Skip to content

Commit

Permalink
release : v1.4.1
Browse files Browse the repository at this point in the history
  • Loading branch information
@ggerganov
ggerganov committed Apr 30, 2023
1 parent 30edc4c commit af745e4
Show file tree
Hide file tree
Showing 3 changed files with 16 additions and 14 deletions.
2 changes: 1 addition & 1 deletion Makefile
View file
Original file line number Diff line number Diff line change
Expand Up @@ -30,7 +30,7 @@ publish: publish-trigger
\n\
cd /path/to/whisper.cpp/bindings/ios\n\
git commit\n\
git tag 1.4.0\n\
git tag 1.4.1\n\
git push origin master --tags\n\
"

Expand Down
24 changes: 12 additions & 12 deletions Sources/whisper/ggml.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -1911,8 +1911,8 @@ static void dequantize_row_q5_0(const void * restrict vx, float * restrict y, in
const uint8_t vi = pp[l/2];

// extract the 5-th bit from qh
const uint8_t vh0 = ((qh & (1 << (l + 0))) >> (l + 0)) << 4;
const uint8_t vh1 = ((qh & (1 << (l + 1))) >> (l + 1)) << 4;
const uint8_t vh0 = ((qh & (1u << (l + 0))) >> (l + 0)) << 4;
const uint8_t vh1 = ((qh & (1u << (l + 1))) >> (l + 1)) << 4;

const int8_t vi0 = (vi & 0x0F) | vh0;
const int8_t vi1 = (vi >> 4) | vh1;
Expand Down Expand Up @@ -1948,8 +1948,8 @@ static void dequantize_row_q5_1(const void * restrict vx, float * restrict y, in
const uint8_t vi = pp[l/2];

// extract the 5-th bit from qh
const uint8_t vh0 = ((qh & (1 << (l + 0))) >> (l + 0)) << 4;
const uint8_t vh1 = ((qh & (1 << (l + 1))) >> (l + 1)) << 4;
const uint8_t vh0 = ((qh & (1u << (l + 0))) >> (l + 0)) << 4;
const uint8_t vh1 = ((qh & (1u << (l + 1))) >> (l + 1)) << 4;

const uint8_t vi0 = (vi & 0x0F) | vh0;
const uint8_t vi1 = (vi >> 4) | vh1;
Expand Down Expand Up @@ -3286,8 +3286,8 @@ static void ggml_vec_dot_q5_0_q8_0(const int n, float * restrict s, const void *
for (int j = 0; j < QK8_0/2; j++) {
const uint8_t v0 = x0[j];

const int x0_0h = ((qh & (1 << (2*j + 0))) >> (2*j + 0)) << 4;
const int x1_0h = ((qh & (1 << (2*j + 1))) >> (2*j + 1)) << 4;
const int x0_0h = ((qh & (1u << (2*j + 0))) >> (2*j + 0)) << 4;
const int x1_0h = ((qh & (1u << (2*j + 1))) >> (2*j + 1)) << 4;

const int x0_0 = ((v0 & 0x0F) | x0_0h) - 16;
const int x1_0 = ((v0 >> 4) | x1_0h) - 16;
Expand Down Expand Up @@ -3491,8 +3491,8 @@ static void ggml_vec_dot_q5_1_q8_1(const int n, float * restrict s, const void *
for (int j = 0; j < QK8_1/2; j++) {
const uint8_t v0 = x0[j];

const int x0_0h = ((qh & (1 << (2*j + 0))) >> (2*j + 0)) << 4;
const int x1_0h = ((qh & (1 << (2*j + 1))) >> (2*j + 1)) << 4;
const int x0_0h = ((qh & (1u << (2*j + 0))) >> (2*j + 0)) << 4;
const int x1_0h = ((qh & (1u << (2*j + 1))) >> (2*j + 1)) << 4;

const int x0_0 = (v0 & 0x0F) | x0_0h;
const int x1_0 = (v0 >> 4) | x1_0h;
Expand Down Expand Up @@ -13057,8 +13057,8 @@ size_t ggml_quantize_q5_0(const float * src, void * dst, int n, int k, int64_t *
memcpy(&qh, &y[i].qh, sizeof(qh));

for (int l = 0; l < QK5_0; l += 2) {
const uint8_t vh0 = ((qh & (1 << (l + 0))) >> (l + 0)) << 4;
const uint8_t vh1 = ((qh & (1 << (l + 1))) >> (l + 1)) << 4;
const uint8_t vh0 = ((qh & (1u << (l + 0))) >> (l + 0)) << 4;
const uint8_t vh1 = ((qh & (1u << (l + 1))) >> (l + 1)) << 4;

// cast to 16 bins
const uint8_t vi0 = ((y[i].qs[l/2] & 0x0F) | vh0) / 2;
Expand Down Expand Up @@ -13087,8 +13087,8 @@ size_t ggml_quantize_q5_1(const float * src, void * dst, int n, int k, int64_t *
memcpy(&qh, &y[i].qh, sizeof(qh));

for (int l = 0; l < QK5_1; l += 2) {
const uint8_t vh0 = ((qh & (1 << (l + 0))) >> (l + 0)) << 4;
const uint8_t vh1 = ((qh & (1 << (l + 1))) >> (l + 1)) << 4;
const uint8_t vh0 = ((qh & (1u << (l + 0))) >> (l + 0)) << 4;
const uint8_t vh1 = ((qh & (1u << (l + 1))) >> (l + 1)) << 4;

// cast to 16 bins
const uint8_t vi0 = ((y[i].qs[l/2] & 0x0F) | vh0) / 2;
Expand Down
4 changes: 3 additions & 1 deletion Sources/whisper/whisper.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -1333,7 +1333,7 @@ static bool whisper_model_load(struct whisper_model_loader * loader, whisper_con
}

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) {
read_safe(loader, ne[i]);
nelements *= ne[i];
Expand All @@ -1352,6 +1352,8 @@ static bool whisper_model_load(struct whisper_model_loader * loader, whisper_con
auto tensor = model.tensors[name.data()];
if (ggml_nelements(tensor) != nelements) {
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;
}

Expand Down

0 comments on commit af745e4

Please sign in to comment.