vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

review: modify codes as review comments

Browse source
This commit is contained in:
zhou.weiguo 2024-06-14 10:54:15 +08:00
parent b7a9d40e51
commit fe59684e32
No known key found for this signature in database
GPG key ID: 952EA81D18BB2FA8
1 changed files with 17 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

18
examples/benchmark/benchmark-matmult.cpp
View file

@ -71,8 +71,23 @@ static float tensor_sum_elements(const ggml_tensor * tensor) {
}
if (tensor->type == GGML_TYPE_Q8_0) {
int blocks = 0;
block_q8_0 * quant_datas = (block_q8_0 *)tensor->data;
#if 1
ggml_type_traits_t qtype = ggml_internal_get_type_traits(tensor->type);
float * float32 = (float*)malloc((tensor->ne[0] * tensor->ne[1]) * sizeof(float));
if (NULL == float32) {
printf("malloc failed\n");
return 0.0;
}
qtype.to_float(quant_datas, float32, tensor->ne[0] * tensor->ne[1]);
for (int j = 0; j < tensor->ne[1]; j++) {
for (int k = 0; k < tensor->ne[0]; k++) {
sum += float32[j * tensor->ne[0] + k];
}
}
free(float32);
#else
int blocks = 0;
for (int j = 0; j < tensor->ne[1]; j++) {
blocks = tensor->ne[0] / QK8_0;
for (int i = 0; i < blocks; i++) {
@ -82,6 +97,7 @@ static float tensor_sum_elements(const ggml_tensor * tensor) {
}
}
}
#endif
}
return sum;