xcassets */; };
+		0AAC5DCB29539EB1003032C3 /* whisper.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 0AAC5DC729539EB0003032C3 /* whisper.cpp */; settings = {COMPILER_FLAGS = "-Wno-shorten-64-to-32"; }; };
+		0AAC5DCC29539EB1003032C3 /* ggml.c in Sources */ = {isa = PBXBuildFile; fileRef = 0AAC5DC929539EB0003032C3 /* ggml.c */; settings = {COMPILER_FLAGS = "-DGGML_USE_ACCELERATE -Wno-shorten-64-to-32"; }; };
 		0AAC5DCE2953A05C003032C3 /* WhisperState.swift in Sources */ = {isa = PBXBuildFile; fileRef = 0AAC5DCD2953A05C003032C3 /* WhisperState.swift */; };
 		0AAC5DD12953A394003032C3 /* LibWhisper.swift in Sources */ = {isa = PBXBuildFile; fileRef = 0AAC5DD02953A394003032C3 /* LibWhisper.swift */; };
-		E3F92DC52AFA8E3800A6A9D4 /* whisper in Frameworks */ = {isa = PBXBuildFile; productRef = E3F92DC42AFA8E3800A6A9D4 /* whisper */; };
+		18ABE1522AF555FA0044A204 /* ggml-backend.c in Sources */ = {isa = PBXBuildFile; fileRef = 18ABE14C2AF555FA0044A204 /* ggml-backend.c */; };
+		18ABE1532AF555FA0044A204 /* ggml-quants.c in Sources */ = {isa = PBXBuildFile; fileRef = 18ABE1512AF555FA0044A204 /* ggml-quants.c */; };
+		18AED4812AB21F2B009D854F /* ggml-alloc.c in Sources */ = {isa = PBXBuildFile; fileRef = 18AED47F2AB21F2B009D854F /* ggml-alloc.c */; };
 /* End PBXBuildFile section */
 
 /* Begin PBXFileReference section */
@@ -31,9 +36,22 @@
 		0AAC5D9C29539CCF003032C3 /* ContentView.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = ContentView.swift; sourceTree = "<group>"; };
 		0AAC5D9E29539CD0003032C3 /* Assets.xcassets */ = {isa = PBXFileReference; lastKnownFileType = folder.assetcatalog; path = Assets.xcassets; sourceTree = "<group>"; };
 		0AAC5DA029539CD0003032C3 /* WhisperCppDemo.entitlements */ = {isa = PBXFileReference; lastKnownFileType = text.plist.entitlements; path = WhisperCppDemo.entitlements; sourceTree = "<group>"; };
+		0AAC5DA229539CD0003032C3 /* Preview Assets.xcassets */ = {isa = PBXFileReference; lastKnownFileType = folder.assetcatalog; path = "Preview Assets.xcassets"; sourceTree = "<group>"; };
+		0AAC5DC629539EAF003032C3 /* WhisperCppDemo-Bridging-Header.h */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.c.h; path = "WhisperCppDemo-Bridging-Header.h"; sourceTree = "<group>"; };
+		0AAC5DC729539EB0003032C3 /* whisper.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; path = whisper.cpp; sourceTree = "<group>"; };
+		0AAC5DC829539EB0003032C3 /* whisper.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = whisper.h; sourceTree = "<group>"; };
+		0AAC5DC929539EB0003032C3 /* ggml.c */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.c; path = ggml.c; sourceTree = "<group>"; };
+		0AAC5DCA29539EB0003032C3 /* ggml.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = ggml.h; sourceTree = "<group>"; };
 		0AAC5DCD2953A05C003032C3 /* WhisperState.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = WhisperState.swift; sourceTree = "<group>"; };
 		0AAC5DD02953A394003032C3 /* LibWhisper.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = LibWhisper.swift; sourceTree = "<group>"; };
-		E3F92DC22AFA8DD800A6A9D4 /* whisper.cpp */ = {isa = PBXFileReference; lastKnownFileType = wrapper; name = whisper.cpp; path = ../..; sourceTree = "<group>"; };
+		18ABE14C2AF555FA0044A204 /* ggml-backend.c */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.c; path = "ggml-backend.c"; sourceTree = "<group>"; };
+		18ABE14D2AF555FA0044A204 /* ggml-backend.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = "ggml-backend.h"; sourceTree = "<group>"; };
+		18ABE14E2AF555FA0044A204 /* ggml-backend-impl.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = "ggml-backend-impl.h"; sourceTree = "<group>"; };
+		18ABE14F2AF555FA0044A204 /* ggml-quants.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = "ggml-quants.h"; sourceTree = "<group>"; };
+		18ABE1502AF555FA0044A204 /* ggml-impl.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = "ggml-impl.h"; sourceTree = "<group>"; };
+		18ABE1512AF555FA0044A204 /* ggml-quants.c */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.c; path = "ggml-quants.c"; sourceTree = "<group>"; };
+		18AED47F2AB21F2B009D854F /* ggml-alloc.c */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.c; path = "ggml-alloc.c"; sourceTree = "<group>"; };
+		18AED4802AB21F2B009D854F /* ggml-alloc.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = "ggml-alloc.h"; sourceTree = "<group>"; };
 /* End PBXFileReference section */
 
 /* Begin PBXFrameworksBuildPhase section */
@@ -41,7 +59,6 @@
 			isa = PBXFrameworksBuildPhase;
 			buildActionMask = 2147483647;
 			files = (
-				E3F92DC52AFA8E3800A6A9D4 /* whisper in Frameworks */,
 			);
 			runOnlyForDeploymentPostprocessing = 0;
 		};
@@ -77,12 +94,11 @@
 		0AAC5D8E29539CCF003032C3 = {
 			isa = PBXGroup;
 			children = (
-				E3F92DC22AFA8DD800A6A9D4 /* whisper.cpp */,
 				0A8E48FF2954B3F100704C1B /* README.md */,
+				0AAC5DC529539E89003032C3 /* whisper.cpp */,
 				0AAC5DCF2953A36C003032C3 /* whisper.cpp.swift */,
 				0AAC5D9929539CCF003032C3 /* whisper.swiftui.demo */,
 				0AAC5D9829539CCF003032C3 /* Products */,
-				E3F92DC32AFA8E3800A6A9D4 /* Frameworks */,
 			);
 			sourceTree = "<group>";
 		};
@@ -107,9 +123,39 @@
 			path = whisper.swiftui.demo;
 			sourceTree = "<group>";
 		};
+		0AAC5DA129539CD0003032C3 /* Preview Content */ = {
+			isa = PBXGroup;
+			children = (
+				0AAC5DA229539CD0003032C3 /* Preview Assets.xcassets */,
+			);
+			name = "Preview Content";
+			path = "../Preview Content";
+			sourceTree = "<group>";
+		};
+		0AAC5DC529539E89003032C3 /* whisper.cpp */ = {
+			isa = PBXGroup;
+			children = (
+				18ABE14E2AF555FA0044A204 /* ggml-backend-impl.h */,
+				18ABE14C2AF555FA0044A204 /* ggml-backend.c */,
+				18ABE14D2AF555FA0044A204 /* ggml-backend.h */,
+				18ABE1502AF555FA0044A204 /* ggml-impl.h */,
+				18ABE1512AF555FA0044A204 /* ggml-quants.c */,
+				18ABE14F2AF555FA0044A204 /* ggml-quants.h */,
+				18AED47F2AB21F2B009D854F /* ggml-alloc.c */,
+				18AED4802AB21F2B009D854F /* ggml-alloc.h */,
+				0AAC5DC929539EB0003032C3 /* ggml.c */,
+				0AAC5DCA29539EB0003032C3 /* ggml.h */,
+				0AAC5DC729539EB0003032C3 /* whisper.cpp */,
+				0AAC5DC829539EB0003032C3 /* whisper.h */,
+			);
+			name = whisper.cpp;
+			path = ../..;
+			sourceTree = "<group>";
+		};
 		0AAC5DCF2953A36C003032C3 /* whisper.cpp.swift */ = {
 			isa = PBXGroup;
 			children = (
+				0AAC5DC629539EAF003032C3 /* WhisperCppDemo-Bridging-Header.h */,
 				0AAC5DD02953A394003032C3 /* LibWhisper.swift */,
 			);
 			path = whisper.cpp.swift;
@@ -128,17 +174,11 @@
 			children = (
 				0AAC5D9E29539CD0003032C3 /* Assets.xcassets */,
 				0AAC5DA029539CD0003032C3 /* WhisperCppDemo.entitlements */,
+				0AAC5DA129539CD0003032C3 /* Preview Content */,
 			);
 			path = "Supporting files";
 			sourceTree = "<group>";
 		};
-		E3F92DC32AFA8E3800A6A9D4 /* Frameworks */ = {
-			isa = PBXGroup;
-			children = (
-			);
-			name = Frameworks;
-			sourceTree = "<group>";
-		};
 /* End PBXGroup section */
 
 /* Begin PBXNativeTarget section */
@@ -155,9 +195,6 @@
 			dependencies = (
 			);
 			name = whisper.swiftui;
-			packageProductDependencies = (
-				E3F92DC42AFA8E3800A6A9D4 /* whisper */,
-			);
 			productName = WhisperCppDemo;
 			productReference = 0AAC5D9729539CCF003032C3 /* whisper.swiftui.app */;
 			productType = "com.apple.product-type.application";
@@ -202,6 +239,7 @@
 			buildActionMask = 2147483647;
 			files = (
 				0AA751482953AC2E001EE061 /* samples in Resources */,
+				0AAC5DA329539CD0003032C3 /* Preview Assets.xcassets in Resources */,
 				0A8E49002954B3F100704C1B /* README.md in Resources */,
 				0AA751492953AC2E001EE061 /* models in Resources */,
 				0AAC5D9F29539CD0003032C3 /* Assets.xcassets in Resources */,
@@ -217,10 +255,15 @@
 			files = (
 				0AAC5D9D29539CCF003032C3 /* ContentView.swift in Sources */,
 				0AAC5D9B29539CCF003032C3 /* WhisperCppDemoApp.swift in Sources */,
+				0AAC5DCC29539EB1003032C3 /* ggml.c in Sources */,
+				18ABE1532AF555FA0044A204 /* ggml-quants.c in Sources */,
 				0AAC5DCE2953A05C003032C3 /* WhisperState.swift in Sources */,
 				0AAC5DD12953A394003032C3 /* LibWhisper.swift in Sources */,
 				0AA7514C2953B569001EE061 /* RiffWaveUtils.swift in Sources */,
+				0AAC5DCB29539EB1003032C3 /* whisper.cpp in Sources */,
 				0AA7514E2953D958001EE061 /* Recorder.swift in Sources */,
+				18AED4812AB21F2B009D854F /* ggml-alloc.c in Sources */,
+				18ABE1522AF555FA0044A204 /* ggml-backend.c in Sources */,
 			);
 			runOnlyForDeploymentPostprocessing = 0;
 		};
@@ -348,7 +391,7 @@
 				CODE_SIGN_STYLE = Automatic;
 				CURRENT_PROJECT_VERSION = 1;
 				DEVELOPMENT_ASSET_PATHS = "\"whisper.swiftui.demo/Supporting files/Preview Content\"";
-				DEVELOPMENT_TEAM = "";
+				DEVELOPMENT_TEAM = P8JZH34X63;
 				ENABLE_HARDENED_RUNTIME = YES;
 				ENABLE_PREVIEWS = YES;
 				GENERATE_INFOPLIST_FILE = YES;
@@ -372,6 +415,7 @@
 				SDKROOT = auto;
 				SUPPORTED_PLATFORMS = "iphoneos iphonesimulator macosx";
 				SWIFT_EMIT_LOC_STRINGS = YES;
+				SWIFT_OBJC_BRIDGING_HEADER = "whisper.cpp.swift/WhisperCppDemo-Bridging-Header.h";
 				SWIFT_OPTIMIZATION_LEVEL = "-Onone";
 				SWIFT_VERSION = 5.0;
 				TARGETED_DEVICE_FAMILY = "1,2";
@@ -388,7 +432,7 @@
 				CODE_SIGN_STYLE = Automatic;
 				CURRENT_PROJECT_VERSION = 1;
 				DEVELOPMENT_ASSET_PATHS = "\"whisper.swiftui.demo/Supporting files/Preview Content\"";
-				DEVELOPMENT_TEAM = "";
+				DEVELOPMENT_TEAM = P8JZH34X63;
 				ENABLE_HARDENED_RUNTIME = YES;
 				ENABLE_PREVIEWS = YES;
 				GENERATE_INFOPLIST_FILE = YES;
@@ -417,6 +461,7 @@
 				SDKROOT = auto;
 				SUPPORTED_PLATFORMS = "iphoneos iphonesimulator macosx";
 				SWIFT_EMIT_LOC_STRINGS = YES;
+				SWIFT_OBJC_BRIDGING_HEADER = "whisper.cpp.swift/WhisperCppDemo-Bridging-Header.h";
 				SWIFT_VERSION = 5.0;
 				TARGETED_DEVICE_FAMILY = "1,2";
 			};
@@ -444,13 +489,6 @@
 			defaultConfigurationName = Release;
 		};
 /* End XCConfigurationList section */
-
-/* Begin XCSwiftPackageProductDependency section */
-		E3F92DC42AFA8E3800A6A9D4 /* whisper */ = {
-			isa = XCSwiftPackageProductDependency;
-			productName = whisper;
-		};
-/* End XCSwiftPackageProductDependency section */
 	};
 	rootObject = 0AAC5D8F29539CCF003032C3 /* Project object */;
 }

examples/whisper.wasm/emscripten.cpp

@@ -24,7 +24,7 @@ EMSCRIPTEN_BINDINGS(whisper) {
 
         for (size_t i = 0; i < g_contexts.size(); ++i) {
             if (g_contexts[i] == nullptr) {
-                g_contexts[i] = whisper_init_from_file_with_params(path_model.c_str(), whisper_context_default_params());
+                g_contexts[i] = whisper_init_from_file(path_model.c_str());
                 if (g_contexts[i] != nullptr) {
                     return i + 1;
                 } else {

extra/convert-all.sh

@@ -1,6 +1,6 @@
 #!/bin/bash
 
-models=( "tiny.en" "tiny" "base.en" "base" "small.en" "small" "medium.en" "medium" "large-v1" "large-v2" "large" )
+models=( "tiny.en" "tiny" "base.en" "base" "small.en" "small" "medium.en" "medium" "large-v1" "large" )
 
 for model in "${models[@]}"; do
     python3 models/convert-pt-to-ggml.py ~/.cache/whisper/$model.pt ../whisper models/

ggml-metal.m

@@ -1030,7 +1030,7 @@ void ggml_metal_graph_compute(
                             [encoder setBytes:&ne00 length:sizeof(ne00) atIndex:2];
                             [encoder setBytes:&ne01 length:sizeof(ne01) atIndex:3];
                             [encoder setBytes:&ne02 length:sizeof(ne02) atIndex:4];
-                            [encoder setThreadgroupMemoryLength:GGML_PAD(nth/32*sizeof(float), 16) atIndex:0];
+                            [encoder setThreadgroupMemoryLength:nth/32*sizeof(float) atIndex:0];
 
                             [encoder dispatchThreadgroups:MTLSizeMake(ne01*ne02*ne03, 1, 1) threadsPerThreadgroup:MTLSizeMake(nth, 1, 1)];
                         } break;
@@ -1342,7 +1342,7 @@ void ggml_metal_graph_compute(
                             [encoder setBytes:&ne00 length:sizeof( int64_t) atIndex:2];
                             [encoder setBytes:&nb01 length:sizeof(uint64_t) atIndex:3];
                             [encoder setBytes:&eps  length:sizeof(   float) atIndex:4];
-                            [encoder setThreadgroupMemoryLength:GGML_PAD(nth/32*sizeof(float), 16) atIndex:0];
+                            [encoder setThreadgroupMemoryLength:nth/32*sizeof(float) atIndex:0];
 
                             const int64_t nrows = ggml_nrows(src0);
 
@@ -1361,7 +1361,7 @@ void ggml_metal_graph_compute(
                             [encoder setBytes:&ne00    length:sizeof( int64_t) atIndex:2];
                             [encoder setBytes:&nb01    length:sizeof(uint64_t) atIndex:3];
                             [encoder setBytes:&eps     length:sizeof(   float) atIndex:4];
-                            [encoder setThreadgroupMemoryLength:GGML_PAD(nth*sizeof(float), 16) atIndex:0];
+                            [encoder setThreadgroupMemoryLength:nth*sizeof(float) atIndex:0];
 
                             const int64_t nrows = ggml_nrows(src0);
 

ggml.c

@@ -143,6 +143,12 @@ void ggml_print_backtrace(void) {
 }
 #endif
 
+#undef MIN
+#undef MAX
+
+#define MIN(a, b) ((a) < (b) ? (a) : (b))
+#define MAX(a, b) ((a) > (b) ? (a) : (b))
+
 /*#define GGML_PERF*/
 #define GGML_DEBUG 0
 #define GGML_GELU_FP16
@@ -271,12 +277,6 @@ inline static void * ggml_aligned_malloc(size_t size) {
 // floating point type used to accumulate sums
 typedef double ggml_float;
 
-#undef MIN
-#undef MAX
-
-#define MIN(a, b) ((a) < (b) ? (a) : (b))
-#define MAX(a, b) ((a) > (b) ? (a) : (b))
-
 //
 // global data
 //
@@ -16391,9 +16391,6 @@ struct ggml_compute_state_shared {
     // synchronization primitives
     atomic_int n_active; // num active threads
     atomic_int node_n;   // active graph node
-
-    bool (*abort_callback)(void * data); // abort ggml_graph_compute when true
-    void * abort_callback_data;
 };
 
 struct ggml_compute_state {
@@ -16744,6 +16741,10 @@ static thread_ret_t ggml_graph_compute_thread(void * data) {
 
                 node_n = atomic_load(&state->shared->node_n);
                 if (node_n != last) break;
+
+                if (cplan->abort_callback && cplan->abort_callback(cplan->abort_callback_data)) {
+                    break;
+                }
             };
         }
 
@@ -17037,8 +17038,6 @@ int ggml_graph_compute(struct ggml_cgraph * cgraph, struct ggml_cplan * cplan) {
         /*.n_threads               =*/ n_threads,
         /*.n_active                =*/ n_threads,
         /*.node_n                  =*/ -1,
-        /*.abort_callback          =*/ NULL,
-        /*.abort_callback_data     =*/ NULL,
     };
     struct ggml_compute_state * workers = alloca(sizeof(struct ggml_compute_state)*n_threads);
 

models/README.md

@@ -50,8 +50,7 @@ https://huggingface.co/ggerganov/whisper.cpp/tree/main
 | medium    | 1.5 GB | ~2.6 GB | `fd9727b6e1217c2f614f9b698455c4ffd82463b4` |
 | medium.en | 1.5 GB | ~2.6 GB | `8c30f0e44ce9560643ebd10bbe50cd20eafd3723` |
 | large-v1  | 2.9 GB | ~4.7 GB | `b1caaf735c4cc1429223d5a74f0f4d0b9b59a299` |
-| large-v2  | 2.9 GB | ~4.7 GB | `0f4c8e34f21cf1a914c59d8b3ce882345ad349d6` |
-| large     | 2.9 GB | ~4.7 GB | `ad82bf6a9043ceed055076d0fd39f5f186ff8062` |
+| large     | 2.9 GB | ~4.7 GB | `0f4c8e34f21cf1a914c59d8b3ce882345ad349d6` |
 
 ## Model files for testing purposes
 

models/convert-h5-to-coreml.py

@@ -78,14 +78,14 @@ def convert_hf_whisper(hf_model_name_or_path: str, whisper_state_path: str):
 # Ported from models/convert-whisper-to-coreml.py
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
-    parser.add_argument("--model-name", type=str, help="name of model to convert (e.g. tiny, tiny.en, base, base.en, small, small.en, medium, medium.en, large, large-v1, large-v2)", required=True)
+    parser.add_argument("--model-name", type=str, help="name of model to convert (e.g. tiny, tiny.en, base, base.en, small, small.en, medium, medium.en, large, large-v1)", required=True)
     parser.add_argument("--model-path", type=str, help="path to the model (e.g. if published on HuggingFace: Oblivion208/whisper-tiny-cantonese)", required=True)
     parser.add_argument("--encoder-only", type=bool, help="only convert encoder", default=False)
     parser.add_argument("--quantize",     type=bool, help="quantize weights to F16", default=False)
     parser.add_argument("--optimize-ane", type=bool, help="optimize for ANE execution (currently broken)", default=False)
     args = parser.parse_args()
 
-    if args.model_name not in ["tiny", "tiny.en", "base", "base.en", "small", "small.en", "medium", "medium.en", "large", "large-v1", "large-v2"]:
+    if args.model_name not in ["tiny", "tiny.en", "base", "base.en", "small", "small.en", "medium", "medium.en", "large", "large-v1"]:
         raise ValueError("Invalid model name")
 
     pt_target_path = f"models/hf-{args.model_name}.pt"

models/convert-pt-to-ggml.py

@@ -228,7 +228,7 @@ with np.load(dir_whisper / "whisper" / "assets" / "mel_filters.npz") as f:
 # for backwards compatibility, also check for older hf_transformers format tokenizer files
 # old format: dir_whisper/whisper/assets/[multilingual/gpt2]/vocab.json
 # new format: dir_whisper/whisper/assets/[multilingual/gpt2].tiktoken
-multilingual = hparams["n_vocab"] >= 51865
+multilingual = hparams["n_vocab"] == 51865
 tokenizer = dir_whisper / "whisper" / "assets" / (multilingual and "multilingual.tiktoken" or "gpt2.tiktoken")
 tokenizer_type = "tiktoken"
 if not tokenizer.is_file():

models/convert-whisper-to-coreml.py

@@ -194,7 +194,7 @@ class TextDecoderANE(TextDecoder):
         x = x.permute(0,2,3,1).squeeze(0)
 
         # ANE can only load tensors with dim size of at most 16,384 - whisper uses 51,864 (en) or 51,865 (multi-lang) tokens so we need to compute in chunks
-        if self.token_embedding.weight.shape[0] >= 51865:
+        if self.token_embedding.weight.shape[0] == 51865:
             # split in 11 chunks - 4715 each
             splits = self.token_embedding.weight.split(self.token_embedding.weight.shape[0]//11, dim=0)
             logits = torch.cat([torch.einsum('bid,jd->bij', x, split) for split in splits]).view(*x.shape[:2], -1)
@@ -296,13 +296,13 @@ def convert_decoder(hparams, model, quantize=False):
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
-    parser.add_argument("--model", type=str, help="model to convert (e.g. tiny, tiny.en, base, base.en, small, small.en, medium, medium.en, large, large-v1, large-v2)", required=True)
+    parser.add_argument("--model", type=str, help="model to convert (e.g. tiny, tiny.en, base, base.en, small, small.en, medium, medium.en, large, large-v1)", required=True)
     parser.add_argument("--encoder-only", type=bool, help="only convert encoder", default=False)
     parser.add_argument("--quantize",     type=bool, help="quantize weights to F16", default=False)
     parser.add_argument("--optimize-ane", type=bool, help="optimize for ANE execution (currently broken)", default=False)
     args = parser.parse_args()
 
-    if args.model not in ["tiny", "tiny.en", "base", "base.en", "small", "small.en", "medium", "medium.en", "large", "large-v1", "large-v2"]:
+    if args.model not in ["tiny", "tiny.en", "base", "base.en", "small", "small.en", "medium", "medium.en", "large", "large-v1"]:
         raise ValueError("Invalid model name")
 
     whisper = load_model(args.model).cpu()

models/convert-whisper-to-openvino.py

@@ -38,10 +38,10 @@ def convert_encoder(hparams, encoder, mname):
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
-    parser.add_argument("--model", type=str, help="model to convert (e.g. tiny, tiny.en, base, base.en, small, small.en, medium, medium.en, large, large-v1, large-v2)", required=True)
+    parser.add_argument("--model", type=str, help="model to convert (e.g. tiny, tiny.en, base, base.en, small, small.en, medium, medium.en, large, large-v1)", required=True)
     args = parser.parse_args()
 
-    if args.model not in ["tiny", "tiny.en", "base", "base.en", "small", "small.en", "medium", "medium.en", "large", "large-v1", "large-v2"]:
+    if args.model not in ["tiny", "tiny.en", "base", "base.en", "small", "small.en", "medium", "medium.en", "large", "large-v1"]:
         raise ValueError("Invalid model name")
 
     whisper = load_model(args.model).cpu()

models/download-coreml-model.sh

@@ -19,7 +19,7 @@ function get_script_path() {
 models_path="$(get_script_path)"
 
 # Whisper models
-models=( "tiny.en" "tiny" "base.en" "base" "small.en" "small" "medium.en" "medium" "large-v1" "large-v2" "large" )
+models=( "tiny.en" "tiny" "base.en" "base" "small.en" "small" "medium.en" "medium" "large-v1" "large" )
 
 # list available models
 function list_models {

models/download-ggml-model.cmd

@@ -8,7 +8,7 @@ popd
 set argc=0
 for %%x in (%*) do set /A argc+=1
 
-set models=tiny.en tiny base.en base small.en small medium.en medium large-v1 large-v2 large
+set models=tiny.en tiny base.en base small.en small medium.en medium large-v1 large
 
 if %argc% neq 1 (
   echo.
@@ -57,8 +57,8 @@ goto :eof
 :list_models
   echo.
   echo Available models:
-  (for %%a in (%models%) do (
-    echo %%a
+  (for %%a in (%models%) do ( 
+    echo %%a 
   ))
   echo.
   exit /b

models/download-ggml-model.sh

@@ -41,7 +41,6 @@ models=(
     "medium-q5_0"
     "medium.en-q5_0"
     "large-v1"
-    "large-v2"
     "large"
     "large-q5_0"
 )

spm-headers/ggml.h

@@ -1 +0,0 @@
-../ggml.h

spm-headers/whisper.h

@@ -1 +0,0 @@
-../whisper.h

tests/run-tests.sh

@@ -19,7 +19,7 @@
 cd `dirname $0`
 
 # Whisper models
-models=( "tiny.en" "tiny" "base.en" "base" "small.en" "small" "medium.en" "medium" "large-v1" "large-v2" "large" )
+models=( "tiny.en" "tiny" "base.en" "base" "small.en" "small" "medium.en" "medium" "large-v1" "large" )
 
 # list available models
 function list_models {

whisper.h

@@ -1,20 +1,10 @@
 #ifndef WHISPER_H
 #define WHISPER_H
 
-#include "ggml.h"
-
 #include <stddef.h>
 #include <stdint.h>
 #include <stdbool.h>
 
-#ifdef __GNUC__
-#    define WHISPER_DEPRECATED(func, hint) func __attribute__((deprecated(hint)))
-#elif defined(_MSC_VER)
-#    define WHISPER_DEPRECATED(func, hint) __declspec(deprecated(hint)) func
-#else
-#    define WHISPER_DEPRECATED(func, hint) func
-#endif
-
 #ifdef WHISPER_SHARED
 #    ifdef _WIN32
 #        ifdef WHISPER_BUILD
@@ -31,6 +21,7 @@
 
 #define WHISPER_SAMPLE_RATE 16000
 #define WHISPER_N_FFT       400
+#define WHISPER_N_MEL       80
 #define WHISPER_HOP_LENGTH  160
 #define WHISPER_CHUNK_SIZE  30
 
@@ -80,10 +71,6 @@ extern "C" {
 
     typedef int whisper_token;
 
-    struct whisper_context_params {
-        bool  use_gpu;
-    };
-
     typedef struct whisper_token_data {
         whisper_token id;  // token id
         whisper_token tid; // forced timestamp token id
@@ -112,40 +99,15 @@ extern "C" {
     // Various functions for loading a ggml whisper model.
     // Allocate (almost) all memory needed for the model.
     // Return NULL on failure
-    WHISPER_API struct whisper_context * whisper_init_from_file_with_params  (const char * path_model,              struct whisper_context_params params);
-    WHISPER_API struct whisper_context * whisper_init_from_buffer_with_params(void * buffer, size_t buffer_size,    struct whisper_context_params params);
-    WHISPER_API struct whisper_context * whisper_init_with_params            (struct whisper_model_loader * loader, struct whisper_context_params params);
+    WHISPER_API struct whisper_context * whisper_init_from_file(const char * path_model);
+    WHISPER_API struct whisper_context * whisper_init_from_buffer(void * buffer, size_t buffer_size);
+    WHISPER_API struct whisper_context * whisper_init(struct whisper_model_loader * loader);
 
     // These are the same as the above, but the internal state of the context is not allocated automatically
     // It is the responsibility of the caller to allocate the state using whisper_init_state() (#523)
-    WHISPER_API struct whisper_context * whisper_init_from_file_with_params_no_state  (const char * path_model,              struct whisper_context_params params);
-    WHISPER_API struct whisper_context * whisper_init_from_buffer_with_params_no_state(void * buffer, size_t buffer_size,    struct whisper_context_params params);
-    WHISPER_API struct whisper_context * whisper_init_with_params_no_state            (struct whisper_model_loader * loader, struct whisper_context_params params);
-
-    WHISPER_DEPRECATED(
-        WHISPER_API struct whisper_context * whisper_init_from_file(const char * path_model),
-        "use whisper_init_from_file_with_params instead"
-    );
-    WHISPER_DEPRECATED(
-        WHISPER_API struct whisper_context * whisper_init_from_buffer(void * buffer, size_t buffer_size),
-        "use whisper_init_from_buffer_with_params instead"
-    );
-    WHISPER_DEPRECATED(
-        WHISPER_API struct whisper_context * whisper_init(struct whisper_model_loader * loader),
-        "use whisper_init_with_params instead"
-    );
-    WHISPER_DEPRECATED(
-        WHISPER_API struct whisper_context * whisper_init_from_file_no_state(const char * path_model),
-        "use whisper_init_from_file_with_params_no_state instead"
-    );
-    WHISPER_DEPRECATED(
-        WHISPER_API struct whisper_context * whisper_init_from_buffer_no_state(void * buffer, size_t buffer_size),
-        "use whisper_init_from_buffer_with_params_no_state instead"
-    );
-    WHISPER_DEPRECATED(
-        WHISPER_API struct whisper_context * whisper_init_no_state(struct whisper_model_loader * loader),
-        "use whisper_init_with_params_no_state instead"
-    );
+    WHISPER_API struct whisper_context * whisper_init_from_file_no_state(const char * path_model);
+    WHISPER_API struct whisper_context * whisper_init_from_buffer_no_state(void * buffer, size_t buffer_size);
+    WHISPER_API struct whisper_context * whisper_init_no_state(struct whisper_model_loader * loader);
 
     WHISPER_API struct whisper_state * whisper_init_state(struct whisper_context * ctx);
 
@@ -170,7 +132,6 @@ extern "C" {
     WHISPER_API void whisper_free      (struct whisper_context * ctx);
     WHISPER_API void whisper_free_state(struct whisper_state * state);
     WHISPER_API void whisper_free_params(struct whisper_full_params * params);
-    WHISPER_API void whisper_free_context_params(struct whisper_context_params * params);
 
     // Convert RAW PCM audio to log mel spectrogram.
     // The resulting spectrogram is stored inside the default state of the provided whisper context.
@@ -481,9 +442,7 @@ extern "C" {
         void * logits_filter_callback_user_data;
     };
 
-    // NOTE: this function allocates memory, and it is the responsibility of the caller to free the pointer - see whisper_free_context_params & whisper_free_params()
-    WHISPER_API struct whisper_context_params * whisper_context_default_params_by_ref();
-    WHISPER_API struct whisper_context_params whisper_context_default_params(void);
+    // NOTE: this function allocates memory, and it is the responsibility of the caller to free the pointer - see whisper_free_params()
     WHISPER_API struct whisper_full_params * whisper_full_default_params_by_ref(enum whisper_sampling_strategy strategy);
     WHISPER_API struct whisper_full_params whisper_full_default_params(enum whisper_sampling_strategy strategy);
 
@@ -572,7 +531,8 @@ extern "C" {
 
     // Control logging output; default behavior is to print to stderr
 
-    WHISPER_API void whisper_log_set(ggml_log_callback log_callback, void * user_data);
+    typedef void (*whisper_log_callback)(const char * line);
+    WHISPER_API void whisper_set_log_callback(whisper_log_callback callback);
 
 #ifdef __cplusplus
 }