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

review: modify codes as review suggestion

Browse source
This commit is contained in:
zhou.weiguo 2024-06-14 23:18:43 +08:00
parent 11c7b1e25a
commit 181c0e3b0f
No known key found for this signature in database
GPG key ID: 952EA81D18BB2FA8
1 changed files with 4 additions and 16 deletions
Download patch file Download diff file Expand all files Collapse all files

20
examples/benchmark/benchmark-matmult.cpp
View file

@ -33,18 +33,10 @@ static void ggml_graph_compute_helper(std::vector<uint8_t> & buf, ggml_cgraph *
} }
static float tensor_sum_elements(const ggml_tensor * tensor) { static float tensor_sum_elements(const ggml_tensor * tensor) {
double sum = 0; double sum = 0.0;
float floatvalue = 0; float floatvalue = 0.0f;
unsigned short shortvalue = 0; unsigned short shortvalue = 0;
if (tensor->type == GGML_TYPE_I8) {
for (int j = 0; j < tensor->ne[1]; j++) {
for (int k = 0; k < tensor->ne[0]; k++) {
sum += ((int8_t *) tensor->data)[j * tensor->ne[0] + k];
}
}
}
if (tensor->type == GGML_TYPE_F16) { if (tensor->type == GGML_TYPE_F16) {
for (int j = 0; j < tensor->ne[1]; j++) { for (int j = 0; j < tensor->ne[1]; j++) {
for (int k = 0; k < tensor->ne[0]; k++) { for (int k = 0; k < tensor->ne[0]; k++) {
@ -53,17 +45,13 @@ static float tensor_sum_elements(const ggml_tensor * tensor) {
sum += floatvalue; sum += floatvalue;
} }
} }
} } else 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++) {
for (int k = 0; k < tensor->ne[0]; k++) { for (int k = 0; k < tensor->ne[0]; k++) {
sum += ((float *) tensor->data)[j * tensor->ne[0] + k]; sum += ((float *) tensor->data)[j * tensor->ne[0] + k];
} }
} }
} } else if (ggml_is_quantized(tensor->type)) {
if (ggml_is_quantized(tensor->type)) {
std::vector<float> f32out(ggml_nelements(tensor)); std::vector<float> f32out(ggml_nelements(tensor));
ggml_type_traits_t qtype = ggml_internal_get_type_traits(tensor->type); ggml_type_traits_t qtype = ggml_internal_get_type_traits(tensor->type);
qtype.to_float((void *)tensor->data, f32out.data(), f32out.size()); qtype.to_float((void *)tensor->data, f32out.data(), f32out.size());