upstream-archive/whisper.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/whisper.cpp.git synced 2026-07-23 11:10:57 -05:00
Code Issues Packages Projects Releases Wiki Activity

whisper : enable flash attention by default (#3441)

Browse Source
This commit is contained in:
Georgi Gerganov
2025-09-30 15:47:20 +03:00
committed by GitHub
parent 1e5ad50f8f
commit 0b3587acdd
10 changed files with 103 additions and 87 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
examples/server/server.cpp
View File

@@ -101,7 +101,7 @@ 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_language_probabilities = false;
@@ -178,7 +178,8 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
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");
@@ -236,6 +237,7 @@ 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; }