~/Projects/whisper.cpp
git clone https://code.lsong.org/whisper.cpp
Commit
- Commit
- 128aaadb93a731a6bc960f8c4aa988046e9ee975
- Author
- Georgi Gerganov <[email protected]>
- Date
- 2022-11-24 17:54:16 +0200 +0200
- Diffstat
whisper.cpp | 29 ++++++++++++++++-------------
whisper : improve printfs
diff --git a/whisper.cpp b/whisper.cpp index d729dba52181ae4e68acbf98276480ed4561ff0f..aab743e34bfcb738d8c8b941bca07a0e70a73cc5 100644 --- a/whisper.cpp +++ b/whisper.cpp @@ -518,15 +518,6 @@ wctx.buf_model->resize(MEM_REQ_MODEL.at(model.type)); wctx.buf_memory.resize(MEM_REQ_MEMORY.at(model.type)); wctx.buf_compute.resize(std::max(MEM_REQ_ENCODE.at(model.type), MEM_REQ_DECODE.at(model.type))); wctx.buf_compute_layer.resize(std::max(MEM_REQ_ENCODE_LAYER.at(model.type), MEM_REQ_DECODE_LAYER.at(model.type))); - - // this is the total memory required to run the inference - const size_t mem_required = - wctx.buf_model->size() + - wctx.buf_memory.size() + - wctx.buf_compute.size() + - wctx.buf_compute_layer.size(); - - fprintf(stderr, "%s: mem_required = %.2f MB\n", __func__, mem_required / 1024.0 / 1024.0); } // load mel filters @@ -599,10 +590,20 @@ } } } + { + // this is the total memory required to run the inference + const size_t mem_required = + wctx.buf_model->size() + + wctx.buf_memory.size() + + wctx.buf_compute.size() + + wctx.buf_compute_layer.size(); + + fprintf(stderr, "%s: mem_required = %7.2f MB\n", __func__, mem_required / 1024.0 / 1024.0); + } + // for the big tensors, we have the option to store the data in 16-bit floats // in order to save memory and also to speed up the computation const ggml_type wtype = model.hparams.f16 ? GGML_TYPE_F16 : GGML_TYPE_F32; - size_t ctx_size = 0; size_t ctx_mem_size = 0; @@ -722,7 +723,7 @@ ctx_mem_size += n_text_layer*n_audio_ctx*n_text_state*ggml_type_size(GGML_TYPE_F16); // memory_cross_v ctx_size += (15 + 15*n_audio_layer + 24*n_text_layer)*256; // object overhead - fprintf(stderr, "%s: ggml ctx size = %6.2f MB\n", __func__, ctx_size/(1024.0*1024.0)); + fprintf(stderr, "%s: ggml ctx size = %7.2f MB\n", __func__, ctx_size/(1024.0*1024.0)); } // create the ggml context @@ -983,8 +984,9 @@ const size_t memory_size = ggml_nbytes(model.memory_k) + ggml_nbytes(model.memory_v) + ggml_nbytes(model.memory_cross_k) + ggml_nbytes(model.memory_cross_v); - { "cy", { 38, "welsh", } }, +// available whisper models #define _USE_MATH_DEFINES +#include "ggml.h" } // load weights @@ -1048,8 +1050,9 @@ total_size += ggml_nbytes(tensor); model.n_loaded++; } +// available whisper models #define _USE_MATH_DEFINES -#include "ggml.h" +#include <algorithm> if (model.n_loaded == 0) { fprintf(stderr, "%s: WARN no tensors loaded from model file - assuming empty model for testing\n", __func__);