mirror/whisper.cpp
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

whisper : fix quantize bug (#842)

Browse Source 
* whisper : debug

* whisper : fix bug during quantization
This commit is contained in:
Georgi Gerganov 2023-04-30 22:50:04 +03:00 committed by GitHub
parent feac80dd3f
commit c94c469592
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 9 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
examples/common-ggml.cpp
View File

@ -90,7 +90,7 @@ bool ggml_common_quantize_0(
} }
int32_t nelements = 1; int32_t nelements = 1;
int32_t ne[2] = { 1, 1 }; int32_t ne[4] = { 1, 1, 1, 1 };
for (int i = 0; i < n_dims; ++i) { for (int i = 0; i < n_dims; ++i) {
finp.read (reinterpret_cast<char *>(&ne[i]), sizeof(ne[i])); finp.read (reinterpret_cast<char *>(&ne[i]), sizeof(ne[i]));
nelements *= ne[i]; nelements *= ne[i];
@ -99,7 +99,7 @@ bool ggml_common_quantize_0(
std::string name(length, 0); std::string name(length, 0);
finp.read (&name[0], length); finp.read (&name[0], length);
printf("%64s - [%5d, %5d], type = %6s ", name.data(), ne[0], ne[1], ggml_type_name((ggml_type) ttype)); printf("%64s - [%5d, %5d, %5d], type = %6s ", name.data(), ne[0], ne[1], ne[2], ggml_type_name((ggml_type) ttype));
bool quantize = false; bool quantize = false;
@ -204,11 +204,11 @@ bool ggml_common_quantize_0(
total_size_new += cur_size; total_size_new += cur_size;
printf("size = %8.2f MB -> %8.2f MB | hist: ", nelements * sizeof(float)/1024.0/1024.0, cur_size/1024.0/1024.0); printf("size = %8.2f MB -> %8.2f MB | hist: ", nelements * sizeof(float)/1024.0/1024.0, cur_size/1024.0/1024.0);
for (int i = 0; i < hist_cur.size(); ++i) { for (int i = 0; i < (int) hist_cur.size(); ++i) {
hist_all[i] += hist_cur[i]; hist_all[i] += hist_cur[i];
} }
for (int i = 0; i < hist_cur.size(); ++i) { for (int i = 0; i < (int) hist_cur.size(); ++i) {
printf("%5.3f ", hist_cur[i] / (float)nelements); printf("%5.3f ", hist_cur[i] / (float)nelements);
} }
printf("\n"); printf("\n");
@ -226,12 +226,12 @@ bool ggml_common_quantize_0(
{ {
int64_t sum_all = 0; int64_t sum_all = 0;
for (int i = 0; i < hist_all.size(); ++i) { for (int i = 0; i < (int) hist_all.size(); ++i) {
sum_all += hist_all[i]; sum_all += hist_all[i];
} }
printf("%s: hist: ", __func__); printf("%s: hist: ", __func__);
for (int i = 0; i < hist_all.size(); ++i) { for (int i = 0; i < (int) hist_all.size(); ++i) {
printf("%5.3f ", hist_all[i] / (float)sum_all); printf("%5.3f ", hist_all[i] / (float)sum_all);
} }
printf("\n"); printf("\n");

4
whisper.cpp
View File

@ -1333,7 +1333,7 @@ static bool whisper_model_load(struct whisper_model_loader * loader, whisper_con
} }
int32_t nelements = 1; int32_t nelements = 1;
int32_t ne[3] = { 1, 1, 1 }; int32_t ne[4] = { 1, 1, 1, 1 };
for (int i = 0; i < n_dims; ++i) { for (int i = 0; i < n_dims; ++i) {
read_safe(loader, ne[i]); read_safe(loader, ne[i]);
nelements *= ne[i]; nelements *= ne[i];
@ -1352,6 +1352,8 @@ static bool whisper_model_load(struct whisper_model_loader * loader, whisper_con
auto tensor = model.tensors[name.data()]; auto tensor = model.tensors[name.data()];
if (ggml_nelements(tensor) != nelements) { if (ggml_nelements(tensor) != nelements) {
fprintf(stderr, "%s: tensor '%s' has wrong size in model file\n", __func__, name.data()); fprintf(stderr, "%s: tensor '%s' has wrong size in model file\n", __func__, name.data());
fprintf(stderr, "%s: shape: [%d, %d, %d], expected: [%d, %d, %d]\n",
__func__, ne[0], ne[1], ne[2], (int) tensor->ne[0], (int) tensor->ne[1], (int) tensor->ne[2]);
return false; return false;
} }