extern/whisper.cpp
1
0
Fork 1
mirror of https://github.com/ggerganov/whisper.cpp.git synced 2025-02-20 20:20:57 +01:00
Code Issues Packages Projects Releases Wiki Activity

whisper : add description of ggml_mul_mat_pad

Browse Source
This commit is contained in:
Georgi Gerganov 2023-09-14 15:36:36 +03:00
parent f36554382a
commit 2b4160af29
No known key found for this signature in database
GPG Key ID: 449E073F9DC10735
2 changed files with 15 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
coreml/whisper-encoder.mm
View File

@ -24,8 +24,8 @@ struct whisper_coreml_context * whisper_coreml_init(const char * path_model) {
// select which device to run the Core ML model on // select which device to run the Core ML model on
MLModelConfiguration *config = [[MLModelConfiguration alloc] init]; MLModelConfiguration *config = [[MLModelConfiguration alloc] init];
//config.computeUnits = MLComputeUnitsCPUAndGPU; config.computeUnits = MLComputeUnitsCPUAndGPU;
config.computeUnits = MLComputeUnitsCPUAndNeuralEngine; //config.computeUnits = MLComputeUnitsCPUAndNeuralEngine;
//config.computeUnits = MLComputeUnitsAll; //config.computeUnits = MLComputeUnitsAll;
const void * data = CFBridgingRetain([[whisper_encoder_impl alloc] initWithContentsOfURL:url_model configuration:config error:nil]); const void * data = CFBridgingRetain([[whisper_encoder_impl alloc] initWithContentsOfURL:url_model configuration:config error:nil]);

13
whisper.cpp
View File

@ -136,6 +136,19 @@ static void ggml_graph_compute_helper(std::vector<uint8_t> & buf, ggml_cgraph *
ggml_graph_compute(graph, &plan); ggml_graph_compute(graph, &plan);
} }
// faster matrix multiplications for tensors that do not have dimension 0 divisible by "pad"
// the idea is to represent the original matrix multiplication:
//
// Z = X @ Y
//
// with two matrix multiplications:
//
// Z = [X_0; X_1] @ [Y_0; Y_1]
//
// here X_0 and Y_0 are views of X and Y that have dimension 0 divisible by "pad"
// and X_1 and Y_1 are the remaining views. X_1 and Y_1 end up being small matrices that can be processed with more
// general-purpose kernels
//
static struct ggml_tensor * ggml_mul_mat_pad(struct ggml_context * ctx, struct ggml_tensor * x, struct ggml_tensor * y, int pad = 32) { static struct ggml_tensor * ggml_mul_mat_pad(struct ggml_context * ctx, struct ggml_tensor * x, struct ggml_tensor * y, int pad = 32) {
//#if !defined(GGML_USE_METAL) //#if !defined(GGML_USE_METAL)
// return ggml_mul_mat(ctx, x, y); // return ggml_mul_mat(ctx, x, y);