mirrors/obs-localvocal
1
0
Fork 0
mirror of https://github.com/occ-ai/obs-localvocal synced 2024-11-07 18:57:14 +00:00
Code Issues Packages Projects Releases Wiki Activity

Update CUDA support and model versions (#78)

Browse Source
This commit is contained in:
Roy Shilkrot 2024-03-24 21:23:06 -04:00 committed by GitHub
parent 6791e5a5d3
commit 0c7d7234af
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
8 changed files with 43 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
CMakeLists.txt
View File

@ -41,6 +41,7 @@ set(USE_SYSTEM_CURL
CACHE STRING "Use system cURL")
if(LOCALVOCAL_WITH_CUDA)
message(STATUS "Building with CUDA support")
add_compile_definitions("LOCALVOCAL_WITH_CUDA")
endif()

2
buildspec.json
View File

@ -45,7 +45,7 @@
}
},
"name": "obs-localvocal",
"version": "0.1.1",
"version": "0.2.0",
"author": "Roy Shilkrot",
"website": "https://github.com/occ-ai/obs-localvocal",
"email": "roy.shil@gmail.com",

2
cmake/BuildWhispercpp.cmake
View File

@ -42,7 +42,7 @@ if(WIN32)
endif(NOT DEFINED CUDA_TOOLKIT_ROOT_DIR)
set(WHISPER_ADDITIONAL_ENV "CUDAToolkit_ROOT=${CUDA_TOOLKIT_ROOT_DIR}")
set(WHISPER_ADDITIONAL_CMAKE_ARGS -DWHISPER_CUBLAS=ON -DWHISPER_OPENBLAS=OFF
set(WHISPER_ADDITIONAL_CMAKE_ARGS -DWHISPER_BLAS=OFF -DWHISPER_CUBLAS=ON -DWHISPER_OPENBLAS=OFF
-DCMAKE_GENERATOR_TOOLSET=cuda=${CUDA_TOOLKIT_ROOT_DIR})
else()
# Build with OpenBLAS

0
data/models/ggml-tiny.en.bin → data/models/ggml-model-whisper-tiny.en.bin
View File

5
src/model-utils/model-downloader-ui.cpp
View File

@ -5,8 +5,7 @@
#include <filesystem>
const std::string MODEL_BASE_PATH = "https://huggingface.co/ggerganov/whisper.cpp";
const std::string MODEL_PREFIX = "resolve/main/";
const std::string MODEL_BASE_PATH = "https://ggml.ggerganov.com/";
size_t write_data(void *ptr, size_t size, size_t nmemb, FILE *stream)
{
@ -143,7 +142,7 @@ void ModelDownloadWorker::download_model()
const std::string model_filename =
this->model_name.substr(this->model_name.find_last_of("/\\") + 1);
std::string model_url = MODEL_BASE_PATH + "/" + MODEL_PREFIX + model_filename;
std::string model_url = MODEL_BASE_PATH + model_filename;
obs_log(LOG_INFO, "Model URL: %s", model_url.c_str());
CURL *curl = curl_easy_init();

44
src/transcription-filter.cpp
View File

@ -654,7 +654,8 @@ void transcription_filter_defaults(obs_data_t *s)
obs_data_set_default_int(s, "log_level", LOG_DEBUG);
obs_data_set_default_bool(s, "log_words", true);
obs_data_set_default_bool(s, "caption_to_stream", false);
obs_data_set_default_string(s, "whisper_model_path", "models/ggml-tiny.en.bin");
obs_data_set_default_string(s, "whisper_model_path",
"models/ggml-model-whisper-tiny.en.bin");
obs_data_set_default_string(s, "whisper_language_select", "en");
obs_data_set_default_string(s, "subtitle_sources", "none");
obs_data_set_default_bool(s, "step_by_step_processing", false);
@ -754,15 +755,38 @@ obs_properties_t *transcription_filter_properties(void *data)
obs_properties_add_list(ppts, "whisper_model_path", MT_("whisper_model"),
OBS_COMBO_TYPE_LIST, OBS_COMBO_FORMAT_STRING);
obs_property_list_add_string(whisper_models_list, "Tiny (Eng) 75Mb",
"models/ggml-tiny.en.bin");
obs_property_list_add_string(whisper_models_list, "Tiny 75Mb", "models/ggml-tiny.bin");
obs_property_list_add_string(whisper_models_list, "Base (Eng) 142Mb",
"models/ggml-base.en.bin");
obs_property_list_add_string(whisper_models_list, "Base 142Mb", "models/ggml-base.bin");
obs_property_list_add_string(whisper_models_list, "Small (Eng) 466Mb",
"models/ggml-small.en.bin");
obs_property_list_add_string(whisper_models_list, "Small 466Mb", "models/ggml-small.bin");
obs_property_list_add_string(whisper_models_list, "Base q5 57M",
"models/ggml-model-whisper-base-q5_1.bin");
obs_property_list_add_string(whisper_models_list, "Base 141M",
"models/ggml-model-whisper-base.bin");
obs_property_list_add_string(whisper_models_list, "Base (Eng) q5 57M",
"models/ggml-model-whisper-base.en-q5_1.bin");
obs_property_list_add_string(whisper_models_list, "Base (Eng) 141M",
"models/ggml-model-whisper-base.en.bin");
obs_property_list_add_string(whisper_models_list, "Large q5 1G",
"models/ggml-model-whisper-large-q5_0.bin");
obs_property_list_add_string(whisper_models_list, "Medium q5 514M",
"models/ggml-model-whisper-medium-q5_0.bin");
obs_property_list_add_string(whisper_models_list, "Medium (Eng) 514M",
"models/ggml-model-whisper-medium.en-q5_0.bin");
obs_property_list_add_string(whisper_models_list, "Small q5 181M",
"models/ggml-model-whisper-small-q5_1.bin");
obs_property_list_add_string(whisper_models_list, "Small 465M",
"models/ggml-model-whisper-small.bin");
obs_property_list_add_string(whisper_models_list, "Small (Eng) q5 181M",
"models/ggml-model-whisper-small.en-q5_1.bin");
obs_property_list_add_string(whisper_models_list, "Small (Eng) 465M",
"models/ggml-model-whisper-small.en.bin");
obs_property_list_add_string(whisper_models_list, "Tiny q5 31M",
"models/ggml-model-whisper-tiny-q5_1.bin");
obs_property_list_add_string(whisper_models_list, "Tiny 74M",
"models/ggml-model-whisper-tiny.bin");
obs_property_list_add_string(whisper_models_list, "Tiny (Eng) q5 31M",
"models/ggml-model-whisper-tiny.en-q5_1.bin");
obs_property_list_add_string(whisper_models_list, "Tiny (Eng) q8 42M",
"models/ggml-model-whisper-tiny.en-q8_0.bin");
obs_property_list_add_string(whisper_models_list, "Tiny (Eng) 74M",
"models/ggml-model-whisper-tiny.en.bin");
obs_property_list_add_string(whisper_models_list, "Load external model file",
"!!!external!!!");

4
src/whisper-utils/whisper-processing.cpp
View File

@ -112,11 +112,13 @@ struct whisper_context *init_whisper_context(const std::string &model_path)
{
obs_log(LOG_INFO, "Loading whisper model from %s", model_path.c_str());
struct whisper_context_params cparams;
struct whisper_context_params cparams = whisper_context_default_params();
#ifdef LOCALVOCAL_WITH_CUDA
cparams.use_gpu = true;
obs_log(LOG_INFO, "Using GPU for inference, device %d", cparams.gpu_device);
#else
cparams.use_gpu = false;
obs_log(LOG_INFO, "Using CPU for inference");
#endif
#ifdef _WIN32

2
vendor/curl vendored

@ -1 +1 @@
Subproject commit 439ff2052e219162708faddedacdf6f1242bb8c8
Subproject commit 98044e81705dc24a56daaf3544f30c13f0fc3a31