whispercpp 1.3.3 → 1.3.5

data/ext/sources/examples/lsp/lsp.cpp

@@ -31,7 +31,7 @@ struct whisper_params {
     bool print_special = false;
     bool print_energy  = false;
     bool use_gpu       = true;
-    bool flash_attn    = false;
+    bool flash_attn    = true;
 
     std::string language  = "en";
     std::string model     = "models/ggml-base.en.bin";
@@ -62,21 +62,22 @@ 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 == "-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 {
             fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
             whisper_print_usage(argc, argv, params);
@@ -105,7 +106,8 @@ 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] 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, "  -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");

data/ext/sources/examples/quantize/CMakeLists.txt

@@ -1,6 +1,7 @@
-set(TARGET quantize)
+set(TARGET whisper-quantize)
 add_executable(${TARGET} quantize.cpp)
 
 include(DefaultTargetOptions)
 
 target_link_libraries(${TARGET} PRIVATE common whisper ${CMAKE_THREAD_LIBS_INIT})
+install(TARGETS ${TARGET} RUNTIME)

data/ext/sources/examples/server/server.cpp

@@ -60,6 +60,7 @@ struct server_params
     std::string public_path = "examples/server/public";
     std::string request_path = "";
     std::string inference_path = "/inference";
+    std::string tmp_dir = ".";
 
     int32_t port          = 8080;
     int32_t read_timeout  = 600;
@@ -101,9 +102,10 @@ struct whisper_params {
     bool print_progress  = false;
     bool no_timestamps   = false;
     bool use_gpu         = true;
-    bool flash_attn      = false;
+    bool flash_attn      = true;
     bool suppress_nst    = false;
-    bool no_context      = false;
+    bool no_context      = true;
+    bool no_language_probabilities = false;
 
     std::string language        = "en";
     std::string prompt          = "";
@@ -173,11 +175,13 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
     fprintf(stderr, "  --request-path PATH,           [%-7s] Request path for all requests\n", sparams.request_path.c_str());
     fprintf(stderr, "  --inference-path PATH,         [%-7s] Inference path for all requests\n", sparams.inference_path.c_str());
     fprintf(stderr, "  --convert,                     [%-7s] Convert audio to WAV, requires ffmpeg on the server\n", sparams.ffmpeg_converter ? "true" : "false");
+    fprintf(stderr, "  --tmp-dir,                     [%-7s] Temporary directory for ffmpeg transcoded files\n", sparams.tmp_dir.c_str());
     fprintf(stderr, "  -sns,      --suppress-nst      [%-7s] suppress non-speech tokens\n", params.suppress_nst ? "true" : "false");
     fprintf(stderr, "  -nth N,    --no-speech-thold N [%-7.2f] no speech threshold\n",   params.no_speech_thold);
-    fprintf(stderr, "  -nc,       --no-context        [%-7s] do not use previous audio context\n", params.no_context ? "true" : "false");
     fprintf(stderr, "  -ng,       --no-gpu            [%-7s] do not use 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, "  -nlp,      --no-language-probabilities [%-7s] exclude language probabilities from verbose_json output\n", params.no_language_probabilities ? "true" : "false");
     // Voice Activity Detection (VAD) parameters
     fprintf(stderr, "\nVoice Activity Detection (VAD) options:\n");
     fprintf(stderr, "             --vad                           [%-7s] enable Voice Activity Detection (VAD)\n",            params.vad ? "true" : "false");
@@ -234,9 +238,10 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params, serve
         else if (arg == "-dtw"  || arg == "--dtw")             { params.dtw             = 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 if (arg == "-sns"  || arg == "--suppress-nst")    { params.suppress_nst    = true; }
         else if (arg == "-nth"  || arg == "--no-speech-thold") { params.no_speech_thold = std::stof(argv[++i]); }
-        else if (arg == "-nc"   || arg == "--no-context")      { params.no_context      = true; }
+        else if (arg == "-nlp"  || arg == "--no-language-probabilities") { params.no_language_probabilities = true; }
 
         // server params
         else if (                  arg == "--port")            { sparams.port        = std::stoi(argv[++i]); }
@@ -245,13 +250,14 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params, serve
         else if (                  arg == "--request-path")    { sparams.request_path = argv[++i]; }
         else if (                  arg == "--inference-path")  { sparams.inference_path = argv[++i]; }
         else if (                  arg == "--convert")         { sparams.ffmpeg_converter     = true; }
+        else if (                  arg == "--tmp-dir")         { sparams.tmp_dir     = argv[++i]; }
 
         // Voice Activity Detection (VAD)
         else if (                  arg == "--vad")                         { params.vad                         = true; }
         else if (arg == "-vm"   || arg == "--vad-model")                   { params.vad_model                   = argv[++i]; }
         else if (arg == "-vt"   || arg == "--vad-threshold")               { params.vad_threshold               = std::stof(argv[++i]); }
         else if (arg == "-vspd" || arg == "--vad-min-speech-duration-ms")  { params.vad_min_speech_duration_ms  = std::stoi(argv[++i]); }
-        else if (arg == "-vsd"  || arg == "--vad-min-silence-duration-ms") { params.vad_min_speech_duration_ms  = std::stoi(argv[++i]); }
+        else if (arg == "-vsd"  || arg == "--vad-min-silence-duration-ms") { params.vad_min_silence_duration_ms = std::stoi(argv[++i]); }
         else if (arg == "-vmsd" || arg == "--vad-max-speech-duration-s")   { params.vad_max_speech_duration_s   = std::stof(argv[++i]); }
         else if (arg == "-vp"   || arg == "--vad-speech-pad-ms")           { params.vad_speech_pad_ms           = std::stoi(argv[++i]); }
         else if (arg == "-vo"   || arg == "--vad-samples-overlap")         { params.vad_samples_overlap         = std::stof(argv[++i]); }
@@ -285,7 +291,7 @@ void check_ffmpeg_availibility() {
     }
 }
 
-std::string generate_temp_filename(const std::string &prefix, const std::string &extension) {
+std::string generate_temp_filename(const std::string &path, const std::string &prefix, const std::string &extension) {
     auto now = std::chrono::system_clock::now();
     auto now_time_t = std::chrono::system_clock::to_time_t(now);
 
@@ -293,7 +299,9 @@ std::string generate_temp_filename(const std::string &prefix, const std::string
     std::uniform_int_distribution<long long> dist(0, 1e9);
 
     std::stringstream ss;
-    ss << prefix
+    ss << path
+       << std::filesystem::path::preferred_separator
+       << prefix
        << "-"
        << std::put_time(std::localtime(&now_time_t), "%Y%m%d-%H%M%S")
        << "-"
@@ -567,10 +575,6 @@ void get_req_parameters(const Request & req, whisper_params & params)
     {
         params.suppress_nst = parse_str_to_bool(req.get_file_value("suppress_nst").content);
     }
-    if (req.has_file("no_context"))
-    {
-        params.no_context = parse_str_to_bool(req.get_file_value("no_context").content);
-    }
     if (req.has_file("vad"))
     {
         params.vad = parse_str_to_bool(req.get_file_value("vad").content);
@@ -599,6 +603,10 @@ void get_req_parameters(const Request & req, whisper_params & params)
     {
         params.vad_samples_overlap = std::stof(req.get_file_value("vad_samples_overlap").content);
     }
+    if (req.has_file("no_language_probabilities"))
+    {
+        params.no_language_probabilities = parse_str_to_bool(req.get_file_value("no_language_probabilities").content);
+    }
 }
 
 }  // namespace
@@ -813,7 +821,7 @@ int main(int argc, char ** argv) {
         if (sparams.ffmpeg_converter) {
             // if file is not wav, convert to wav
             // write to temporary file
-            const std::string temp_filename = generate_temp_filename("whisper-server", ".wav");
+            const std::string temp_filename = generate_temp_filename(sparams.tmp_dir, "whisper-server", ".wav");
             std::ofstream temp_file{temp_filename, std::ios::binary};
             temp_file << audio_file.content;
             temp_file.close();
@@ -1024,23 +1032,25 @@ int main(int argc, char ** argv) {
         } else if (params.response_format == vjson_format) {
             /* try to match openai/whisper's Python format */
             std::string results = output_str(ctx, params, pcmf32s); 
-            // Get language probabilities
-            std::vector<float> lang_probs(whisper_lang_max_id() + 1, 0.0f);
-            const auto detected_lang_id = whisper_lang_auto_detect(ctx, 0, params.n_threads, lang_probs.data());
             json jres = json{
                 {"task", params.translate ? "translate" : "transcribe"},
                 {"language", whisper_lang_str_full(whisper_full_lang_id(ctx))},
                 {"duration", float(pcmf32.size())/WHISPER_SAMPLE_RATE},
                 {"text", results},
-                {"segments", json::array()},
-                {"detected_language", whisper_lang_str_full(detected_lang_id)},
-                {"detected_language_probability", lang_probs[detected_lang_id]},
-                {"language_probabilities", json::object()}
+                {"segments", json::array()}
             };
-            // Add all language probabilities
-            for (int i = 0; i <= whisper_lang_max_id(); ++i) {
-                if (lang_probs[i] > 0.001f) { // Only include non-negligible probabilities
-                    jres["language_probabilities"][whisper_lang_str(i)] = lang_probs[i];
+            // Only compute language probabilities if requested (expensive operation)
+            if (!params.no_language_probabilities) {
+                std::vector<float> lang_probs(whisper_lang_max_id() + 1, 0.0f);
+                const auto detected_lang_id = whisper_lang_auto_detect(ctx, 0, params.n_threads, lang_probs.data());
+                jres["detected_language"] = whisper_lang_str_full(detected_lang_id);
+                jres["detected_language_probability"] = lang_probs[detected_lang_id];
+                jres["language_probabilities"] = json::object();
+                // Add all language probabilities
+                for (int i = 0; i <= whisper_lang_max_id(); ++i) {
+                    if (lang_probs[i] > 0.001f) { // Only include non-negligible probabilities
+                        jres["language_probabilities"][whisper_lang_str(i)] = lang_probs[i];
+                    }
                 }
             }
             const int n_segments = whisper_full_n_segments(ctx);

data/ext/sources/examples/server.py

@@ -47,7 +47,12 @@ class CustomHTTPRequestHandler(http.server.SimpleHTTPRequestHandler):
 
             elif actual_path == '/':
                 self.path = '/whisper.wasm/index.html'
-            elif actual_path.startswith('/bench.wasm/') or actual_path.startswith('/command.wasm/') or actual_path.startswith('/stream.wasm/'):
+            elif any(actual_path.startswith(prefix) for prefix in (
+                '/bench.wasm/',
+                '/command.wasm/',
+                '/stream.wasm/',
+                '/wchess.wasm/'
+            )):
                 # Keep the path as is, just remove the context root
                 self.path = actual_path
             # For all other paths under the context root

data/ext/sources/examples/stream/stream.cpp

@@ -36,7 +36,7 @@ struct whisper_params {
     bool tinydiarize   = false;
     bool save_audio    = false; // save audio to wav file
     bool use_gpu       = true;
-    bool flash_attn    = false;
+    bool flash_attn    = true;
 
     std::string language  = "en";
     std::string model     = "models/ggml-base.en.bin";
@@ -74,6 +74,7 @@ static bool whisper_params_parse(int argc, char ** argv, whisper_params & params
         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 == "-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());
@@ -111,7 +112,8 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
     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, "  -fa,      --flash-attn    [%-7s] flash attention during inference\n",               params.flash_attn ? "true" : "false");
+    fprintf(stderr, "  -fa,      --flash-attn    [%-7s] enable flash attention during inference\n",        params.flash_attn ? "true" : "false");
+    fprintf(stderr, "  -nfa,     --no-flash-attn [%-7s] disable flash attention during inference\n",       params.flash_attn ? "false" : "true");
     fprintf(stderr, "\n");
 }
 

data/ext/sources/examples/stream.wasm/emscripten.cpp

@@ -31,10 +31,11 @@ void stream_set_status(const std::string & status) {
     g_status = status;
 }
 
-void stream_main(size_t index) {
+void stream_main(size_t index, const std::string & lang) {
     stream_set_status("loading data ...");
 
     struct whisper_full_params wparams = whisper_full_default_params(whisper_sampling_strategy::WHISPER_SAMPLING_GREEDY);
+    bool is_multilingual = whisper_is_multilingual(g_contexts[index]);
 
     wparams.n_threads        = std::min(N_THREAD, (int) std::thread::hardware_concurrency());
     wparams.offset_ms        = 0;
@@ -52,7 +53,7 @@ void stream_main(size_t index) {
     // disable temperature fallback
     wparams.temperature_inc  = -1.0f;
 
-    wparams.language         = "en";
+    wparams.language         = is_multilingual ? lang.c_str() : "en";
 
     printf("stream: using %d threads\n", wparams.n_threads);
 
@@ -127,9 +128,8 @@ void stream_main(size_t index) {
         g_contexts[index] = nullptr;
     }
 }
-
 EMSCRIPTEN_BINDINGS(stream) {
-    emscripten::function("init", emscripten::optional_override([](const std::string & path_model) {
+    emscripten::function("init", emscripten::optional_override([](const std::string & path_model, const std::string & lang) {
         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());
@@ -138,8 +138,8 @@ EMSCRIPTEN_BINDINGS(stream) {
                     if (g_worker.joinable()) {
                         g_worker.join();
                     }
-                    g_worker = std::thread([i]() {
-                        stream_main(i);
+                    g_worker = std::thread([i, lang]() {
+                        stream_main(i, lang);
                     });
 
                     return i + 1;

data/ext/sources/examples/stream.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>
 
@@ -55,6 +56,7 @@
                 Whisper model: <span id="model-whisper-status"></span>
                 <button id="fetch-whisper-tiny-en" onclick="loadWhisper('tiny.en')">tiny.en (75 MB)</button>
                 <button id="fetch-whisper-base-en" onclick="loadWhisper('base.en')">base.en (142 MB)</button>
+                <button id="fetch-whisper-base" onclick="loadWhisper('base')">base (142 MB)</button>
                 <br><br>
                 Quantized models:<br><br>
                 <button id="fetch-whisper-tiny-en-q5_1"   onclick="loadWhisper('tiny-en-q5_1')">tiny.en (Q5_1, 31 MB)</button>
@@ -66,6 +68,77 @@
                 -->
             </div>
 
+            <table>
+                <tr>
+                    <td>
+                        Language:
+                        <select id="language" name="language">
+                            <option value="en">English</option>
+                            <option value="ar">Arabic</option>
+                            <option value="hy">Armenian</option>
+                            <option value="az">Azerbaijani</option>
+                            <option value="eu">Basque</option>
+                            <option value="be">Belarusian</option>
+                            <option value="bn">Bengali</option>
+                            <option value="bg">Bulgarian</option>
+                            <option value="ca">Catalan</option>
+                            <option value="zh">Chinese</option>
+                            <option value="hr">Croatian</option>
+                            <option value="cs">Czech</option>
+                            <option value="da">Danish</option>
+                            <option value="nl">Dutch</option>
+                            <option value="en">English</option>
+                            <option value="et">Estonian</option>
+                            <option value="tl">Filipino</option>
+                            <option value="fi">Finnish</option>
+                            <option value="fr">French</option>
+                            <option value="gl">Galician</option>
+                            <option value="ka">Georgian</option>
+                            <option value="de">German</option>
+                            <option value="el">Greek</option>
+                            <option value="gu">Gujarati</option>
+                            <option value="iw">Hebrew</option>
+                            <option value="hi">Hindi</option>
+                            <option value="hu">Hungarian</option>
+                            <option value="is">Icelandic</option>
+                            <option value="id">Indonesian</option>
+                            <option value="ga">Irish</option>
+                            <option value="it">Italian</option>
+                            <option value="ja">Japanese</option>
+                            <option value="kn">Kannada</option>
+                            <option value="ko">Korean</option>
+                            <option value="la">Latin</option>
+                            <option value="lv">Latvian</option>
+                            <option value="lt">Lithuanian</option>
+                            <option value="mk">Macedonian</option>
+                            <option value="ms">Malay</option>
+                            <option value="mt">Maltese</option>
+                            <option value="no">Norwegian</option>
+                            <option value="fa">Persian</option>
+                            <option value="pl">Polish</option>
+                            <option value="pt">Portuguese</option>
+                            <option value="ro">Romanian</option>
+                            <option value="ru">Russian</option>
+                            <option value="sr">Serbian</option>
+                            <option value="sk">Slovak</option>
+                            <option value="sl">Slovenian</option>
+                            <option value="es">Spanish</option>
+                            <option value="sw">Swahili</option>
+                            <option value="sv">Swedish</option>
+                            <option value="ta">Tamil</option>
+                            <option value="te">Telugu</option>
+                            <option value="th">Thai</option>
+                            <option value="tr">Turkish</option>
+                            <option value="uk">Ukrainian</option>
+                            <option value="ur">Urdu</option>
+                            <option value="vi">Vietnamese</option>
+                            <option value="cy">Welsh</option>
+                            <option value="yi">Yiddish</option>
+                        </select>
+                    </td>
+                </tr>
+            </table>
+
             <br>
 
             <div id="input">
@@ -174,16 +247,18 @@
 
             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',
+                    'base'   : 'https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.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 = {
                     'tiny.en': 75,
                     'base.en': 142,
+                    'base':     142,
 
                     'tiny-en-q5_1':   31,
                     'base-en-q5_1':   57,
@@ -197,6 +272,7 @@
 
                 document.getElementById('fetch-whisper-tiny-en').style.display = 'none';
                 document.getElementById('fetch-whisper-base-en').style.display = 'none';
+                document.getElementById('fetch-whisper-base').style.display = 'none';
 
                 document.getElementById('fetch-whisper-tiny-en-q5_1').style.display = 'none';
                 document.getElementById('fetch-whisper-base-en-q5_1').style.display = 'none';
@@ -212,6 +288,7 @@
                     var el;
                     el = document.getElementById('fetch-whisper-tiny-en'); if (el) el.style.display = 'inline-block';
                     el = document.getElementById('fetch-whisper-base-en'); if (el) el.style.display = 'inline-block';
+                    el = document.getElementById('fetch-whisper-base'); if (el) el.style.display = 'inline-block';
 
                     el = document.getElementById('fetch-whisper-tiny-en-q5_1'); if (el) el.style.display = 'inline-block';
                     el = document.getElementById('fetch-whisper-base-en-q5_1'); if (el) el.style.display = 'inline-block';
@@ -368,7 +445,7 @@
 
             function onStart() {
                 if (!instance) {
-                    instance = Module.init('whisper.bin');
+                    instance = Module.init('whisper.bin', document.getElementById('language').value);
 
                     if (instance) {
                         printTextarea("js: whisper initialized, instance: " + instance);

data/ext/sources/examples/talk-llama/CMakeLists.txt

@@ -2,6 +2,8 @@ if (WHISPER_SDL2)
     set(CMAKE_CXX_STANDARD 17)
     set(CMAKE_CXX_STANDARD_REQUIRED ON)
 
+    file(GLOB SRC_MODELS models/*.cpp)
+
     set(TARGET whisper-talk-llama)
     add_executable(${TARGET} talk-llama.cpp
         llama.cpp
@@ -16,8 +18,8 @@ if (WHISPER_SDL2)
         llama-hparams.cpp
         llama-impl.cpp
         llama-io.cpp
-        llama-kv-cache-unified.cpp
-        llama-kv-cache-unified-iswa.cpp
+        llama-kv-cache.cpp
+        llama-kv-cache-iswa.cpp
         llama-memory-recurrent.cpp
         llama-memory-hybrid.cpp
         llama-memory.cpp
@@ -29,10 +31,12 @@ if (WHISPER_SDL2)
         llama-sampling.cpp
         llama-vocab.cpp
         unicode.cpp
-        unicode-data.cpp)
+        unicode-data.cpp
+        ${SRC_MODELS})
     target_include_directories(${TARGET} PRIVATE ${SDL2_INCLUDE_DIRS})
 
     target_link_libraries(${TARGET} PRIVATE common common-sdl whisper ${SDL2_LIBRARIES} ${CMAKE_THREAD_LIBS_INIT})
+    install(TARGETS ${TARGET} RUNTIME)
 
     if(WIN32)
         # It requires Windows 8.1 or later for PrefetchVirtualMemory