ExternalVendorCode/whisper.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/whisper.cpp.git synced 2025-06-17 14:28:07 +00:00
Code Issues Actions 4 Packages Projects Releases Wiki Activity

whisper : reduce memory usage during inference (#431)

Browse Source 
* ggml : add "scratch" buffer support

* ggml : support for scratch ring-buffer

* ggml : bug fix in ggml_repeat()

* ggml : error on scratch buffer overflow

* whisper : use scratch buffers during inference (base model only)

* whisper : update memory usage for all models

* whisper : fix encoder memory usage

* whisper : use whisper_context functions instead of macros

* whisper : fix FF + remove it from README

* ggml : reuse ggml_new_i32

* ggml : refactor the scratch buffer storage

* whisper : reorder scratch buffers in the decoder

* main : add option to disable temp fallback

* Update README.md
This commit is contained in:
Georgi Gerganov
2023-02-04 09:45:52 +02:00
committed by GitHub
parent c306a7fd89
commit f3ee4a9673
7 changed files with 702 additions and 472 deletions
Download Patch File Download Diff File Expand all files Collapse all files

52
examples/main/README.md
View File

@ -9,25 +9,35 @@ It can be used as a reference for using the `whisper.cpp` library in other proje
usage: ./main [options] file0.wav file1.wav ...
options:
-h, --help [default] show this help message and exit
-t N, --threads N [4 ] number of threads to use during computation
-p N, --processors N [1 ] number of processors to use during computation
-ot N, --offset-t N [0 ] time offset in milliseconds
-on N, --offset-n N [0 ] segment index offset
-d N, --duration N [0 ] duration of audio to process in milliseconds
-mc N, --max-context N [-1 ] maximum number of text context tokens to store
-ml N, --max-len N [0 ] maximum segment length in characters
-wt N, --word-thold N [0.01 ] word timestamp probability threshold
-su, --speed-up [false ] speed up audio by x2 (reduced accuracy)
-tr, --translate [false ] translate from source language to english
-otxt, --output-txt [false ] output result in a text file
-ovtt, --output-vtt [false ] output result in a vtt file
-osrt, --output-srt [false ] output result in a srt file
-owts, --output-words [false ] output script for generating karaoke video
-ps, --print-special [false ] print special tokens
-pc, --print-colors [false ] print colors
-nt, --no-timestamps [true ] do not print timestamps
-l LANG, --language LANG [en ] spoken language
-m FNAME, --model FNAME [models/ggml-base.en.bin] model path
-f FNAME, --file FNAME [ ] input WAV file path
-h, --help [default] show this help message and exit
-t N, --threads N [4 ] number of threads to use during computation
-p N, --processors N [1 ] number of processors to use during computation
-ot N, --offset-t N [0 ] time offset in milliseconds
-on N, --offset-n N [0 ] segment index offset
-d N, --duration N [0 ] duration of audio to process in milliseconds
-mc N, --max-context N [-1 ] maximum number of text context tokens to store
-ml N, --max-len N [0 ] maximum segment length in characters
-bo N, --best-of N [5 ] number of best candidates to keep
-bs N, --beam-size N [-1 ] beam size for beam search
-wt N, --word-thold N [0.01 ] word timestamp probability threshold
-et N, --entropy-thold N [2.40 ] entropy threshold for decoder fail
-lpt N, --logprob-thold N [-1.00 ] log probability threshold for decoder fail
-su, --speed-up [false ] speed up audio by x2 (reduced accuracy)
-tr, --translate [false ] translate from source language to english
-di, --diarize [false ] stereo audio diarization
-nf, --no-fallback [false ] do not use temperature fallback while decoding
-otxt, --output-txt [false ] output result in a text file
-ovtt, --output-vtt [false ] output result in a vtt file
-osrt, --output-srt [false ] output result in a srt file
-owts, --output-words [false ] output script for generating karaoke video
-ocsv, --output-csv [false ] output result in a CSV file
-of FNAME, --output-file FNAME [ ] output file path (without file extension)
-ps, --print-special [false ] print special tokens
-pc, --print-colors [false ] print colors
-pp, --print-progress [false ] print progress
-nt, --no-timestamps [true ] do not print timestamps
-l LANG, --language LANG [en ] spoken language ('auto' for auto-detect)
--prompt PROMPT [ ] initial prompt
-m FNAME, --model FNAME [models/ggml-base.en.bin] model path
-f FNAME, --file FNAME [ ] input WAV file path
```

33
examples/main/main.cpp
View File

@ -53,22 +53,23 @@ void replace_all(std::string & s, const std::string & search, const std::string
// command-line parameters
struct whisper_params {
int32_t n_threads = std::min(4, (int32_t) std::thread::hardware_concurrency());
int32_t n_processors = 1;
int32_t offset_t_ms = 0;
int32_t offset_n = 0;
int32_t duration_ms = 0;
int32_t n_processors = 1;
int32_t offset_t_ms = 0;
int32_t offset_n = 0;
int32_t duration_ms = 0;
int32_t max_context = -1;
int32_t max_len = 0;
int32_t best_of = 5;
int32_t max_len = 0;
int32_t best_of = 5;
int32_t beam_size = -1;
float word_thold = 0.01f;
float entropy_thold = 2.4f;
float logprob_thold = -1.0f;
float word_thold = 0.01f;
float entropy_thold = 2.40f;
float logprob_thold = -1.00f;
bool speed_up = false;
bool translate = false;
bool diarize = false;
bool no_fallback = false;
bool output_txt = false;
bool output_vtt = false;
bool output_srt = false;
@ -117,6 +118,7 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params) {
else if (arg == "-su" || arg == "--speed-up") { params.speed_up = true; }
else if (arg == "-tr" || arg == "--translate") { params.translate = true; }
else if (arg == "-di" || arg == "--diarize") { params.diarize = true; }
else if (arg == "-nf" || arg == "--no-fallback") { params.no_fallback = true; }
else if (arg == "-otxt" || arg == "--output-txt") { params.output_txt = true; }
else if (arg == "-ovtt" || arg == "--output-vtt") { params.output_vtt = true; }
else if (arg == "-osrt" || arg == "--output-srt") { params.output_srt = true; }
@ -162,6 +164,7 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
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, " -di, --diarize [%-7s] stereo audio diarization\n", params.diarize ? "true" : "false");
fprintf(stderr, " -nf, --no-fallback [%-7s] do not use temperature fallback while decoding\n", params.no_fallback ? "true" : "false");
fprintf(stderr, " -otxt, --output-txt [%-7s] output result in a text file\n", params.output_txt ? "true" : "false");
fprintf(stderr, " -ovtt, --output-vtt [%-7s] output result in a vtt file\n", params.output_vtt ? "true" : "false");
fprintf(stderr, " -osrt, --output-srt [%-7s] output result in a srt file\n", params.output_srt ? "true" : "false");
@ -514,7 +517,7 @@ int main(int argc, char ** argv) {
for (int f = 0; f < (int) params.fname_inp.size(); ++f) {
const auto fname_inp = params.fname_inp[f];
const auto fname_outp = f < params.fname_outp.size() && !params.fname_outp[f].empty() ? params.fname_outp[f] : params.fname_inp[f];
const auto fname_outp = f < (int) params.fname_outp.size() && !params.fname_outp[f].empty() ? params.fname_outp[f] : params.fname_inp[f];
std::vector<float> pcmf32; // mono-channel F32 PCM
std::vector<std::vector<float>> pcmf32s; // stereo-channel F32 PCM
@ -647,17 +650,19 @@ int main(int argc, char ** argv) {
wparams.token_timestamps = params.output_wts || params.max_len > 0;
wparams.thold_pt = params.word_thold;
wparams.entropy_thold = params.entropy_thold;
wparams.logprob_thold = params.logprob_thold;
wparams.max_len = params.output_wts && params.max_len == 0 ? 60 : params.max_len;
wparams.speed_up = params.speed_up;
wparams.prompt_tokens = prompt_tokens.empty() ? nullptr : prompt_tokens.data();
wparams.prompt_n_tokens = prompt_tokens.empty() ? 0 : prompt_tokens.size();
wparams.greedy.best_of = params.best_of;
wparams.beam_search.beam_size = params.beam_size;
wparams.prompt_tokens = prompt_tokens.empty() ? nullptr : prompt_tokens.data();
wparams.prompt_n_tokens = prompt_tokens.empty() ? 0 : prompt_tokens.size();
wparams.temperature_inc = params.no_fallback ? 0.0f : wparams.temperature_inc;
wparams.entropy_thold = params.entropy_thold;
wparams.logprob_thold = params.logprob_thold;
whisper_print_user_data user_data = { &params, &pcmf32s };