root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-25 02:44:36 +00:00
Code Issues Actions 9 Packages Projects Releases Wiki Activity

benchmark-matmult : do not use integer abs() on a float (#3277)

Browse Source
This commit is contained in:
Cebtenzzre 2023-09-20 12:06:08 -04:00 committed by GitHub
parent 80834daecf
commit 65c2c1c5ab
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 5 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
examples/benchmark/benchmark-matmult.cpp
View File

@ -21,7 +21,7 @@
#pragma warning(disable: 4244 4267) // possible loss of data #pragma warning(disable: 4244 4267) // possible loss of data
#endif #endif
void ggml_graph_compute_helper(std::vector<uint8_t> & buf, ggml_cgraph * graph, int n_threads) { static void ggml_graph_compute_helper(std::vector<uint8_t> & buf, ggml_cgraph * graph, int n_threads) {
struct ggml_cplan plan = ggml_graph_plan(graph, n_threads); struct ggml_cplan plan = ggml_graph_plan(graph, n_threads);
if (plan.work_size > 0) { if (plan.work_size > 0) {
@ -32,7 +32,7 @@ void ggml_graph_compute_helper(std::vector<uint8_t> & buf, ggml_cgraph * graph,
ggml_graph_compute(graph, &plan); ggml_graph_compute(graph, &plan);
} }
float tensor_sum_elements(const ggml_tensor * tensor) { static float tensor_sum_elements(const ggml_tensor * tensor) {
double sum = 0; double sum = 0;
if (tensor->type == GGML_TYPE_F32) { if (tensor->type == GGML_TYPE_F32) {
for (int j = 0; j < tensor->ne[1]; j++) { for (int j = 0; j < tensor->ne[1]; j++) {
@ -44,7 +44,7 @@ float tensor_sum_elements(const ggml_tensor * tensor) {
return sum; return sum;
} }
void tensor_dump(const ggml_tensor * tensor, const char * name) { static void tensor_dump(const ggml_tensor * tensor, const char * name) {
printf("%15s: type = %i (%5s) ne = %5" PRIi64 " x %5" PRIi64 " x %5" PRIi64 ", nb = (%5zi, %5zi, %5zi) - ", name, printf("%15s: type = %i (%5s) ne = %5" PRIi64 " x %5" PRIi64 " x %5" PRIi64 ", nb = (%5zi, %5zi, %5zi) - ", name,
tensor->type, ggml_type_name(tensor->type), tensor->type, ggml_type_name(tensor->type),
tensor->ne[0], tensor->ne[1], tensor->ne[2], tensor->nb[0], tensor->nb[1], tensor->nb[2]); tensor->ne[0], tensor->ne[1], tensor->ne[2], tensor->nb[0], tensor->nb[1], tensor->nb[2]);
@ -59,7 +59,7 @@ struct benchmark_params_struct {
int32_t n_iterations = 10; int32_t n_iterations = 10;
}; };
void print_usage(int /*argc*/, char ** argv, struct benchmark_params_struct params) { static void print_usage(int /*argc*/, char ** argv, struct benchmark_params_struct params) {
fprintf(stderr, "usage: %s [options]\n", argv[0]); fprintf(stderr, "usage: %s [options]\n", argv[0]);
fprintf(stderr, "\n"); fprintf(stderr, "\n");
fprintf(stderr, "options:\n"); fprintf(stderr, "options:\n");
@ -253,7 +253,7 @@ int main(int argc, char ** argv) {
// Check that the matrix multiplication result is in the right ballpark // Check that the matrix multiplication result is in the right ballpark
// We cannot use the exact value from the F32 multiplication because the quantizuation will be slightly different // We cannot use the exact value from the F32 multiplication because the quantizuation will be slightly different
float sum_of_Q4_result = tensor_sum_elements(gf31.nodes[0]); float sum_of_Q4_result = tensor_sum_elements(gf31.nodes[0]);
float delta = abs(sum_of_Q4_result - sum_of_F32_reference); float delta = std::abs(sum_of_Q4_result - sum_of_F32_reference);
float allowed_delta = (sum_of_F32_reference) / 1000 / 1000; // Let's accept an epsilon of 10^-6 float allowed_delta = (sum_of_F32_reference) / 1000 / 1000; // Let's accept an epsilon of 10^-6
if (delta > allowed_delta) { if (delta > allowed_delta) {