whispercpp 1.3.3 → 1.3.4

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: f3fdcc54e3e57a80116d06cd42636f910885f6e463387950414af3181ce956f6
-  data.tar.gz: de739103554359ce18d07666399670602fd1cbb62a0991352d32e9d625a4678e
+  metadata.gz: d790b742ddf3f6071ba7cc41ea9a16d2daa1363d4fee9c984ad2609fb19ed9fc
+  data.tar.gz: 97b213ff5bc5cee1a2880d0e69d4ba9b89da309f003957a66f89ea7bdbfa1ee7
 SHA512:
-  metadata.gz: e1ffb0f1b2874cf5cc210df61eaa9f8b716639c05b6bd05cc3e94abea0fc46544b15b7ed2ee8a251014f56da89af029bfaa105f8d52ae93372e64018e7c93ebd
-  data.tar.gz: 591eae5d183c25a37911a233020f1aaa5826734b4259443bb4c9f5bc7d061ace06095504a5b7426f3c9a5e43eab29c5c630b6ad51d6b4dc3c315596fb1a37114
+  metadata.gz: e1ea6c6082532e90b26e17041bd0649f91e32fc70cb489bff0c9d83e4d92431e8d36a54d47904070f1d799e9df9d2fe7dc5f4cc668bf84c6339b7edf4c2b4d45
+  data.tar.gz: e2007190f93a5f6cb21d8014e07c650aa0e835286eb13131d268d02e563c519c23e1ccce60c90f7da5f921fb4ed1eaec143239627015fb6c40e02188e5eaf34c

data/ext/ruby_whisper_params.c

@@ -26,7 +26,7 @@
   rb_define_method(cParams, #param_name, ruby_whisper_params_get_ ## param_name, 0); \
   rb_define_method(cParams, #param_name "=", ruby_whisper_params_set_ ## param_name, 1);
 
-#define RUBY_WHISPER_PARAMS_PARAM_NAMES_COUNT 35
+#define RUBY_WHISPER_PARAMS_PARAM_NAMES_COUNT 36
 
 extern VALUE cParams;
 extern VALUE cVADParams;
@@ -49,6 +49,7 @@ static ID id_print_timestamps;
 static ID id_suppress_blank;
 static ID id_suppress_nst;
 static ID id_token_timestamps;
+static ID id_max_len;
 static ID id_split_on_word;
 static ID id_initial_prompt;
 static ID id_diarize;
@@ -514,6 +515,33 @@ ruby_whisper_params_set_token_timestamps(VALUE self, VALUE value)
 {
   BOOL_PARAMS_SETTER(self, token_timestamps, value)
 }
+
+/*
+ * max segment length in characters.
+ *
+ * call-seq:
+ *   max_len -> Integer
+ */
+static VALUE
+ruby_whisper_params_get_max_len(VALUE self)
+{
+  ruby_whisper_params *rwp;
+  TypedData_Get_Struct(self, ruby_whisper_params, &ruby_whisper_params_type, rwp);
+  return INT2NUM(rwp->params.max_len);
+}
+/*
+ * call-seq:
+ *   max_len = length -> length
+ */
+static VALUE
+ruby_whisper_params_set_max_len(VALUE self, VALUE value)
+{
+  ruby_whisper_params *rwp;
+  TypedData_Get_Struct(self, ruby_whisper_params, &ruby_whisper_params_type, rwp);
+  rwp->params.max_len = NUM2INT(value);
+  return value;
+}
+
 /*
  * If true, split on word rather than on token (when used with max_len).
  *
@@ -1137,6 +1165,7 @@ ruby_whisper_params_initialize(int argc, VALUE *argv, VALUE self)
       SET_PARAM_IF_SAME(suppress_blank)
       SET_PARAM_IF_SAME(suppress_nst)
       SET_PARAM_IF_SAME(token_timestamps)
+      SET_PARAM_IF_SAME(max_len)
       SET_PARAM_IF_SAME(split_on_word)
       SET_PARAM_IF_SAME(initial_prompt)
       SET_PARAM_IF_SAME(offset)
@@ -1271,30 +1300,31 @@ init_ruby_whisper_params(VALUE *mWhisper)
   DEFINE_PARAM(suppress_blank, 8)
   DEFINE_PARAM(suppress_nst, 9)
   DEFINE_PARAM(token_timestamps, 10)
-  DEFINE_PARAM(split_on_word, 11)
-  DEFINE_PARAM(initial_prompt, 12)
-  DEFINE_PARAM(diarize, 13)
-  DEFINE_PARAM(offset, 14)
-  DEFINE_PARAM(duration, 15)
-  DEFINE_PARAM(max_text_tokens, 16)
-  DEFINE_PARAM(temperature, 17)
-  DEFINE_PARAM(max_initial_ts, 18)
-  DEFINE_PARAM(length_penalty, 19)
-  DEFINE_PARAM(temperature_inc, 20)
-  DEFINE_PARAM(entropy_thold, 21)
-  DEFINE_PARAM(logprob_thold, 22)
-  DEFINE_PARAM(no_speech_thold, 23)
-  DEFINE_PARAM(new_segment_callback, 24)
-  DEFINE_PARAM(new_segment_callback_user_data, 25)
-  DEFINE_PARAM(progress_callback, 26)
-  DEFINE_PARAM(progress_callback_user_data, 27)
-  DEFINE_PARAM(encoder_begin_callback, 28)
-  DEFINE_PARAM(encoder_begin_callback_user_data, 29)
-  DEFINE_PARAM(abort_callback, 30)
-  DEFINE_PARAM(abort_callback_user_data, 31)
-  DEFINE_PARAM(vad, 32)
-  DEFINE_PARAM(vad_model_path, 33)
-  DEFINE_PARAM(vad_params, 34)
+  DEFINE_PARAM(max_len, 11)
+  DEFINE_PARAM(split_on_word, 12)
+  DEFINE_PARAM(initial_prompt, 13)
+  DEFINE_PARAM(diarize, 14)
+  DEFINE_PARAM(offset, 15)
+  DEFINE_PARAM(duration, 16)
+  DEFINE_PARAM(max_text_tokens, 17)
+  DEFINE_PARAM(temperature, 18)
+  DEFINE_PARAM(max_initial_ts, 19)
+  DEFINE_PARAM(length_penalty, 20)
+  DEFINE_PARAM(temperature_inc, 21)
+  DEFINE_PARAM(entropy_thold, 22)
+  DEFINE_PARAM(logprob_thold, 23)
+  DEFINE_PARAM(no_speech_thold, 24)
+  DEFINE_PARAM(new_segment_callback, 25)
+  DEFINE_PARAM(new_segment_callback_user_data, 26)
+  DEFINE_PARAM(progress_callback, 27)
+  DEFINE_PARAM(progress_callback_user_data, 28)
+  DEFINE_PARAM(encoder_begin_callback, 29)
+  DEFINE_PARAM(encoder_begin_callback_user_data, 30)
+  DEFINE_PARAM(abort_callback, 31)
+  DEFINE_PARAM(abort_callback_user_data, 32)
+  DEFINE_PARAM(vad, 33)
+  DEFINE_PARAM(vad_model_path, 34)
+  DEFINE_PARAM(vad_params, 35)
 
   rb_define_method(cParams, "on_new_segment", ruby_whisper_params_on_new_segment, 0);
   rb_define_method(cParams, "on_progress", ruby_whisper_params_on_progress, 0);

data/ext/sources/CMakeLists.txt

@@ -1,6 +1,6 @@
 cmake_minimum_required(VERSION 3.5) # for add_link_options and implicit target directories.
 project("whisper.cpp" C CXX)
-project("whisper.cpp" VERSION 1.7.6)
+project("whisper.cpp" VERSION 1.8.0)
 include(CheckIncludeFileCXX)
 
 set(SOVERSION 1)

data/ext/sources/bindings/javascript/package.json

@@ -1,6 +1,6 @@
 {
   "name": "whisper.cpp",
-  "version": "1.7.6",
+  "version": "1.8.0",
   "description": "Whisper speech recognition",
   "main": "whisper.js",
   "scripts": {

data/ext/sources/build-xcframework.sh

@@ -15,6 +15,7 @@ GGML_METAL_EMBED_LIBRARY=ON
 GGML_BLAS_DEFAULT=ON
 GGML_METAL_USE_BF16=ON
 GGML_OPENMP=OFF
+BUILD_STATIC_XCFRAMEWORK=${BUILD_STATIC_XCFRAMEWORK:-OFF}
 
 COMMON_C_FLAGS="-Wno-macro-redefined -Wno-shorten-64-to-32 -Wno-unused-command-line-argument -g"
 COMMON_CXX_FLAGS="-Wno-macro-redefined -Wno-shorten-64-to-32 -Wno-unused-command-line-argument -g"
@@ -327,6 +328,15 @@ combine_static_libraries() {
         arch_flags+=" -arch $arch"
     done
 
+
+    if [[ "${BUILD_STATIC_XCFRAMEWORK}" == "ON" ]]; then
+        echo "Packaging static framework for ${platform}."
+        mkdir -p "$(dirname "${base_dir}/${output_lib}")"
+        cp "${temp_dir}/combined.a" "${base_dir}/${output_lib}"
+        rm -rf "${temp_dir}"
+        return
+    fi
+
     # Create dynamic library
     echo "Creating dynamic library for ${platform}."
     xcrun -sdk $sdk clang++ -dynamiclib \
@@ -529,6 +539,20 @@ combine_static_libraries "build-tvos-device" "Release-appletvos" "tvos" "false"
 
 # Create XCFramework with correct debug symbols paths
 echo "Creating XCFramework..."
+
+if [[ "${BUILD_STATIC_XCFRAMEWORK}" == "ON" ]]; then
+    xcodebuild -create-xcframework \
+        -framework $(pwd)/build-ios-sim/framework/whisper.framework \
+        -framework $(pwd)/build-ios-device/framework/whisper.framework \
+        -framework $(pwd)/build-macos/framework/whisper.framework \
+        -framework $(pwd)/build-visionos/framework/whisper.framework \
+        -framework $(pwd)/build-visionos-sim/framework/whisper.framework \
+        -framework $(pwd)/build-tvos-device/framework/whisper.framework \
+        -framework $(pwd)/build-tvos-sim/framework/whisper.framework \
+        -output $(pwd)/build-apple/whisper.xcframework
+    exit 0
+fi
+
 xcodebuild -create-xcframework \
     -framework $(pwd)/build-ios-sim/framework/whisper.framework \
     -debug-symbols $(pwd)/build-ios-sim/dSYMs/whisper.dSYM \

data/ext/sources/examples/CMakeLists.txt

@@ -98,6 +98,7 @@ if (EMSCRIPTEN)
     add_subdirectory(stream.wasm)
     add_subdirectory(command.wasm)
     add_subdirectory(bench.wasm)
+    add_subdirectory(wchess)
 elseif(CMAKE_JS_VERSION)
     add_subdirectory(addon.node)
 else()

data/ext/sources/examples/addon.node/addon.cpp

@@ -207,7 +207,7 @@ class ProgressWorker : public Napi::AsyncWorker {
             auto callback = [progress](Napi::Env env, Napi::Function jsCallback) {
                 jsCallback.Call({Napi::Number::New(env, progress)});
             };
-            
+
             tsfn.BlockingCall(callback);
         }
     }
@@ -396,59 +396,59 @@ 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 = true;
   if (whisper_params.Has("use_gpu") && whisper_params.Get("use_gpu").IsBoolean()) {
     use_gpu = whisper_params.Get("use_gpu").As<Napi::Boolean>();
   }
-  
+
   bool flash_attn = false;
   if (whisper_params.Has("flash_attn") && whisper_params.Get("flash_attn").IsBoolean()) {
     flash_attn = whisper_params.Get("flash_attn").As<Napi::Boolean>();
   }
-  
+
   bool no_prints = false;
   if (whisper_params.Has("no_prints") && whisper_params.Get("no_prints").IsBoolean()) {
     no_prints = whisper_params.Get("no_prints").As<Napi::Boolean>();
   }
-  
+
   bool no_timestamps = false;
   if (whisper_params.Has("no_timestamps") && whisper_params.Get("no_timestamps").IsBoolean()) {
     no_timestamps = whisper_params.Get("no_timestamps").As<Napi::Boolean>();
   }
-  
+
   bool detect_language = false;
   if (whisper_params.Has("detect_language") && whisper_params.Get("detect_language").IsBoolean()) {
     detect_language = whisper_params.Get("detect_language").As<Napi::Boolean>();
   }
-  
+
   int32_t audio_ctx = 0;
   if (whisper_params.Has("audio_ctx") && whisper_params.Get("audio_ctx").IsNumber()) {
     audio_ctx = whisper_params.Get("audio_ctx").As<Napi::Number>();
   }
-  
+
   bool comma_in_time = true;
   if (whisper_params.Has("comma_in_time") && whisper_params.Get("comma_in_time").IsBoolean()) {
     comma_in_time = whisper_params.Get("comma_in_time").As<Napi::Boolean>();
   }
-  
+
   int32_t max_len = 0;
   if (whisper_params.Has("max_len") && whisper_params.Get("max_len").IsNumber()) {
     max_len = whisper_params.Get("max_len").As<Napi::Number>();
   }
-  
+
   // Add support for max_context
   int32_t max_context = -1;
   if (whisper_params.Has("max_context") && whisper_params.Get("max_context").IsNumber()) {
     max_context = whisper_params.Get("max_context").As<Napi::Number>();
   }
-  
+
   // support prompt
   std::string prompt = "";
   if (whisper_params.Has("prompt") && whisper_params.Get("prompt").IsString()) {
     prompt = whisper_params.Get("prompt").As<Napi::String>();
   }
-  
+
   // Add support for print_progress
   bool print_progress = false;
   if (whisper_params.Has("print_progress") && whisper_params.Get("print_progress").IsBoolean()) {
@@ -465,37 +465,37 @@ Napi::Value whisper(const Napi::CallbackInfo& info) {
   if (whisper_params.Has("vad") && whisper_params.Get("vad").IsBoolean()) {
     vad = whisper_params.Get("vad").As<Napi::Boolean>();
   }
-  
+
   std::string vad_model = "";
   if (whisper_params.Has("vad_model") && whisper_params.Get("vad_model").IsString()) {
     vad_model = whisper_params.Get("vad_model").As<Napi::String>();
   }
-  
+
   float vad_threshold = 0.5f;
   if (whisper_params.Has("vad_threshold") && whisper_params.Get("vad_threshold").IsNumber()) {
     vad_threshold = whisper_params.Get("vad_threshold").As<Napi::Number>();
   }
-  
+
   int vad_min_speech_duration_ms = 250;
   if (whisper_params.Has("vad_min_speech_duration_ms") && whisper_params.Get("vad_min_speech_duration_ms").IsNumber()) {
     vad_min_speech_duration_ms = whisper_params.Get("vad_min_speech_duration_ms").As<Napi::Number>();
   }
-  
+
   int vad_min_silence_duration_ms = 100;
   if (whisper_params.Has("vad_min_silence_duration_ms") && whisper_params.Get("vad_min_silence_duration_ms").IsNumber()) {
     vad_min_silence_duration_ms = whisper_params.Get("vad_min_silence_duration_ms").As<Napi::Number>();
   }
-  
+
   float vad_max_speech_duration_s = FLT_MAX;
   if (whisper_params.Has("vad_max_speech_duration_s") && whisper_params.Get("vad_max_speech_duration_s").IsNumber()) {
     vad_max_speech_duration_s = whisper_params.Get("vad_max_speech_duration_s").As<Napi::Number>();
   }
-  
+
   int vad_speech_pad_ms = 30;
   if (whisper_params.Has("vad_speech_pad_ms") && whisper_params.Get("vad_speech_pad_ms").IsNumber()) {
     vad_speech_pad_ms = whisper_params.Get("vad_speech_pad_ms").As<Napi::Number>();
   }
-  
+
   float vad_samples_overlap = 0.1f;
   if (whisper_params.Has("vad_samples_overlap") && whisper_params.Get("vad_samples_overlap").IsNumber()) {
     vad_samples_overlap = whisper_params.Get("vad_samples_overlap").As<Napi::Number>();

data/ext/sources/examples/addon.node/index.js

@@ -1,8 +1,10 @@
-const path = require("path");
-const { whisper } = require(path.join(
-  __dirname,
-  "../../build/Release/addon.node"
-));
+const path = require('path');
+const os = require('os');
+
+const isWindows = os.platform() === 'win32';
+const buildPath = isWindows ? "../../build/bin/Release/addon.node" : "../../build/Release/addon.node";
+
+const { whisper } = require(path.join(__dirname, buildPath));
 const { promisify } = require("util");
 
 const whisperAsync = promisify(whisper);

data/ext/sources/examples/bench/bench.cpp

@@ -13,7 +13,7 @@ struct whisper_params {
     std::string model = "models/ggml-base.en.bin";
 
     bool use_gpu    = true;
-    bool flash_attn = false;
+    bool flash_attn = true;
 };
 
 void whisper_print_usage(int argc, char ** argv, const whisper_params & params);
@@ -26,11 +26,12 @@ static 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 == "-fa" || arg == "--flash-attn") { params.flash_attn = true; }
+        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 == "-fa"    || arg == "--flash-attn")    { params.flash_attn = true; }
+        else if (arg == "-nfa"   || arg == "--no-flash-attn") { params.flash_attn = false; }
         else {
             fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
             whisper_print_usage(argc, argv, params);
@@ -46,15 +47,16 @@ 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, "  -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, "                           %-7s  0 - whisper\n",                                 "");
-    fprintf(stderr, "                           %-7s  1 - memcpy\n",                                  "");
-    fprintf(stderr, "                           %-7s  2 - ggml_mul_mat\n",                            "");
-    fprintf(stderr, "  -ng,      --no-gpu      [%-7s] disable GPU\n",                                 params.use_gpu ? "false" : "true");
-    fprintf(stderr, "  -fa,      --flash-attn  [%-7s] enable flash attention\n",                      params.flash_attn ? "true" : "false");
+    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, "  -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, "                             %-7s  0 - whisper\n",                                 "");
+    fprintf(stderr, "                             %-7s  1 - memcpy\n",                                  "");
+    fprintf(stderr, "                             %-7s  2 - ggml_mul_mat\n",                            "");
+    fprintf(stderr, "  -ng,      --no-gpu        [%-7s] disable GPU\n",                                 params.use_gpu ? "false" : "true");
+    fprintf(stderr, "  -fa,      --flash-attn    [%-7s] enable flash attention\n",                      params.flash_attn ? "true" : "false");
+    fprintf(stderr, "  -nfa,     --no-flash-attn [%-7s] disable flash attention\n",                     params.flash_attn ? "false" : "true");
     fprintf(stderr, "\n");
 }
 
@@ -99,7 +101,15 @@ static int whisper_bench_full(const whisper_params & params) {
     }
 
     // text-generation heat
-    if (int ret = whisper_decode(ctx, tokens, 1, 256, params.n_threads) != 0) {
+    for (int i = 0; i < 256; i++) {
+        if (int ret = whisper_decode(ctx, tokens, 1, i, params.n_threads) != 0) {
+            fprintf(stderr, "error: failed to decode: %d\n", ret);
+            return 4;
+        }
+    }
+
+    // batched heat
+    if (int ret = whisper_decode(ctx, tokens, 5, 0, params.n_threads) != 0) {
         fprintf(stderr, "error: failed to decode: %d\n", ret);
         return 4;
     }

data/ext/sources/examples/bench.wasm/index-tmpl.html

@@ -42,6 +42,7 @@
                 <a href="../bench.wasm/">bench</a> |
                 <a href="../stream.wasm">stream</a> |
                 <a href="../command.wasm/">command</a> |
+                <a href="../wchess.wasm/">wchess</a> |
 
             <br><br>
 
@@ -191,15 +192,15 @@
 
             function loadWhisper(model) {
                 let urls = {
-                    'tiny.en': 'https://whisper.ggerganov.com/ggml-model-whisper-tiny.en.bin',
-                    'base.en': 'https://whisper.ggerganov.com/ggml-model-whisper-base.en.bin',
-                    'small.en': 'https://whisper.ggerganov.com/ggml-model-whisper-small.en.bin',
-
-                    'tiny-en-q5_1':  'https://whisper.ggerganov.com/ggml-model-whisper-tiny.en-q5_1.bin',
-                    'base-en-q5_1':  'https://whisper.ggerganov.com/ggml-model-whisper-base.en-q5_1.bin',
-                    'small-en-q5_1': 'https://whisper.ggerganov.com/ggml-model-whisper-small.en-q5_1.bin',
-                    'medium-en-q5_0':'https://whisper.ggerganov.com/ggml-model-whisper-medium.en-q5_0.bin',
-                    'large-q5_0':    'https://whisper.ggerganov.com/ggml-model-whisper-large-q5_0.bin',
+                    'tiny.en': 'https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-tiny.en.bin',
+                    'base.en': 'https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.en.bin',
+                    'small.en': 'https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-small.en.bin',
+
+                    'tiny-en-q5_1':  'https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-tiny.en-q5_1.bin',
+                    'base-en-q5_1':  'https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.en-q5_1.bin',
+                    'small-en-q5_1': 'https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-small.en-q5_1.bin',
+                    'medium-en-q5_0':'https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-medium.en-q5_0.bin',
+                    'large-q5_0':    'https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-large-q5_0.bin',
                 };
 
                 let sizes = {

data/ext/sources/examples/cli/cli.cpp

@@ -75,7 +75,7 @@ struct whisper_params {
     bool no_timestamps   = false;
     bool log_score       = false;
     bool use_gpu         = true;
-    bool flash_attn      = false;
+    bool flash_attn      = true;
     bool suppress_nst    = false;
 
     std::string language  = "en";
@@ -193,6 +193,7 @@ static bool whisper_params_parse(int argc, char ** argv, whisper_params & params
         else if (arg == "-ls"   || arg == "--log-score")       { params.log_score       = true; }
         else if (arg == "-ng"   || arg == "--no-gpu")          { params.use_gpu         = false; }
         else if (arg == "-fa"   || arg == "--flash-attn")      { params.flash_attn      = true; }
+        else if (arg == "-nfa"  || arg == "--no-flash-attn")   { params.flash_attn      = false; }
         else if (arg == "-sns"  || arg == "--suppress-nst")    { params.suppress_nst    = true; }
         else if (                  arg == "--suppress-regex")  { params.suppress_regex  = ARGV_NEXT; }
         else if (                  arg == "--grammar")         { params.grammar         = ARGV_NEXT; }
@@ -271,7 +272,8 @@ static void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params
     fprintf(stderr, "  -dtw MODEL --dtw MODEL         [%-7s] compute token-level timestamps\n",                 params.dtw.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, "  -fa,       --flash-attn        [%-7s] flash attention\n",                                params.flash_attn ? "true" : "false");
+    fprintf(stderr, "  -fa,       --flash-attn        [%-7s] enable flash attention\n",                         params.flash_attn ? "true" : "false");
+    fprintf(stderr, "  -nfa,      --no-flash-attn     [%-7s] disable flash attention\n",                        params.flash_attn ? "false" : "true");
     fprintf(stderr, "  -sns,      --suppress-nst      [%-7s] suppress non-speech tokens\n",                     params.suppress_nst ? "true" : "false");
     fprintf(stderr, "  --suppress-regex REGEX         [%-7s] regular expression matching tokens to suppress\n", params.suppress_regex.c_str());
     fprintf(stderr, "  --grammar GRAMMAR              [%-7s] GBNF grammar to guide decoding\n",                 params.grammar.c_str());

data/ext/sources/examples/command/command.cpp

@@ -42,7 +42,7 @@ struct whisper_params {
     bool print_energy  = false;
     bool no_timestamps = true;
     bool use_gpu       = true;
-    bool flash_attn    = false;
+    bool flash_attn    = true;
 
     std::string language  = "en";
     std::string model     = "models/ggml-base.en.bin";
@@ -66,28 +66,29 @@ static 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 == "-pms" || arg == "--prompt-ms")     { params.prompt_ms     = std::stoi(argv[++i]); }
-        else if (arg == "-cms" || arg == "--command-ms")    { params.command_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 == "-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 == "-fa"  || arg == "--flash-attn")    { params.flash_attn    = true; }
-        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 == "-cmd" || arg == "--commands")      { params.commands      = argv[++i]; }
-        else if (arg == "-p"   || arg == "--prompt")        { params.prompt        = argv[++i]; }
-        else if (arg == "-ctx" || arg == "--context")       { params.context       = argv[++i]; }
-        else if (                 arg == "--grammar")       { params.grammar       = argv[++i]; }
-        else if (                 arg == "--grammar-penalty") { params.grammar_penalty = std::stof(argv[++i]); }
-        else if (                 arg == "--suppress-regex") { params.suppress_regex = argv[++i]; }
+        else if (arg == "-t"     || arg == "--threads")       { params.n_threads     = std::stoi(argv[++i]); }
+        else if (arg == "-pms"   || arg == "--prompt-ms")     { params.prompt_ms     = std::stoi(argv[++i]); }
+        else if (arg == "-cms"   || arg == "--command-ms")    { params.command_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 == "-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 == "-fa"    || arg == "--flash-attn")    { params.flash_attn    = true; }
+        else if (arg == "-nfa"   || arg == "--no-flash-attn") { params.flash_attn    = 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 == "-cmd"   || arg == "--commands")      { params.commands      = argv[++i]; }
+        else if (arg == "-p"     || arg == "--prompt")        { params.prompt        = argv[++i]; }
+        else if (arg == "-ctx"   || arg == "--context")       { params.context       = argv[++i]; }
+        else if (                   arg == "--grammar")       { params.grammar       = argv[++i]; }
+        else if (                   arg == "--grammar-penalty") { params.grammar_penalty = std::stof(argv[++i]); }
+        else if (                   arg == "--suppress-regex") { params.suppress_regex = argv[++i]; }
         else {
             fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
             whisper_print_usage(argc, argv, params);
@@ -116,7 +117,8 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
     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, "  -fa,        --flash-attn     [%-7s] flash attention\n",                             params.flash_attn ? "true" : "false");
+    fprintf(stderr, "  -fa,        --flash-attn     [%-7s] enbale flash attention\n",                      params.flash_attn ? "true" : "false");
+    fprintf(stderr, "  -nfa,       --no-flash-attn  [%-7s] disable flash attention\n",                     params.flash_attn ? "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());

data/ext/sources/examples/command.wasm/index-tmpl.html

@@ -42,6 +42,7 @@
                 <a href="../bench.wasm/">bench</a> |
                 <a href="../stream.wasm">stream</a> |
                 <a href="../command.wasm/">command</a> |
+                <a href="../wchess.wasm/">wchess</a> |
 
             <br><br>
 
@@ -174,11 +175,11 @@
 
             function loadWhisper(model) {
                 let urls = {
-                    'tiny.en': 'https://whisper.ggerganov.com/ggml-model-whisper-tiny.en.bin',
-                    'base.en': 'https://whisper.ggerganov.com/ggml-model-whisper-base.en.bin',
+                    'tiny.en': 'https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-tiny.en.bin',
+                    'base.en': 'https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.en.bin',
 
-                    'tiny-en-q5_1':  'https://whisper.ggerganov.com/ggml-model-whisper-tiny.en-q5_1.bin',
-                    'base-en-q5_1':  'https://whisper.ggerganov.com/ggml-model-whisper-base.en-q5_1.bin',
+                    'tiny-en-q5_1':  'https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-tiny.en-q5_1.bin',
+                    'base-en-q5_1':  'https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.en-q5_1.bin',
                 };
 
                 let sizes = {

data/ext/sources/examples/common-ggml.cpp

@@ -72,6 +72,7 @@ bool ggml_common_quantize_0(
         case GGML_FTYPE_MOSTLY_IQ4_XS:
         case GGML_FTYPE_MOSTLY_IQ1_M:
         case GGML_FTYPE_MOSTLY_BF16:
+        case GGML_FTYPE_MOSTLY_MXFP4:
                 {
                     fprintf(stderr, "%s: invalid model type %d\n", __func__, ftype);
                     return false;
@@ -211,6 +212,7 @@ bool ggml_common_quantize_0(
                 case GGML_TYPE_BF16:
                 case GGML_TYPE_TQ1_0:
                 case GGML_TYPE_TQ2_0:
+                case GGML_TYPE_MXFP4:
                 case GGML_TYPE_COUNT:
                     {
                         fprintf(stderr, "%s: unsupported quantization type %d (%s)\n", __func__, ttype, ggml_type_name((ggml_type) ttype));