public_git/whisper.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/whisper.cpp.git synced 2026-10-02 12:31:21 +02:00
Code Issues Packages Projects Releases Wiki Activity

whisper : add context param to disable gpu (#1293)

Browse Source 
* whisper : check state->ctx_metal not null

* whisper : add whisper_context_params { use_gpu }

* whisper : new API with params & deprecate old API

* examples : use no-gpu param && whisper_init_from_file_with_params

* whisper.objc : enable metal & disable on simulator

* whisper.swiftui, metal : enable metal & support load default.metallib

* whisper.android : use new API

* bindings : use new API

* addon.node : fix build & test

* bindings : updata java binding

* bindings : add missing whisper_context_default_params_by_ref WHISPER_API for java

* metal : use SWIFTPM_MODULE_BUNDLE for GGML_SWIFT and reuse library load

* metal : move bundle var into block

* metal : use SWIFT_PACKAGE instead of GGML_SWIFT

* style : minor updates

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
Jhen-Jie Hong
2023-11-06 11:04:24 +02:00
committed by GitHub
co-authored by Georgi Gerganov
parent 39cfad0dee
commit 0463028bc2
29 changed files with 439 additions and 188 deletions
Download Patch File Download Diff File Expand all files Collapse all files
examples/addon.node/__test__/whisper.spec.js
+1
View File
@@ -11,6 +11,7 @@ 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
+6 -1
View File
@@ -36,6 +36,7 @@ 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;
@@ -153,7 +154,9 @@ int run(whisper_params &params, std::vector<std::vector<std::string>> &result) {
// whisper init
struct whisper_context * ctx = whisper_init_from_file(params.model.c_str());
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);
if (ctx == nullptr) {
fprintf(stderr, "error: failed to initialize whisper context\n");
@@ -315,10 +318,12 @@ 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
+1
View File
@@ -11,6 +11,7 @@ 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
+1 -1
View File
@@ -57,7 +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(path_model.c_str());
g_contexts[i] = whisper_init_from_file_with_params(path_model.c_str(), whisper_context_default_params());
if (g_contexts[i] != nullptr) {
if (g_worker.joinable()) {
g_worker.join();
examples/bench/bench.cpp
+11 -4
View File
@@ -11,6 +11,8 @@ 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);
@@ -23,9 +25,10 @@ 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 == "-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 {
fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
whisper_print_usage(argc, argv, params);
@@ -45,6 +48,7 @@ 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", "");
@@ -54,7 +58,10 @@ 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 * ctx = whisper_init_from_file(params.model.c_str());
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);
{
fprintf(stderr, "\n");
examples/command.wasm/emscripten.cpp
+1 -1
View File
@@ -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(path_model.c_str());
g_contexts[i] = whisper_init_from_file_with_params(path_model.c_str(), whisper_context_default_params());
if (g_contexts[i] != nullptr) {
g_running = true;
if (g_worker.joinable()) {
examples/command/command.cpp
+7 -1
View File
@@ -38,6 +38,7 @@ 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";
@@ -68,6 +69,7 @@ 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]; }
@@ -101,6 +103,7 @@ 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());
@@ -610,7 +613,10 @@ int main(int argc, char ** argv) {
// whisper init
struct whisper_context * ctx = whisper_init_from_file(params.model.c_str());
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);
// print some info about the processing
{
examples/lsp/lsp.cpp
+6 -1
View File
@@ -30,6 +30,7 @@ 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";
@@ -72,6 +73,7 @@ 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 {
@@ -102,6 +104,7 @@ 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");
@@ -432,7 +435,9 @@ int main(int argc, char ** argv) {
}
// whisper init
struct whisper_context * ctx = whisper_init_from_file(params.model.c_str());
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);
// init audio
audio_async audio(30*1000);
examples/main/main.cpp
+7 -1
View File
@@ -90,6 +90,7 @@ 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;
@@ -165,6 +166,7 @@ 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);
@@ -221,6 +223,7 @@ 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");
}
@@ -877,7 +880,10 @@ int main(int argc, char ** argv) {
// whisper init
struct whisper_context * ctx = whisper_init_from_file(params.model.c_str());
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);
if (ctx == nullptr) {
fprintf(stderr, "error: failed to initialize whisper context\n");
examples/stream.wasm/emscripten.cpp
+1 -1
View File
@@ -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(path_model.c_str());
g_contexts[i] = whisper_init_from_file_with_params(path_model.c_str(), whisper_context_default_params());
if (g_contexts[i] != nullptr) {
g_running = true;
if (g_worker.joinable()) {
examples/stream/stream.cpp
+29 -23
View File
@@ -48,11 +48,12 @@ 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);
@@ -65,25 +66,26 @@ 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 == "-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 {
fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
@@ -118,8 +120,9 @@ 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");
}
@@ -163,7 +166,10 @@ int main(int argc, char ** argv) {
exit(0);
}
struct whisper_context * ctx = whisper_init_from_file(params.model.c_str());
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);
std::vector<float> pcmf32 (n_samples_30s, 0.0f);
std::vector<float> pcmf32_old;
@@ -424,4 +430,4 @@ int main(int argc, char ** argv) {
whisper_free(ctx);
return 0;
}
}
examples/talk-llama/talk-llama.cpp
+51 -41
View File
@@ -63,6 +63,7 @@ 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";
@@ -84,25 +85,26 @@ 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 == "--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") {
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") {
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') {
@@ -110,6 +112,7 @@ 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);
@@ -125,27 +128,28 @@ 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, " -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, " -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, "\n");
}
@@ -252,7 +256,10 @@ int main(int argc, char ** argv) {
// whisper init
struct whisper_context * ctx_wsp = whisper_init_from_file(params.model_wsp.c_str());
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);
// llama init
@@ -269,6 +276,9 @@ int main(int argc, char ** argv) {
lcparams.seed = 1;
lcparams.f16_kv = true;
lcparams.n_threads = params.n_threads;
if (!params.use_gpu) {
lcparams.n_gpu_layers = 0;
}
struct llama_context * ctx_llama = llama_new_context_with_model(model_llama, lcparams);
examples/talk.wasm/emscripten.cpp
+1 -1
View File
@@ -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(path_model.c_str());
g_contexts[i] = whisper_init_from_file_with_params(path_model.c_str(), whisper_context_default_params());
if (g_contexts[i] != nullptr) {
g_running = true;
if (g_worker.joinable()) {
examples/talk/talk.cpp
+6 -1
View File
@@ -31,6 +31,7 @@ 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";
@@ -61,6 +62,7 @@ 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]; }
@@ -94,6 +96,7 @@ 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());
@@ -181,8 +184,10 @@ 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(params.model_wsp.c_str());
struct whisper_context * ctx_wsp = whisper_init_from_file_with_params(params.model_wsp.c_str(), cparams);
// gpt init
examples/whisper.android/app/src/main/jni/whisper/jni.c
+2 -2
View File
@@ -127,7 +127,7 @@ static struct whisper_context *whisper_init_from_asset(
.close = &asset_close
};
return whisper_init(&loader);
return whisper_init_with_params(&loader, whisper_context_default_params());
}
JNIEXPORT jlong JNICALL
@@ -147,7 +147,7 @@ Java_com_whispercppdemo_whisper_WhisperLib_00024Companion_initContext(
UNUSED(thiz);
struct whisper_context *context = NULL;
const char *model_path_chars = (*env)->GetStringUTFChars(env, model_path_str, NULL);
context = whisper_init_from_file(model_path_chars);
context = whisper_init_from_file_with_params(model_path_chars, whisper_context_default_params());
(*env)->ReleaseStringUTFChars(env, model_path_str, model_path_chars);
return (jlong) context;
}
examples/whisper.objc/whisper.objc.xcodeproj/project.pbxproj
+2 -2
View File
@@ -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"; }; };
18627C9629052C5800BD2A04 /* ggml.c in Sources */ = {isa = PBXBuildFile; fileRef = 18627C9529052C5800BD2A04 /* ggml.c */; settings = {COMPILER_FLAGS = "-DGGML_USE_ACCELERATE"; }; };
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"; }; };
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
+7 -1
View File
@@ -61,7 +61,13 @@ void AudioInputCallback(void * inUserData,
NSLog(@"Loading model from %@", modelPath);
// create ggml context
stateInp.ctx = whisper_init_from_file([modelPath UTF8String]);
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);
// check if the model was loaded successfully
if (stateInp.ctx == NULL) {
examples/whisper.swiftui/whisper.cpp.swift/LibWhisper.swift
+6 -1
View File
@@ -55,7 +55,12 @@ actor WhisperContext {
}
static func createContext(path: String) throws -> WhisperContext {
let context = whisper_init_from_file(path)
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)
if let context {
return WhisperContext(context: context)
} else {
examples/whisper.swiftui/whisper.swiftui.xcodeproj/project.pbxproj
+12 -2
View File
@@ -16,13 +16,15 @@
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"; }; };
0AAC5DCB29539EB1003032C3 /* whisper.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 0AAC5DC729539EB0003032C3 /* whisper.cpp */; settings = {COMPILER_FLAGS = "-DGGML_USE_METAL -Wno-shorten-64-to-32"; }; };
0AAC5DCC29539EB1003032C3 /* ggml.c in Sources */ = {isa = PBXBuildFile; fileRef = 0AAC5DC929539EB0003032C3 /* ggml.c */; settings = {COMPILER_FLAGS = "-DGGML_USE_ACCELERATE -DGGML_USE_METAL -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 */; };
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 */; };
7FCB08262ACFA3A400AF3530 /* ggml-metal.m in Sources */ = {isa = PBXBuildFile; fileRef = 7FCB08252ACFA3A400AF3530 /* ggml-metal.m */; settings = {COMPILER_FLAGS = "-framework Foundation -framework Metal -framework MetalKit -fno-objc-arc"; }; };
7FCB08282ACFA48500AF3530 /* ggml-metal.metal in Sources */ = {isa = PBXBuildFile; fileRef = 7FCB08272ACFA48500AF3530 /* ggml-metal.metal */; };
/* End PBXBuildFile section */
/* Begin PBXFileReference section */
@@ -52,6 +54,9 @@
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>"; };
7FCB081E2ACFA04400AF3530 /* ggml-metal.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = "ggml-metal.h"; sourceTree = "<group>"; };
7FCB08252ACFA3A400AF3530 /* ggml-metal.m */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.objc; path = "ggml-metal.m"; sourceTree = "<group>"; };
7FCB08272ACFA48500AF3530 /* ggml-metal.metal */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.metal; path = "ggml-metal.metal"; sourceTree = "<group>"; };
/* End PBXFileReference section */
/* Begin PBXFrameworksBuildPhase section */
@@ -135,6 +140,9 @@
0AAC5DC529539E89003032C3 /* whisper.cpp */ = {
isa = PBXGroup;
children = (
7FCB08272ACFA48500AF3530 /* ggml-metal.metal */,
7FCB081E2ACFA04400AF3530 /* ggml-metal.h */,
7FCB08252ACFA3A400AF3530 /* ggml-metal.m */,
18ABE14E2AF555FA0044A204 /* ggml-backend-impl.h */,
18ABE14C2AF555FA0044A204 /* ggml-backend.c */,
18ABE14D2AF555FA0044A204 /* ggml-backend.h */,
@@ -258,10 +266,12 @@
0AAC5DCC29539EB1003032C3 /* ggml.c in Sources */,
18ABE1532AF555FA0044A204 /* ggml-quants.c in Sources */,
0AAC5DCE2953A05C003032C3 /* WhisperState.swift in Sources */,
7FCB08282ACFA48500AF3530 /* ggml-metal.metal in Sources */,
0AAC5DD12953A394003032C3 /* LibWhisper.swift in Sources */,
0AA7514C2953B569001EE061 /* RiffWaveUtils.swift in Sources */,
0AAC5DCB29539EB1003032C3 /* whisper.cpp in Sources */,
0AA7514E2953D958001EE061 /* Recorder.swift in Sources */,
7FCB08262ACFA3A400AF3530 /* ggml-metal.m in Sources */,
18AED4812AB21F2B009D854F /* ggml-alloc.c in Sources */,
18ABE1522AF555FA0044A204 /* ggml-backend.c in Sources */,
);
examples/whisper.wasm/emscripten.cpp
+1 -1
View File
@@ -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(path_model.c_str());
g_contexts[i] = whisper_init_from_file_with_params(path_model.c_str(), whisper_context_default_params());
if (g_contexts[i] != nullptr) {
return i + 1;
} else {