Spaces:
Sleeping
Sleeping
Revert GELU change
Browse filesSeems it does not work on x86 for some reason
ggml.c
CHANGED
|
@@ -723,21 +723,21 @@ inline static void ggml_vec_gelu_f16(const int n, ggml_fp16_t * y, const ggml_fp
|
|
| 723 |
}
|
| 724 |
}
|
| 725 |
|
| 726 |
-
inline static void ggml_vec_gelu_f32(const int n, float * y, const float * x) {
|
| 727 |
-
uint16_t t;
|
| 728 |
-
for (int i = 0; i < n; ++i) {
|
| 729 |
-
ggml_fp16_t fp16 = ggml_fp32_to_fp16(x[i]);
|
| 730 |
-
memcpy(&t, &fp16, sizeof(uint16_t));
|
| 731 |
-
y[i] = table_gelu_f16[t];
|
| 732 |
-
}
|
| 733 |
-
}
|
| 734 |
-
|
| 735 |
//inline static void ggml_vec_gelu_f32(const int n, float * y, const float * x) {
|
|
|
|
| 736 |
// for (int i = 0; i < n; ++i) {
|
| 737 |
-
//
|
|
|
|
|
|
|
| 738 |
// }
|
| 739 |
//}
|
| 740 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 741 |
inline static void ggml_vec_sum_f32 (const int n, float * s, const float * x) { ggml_float sum = 0.0; for (int i = 0; i < n; ++i) sum += x[i]; *s += sum; }
|
| 742 |
inline static void ggml_vec_norm_inv_f32(const int n, float * s, const float * x) { ggml_vec_norm_f32(n, s, x); *s = 1./(*s); }
|
| 743 |
|
|
|
|
| 723 |
}
|
| 724 |
}
|
| 725 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 726 |
//inline static void ggml_vec_gelu_f32(const int n, float * y, const float * x) {
|
| 727 |
+
// uint16_t t;
|
| 728 |
// for (int i = 0; i < n; ++i) {
|
| 729 |
+
// ggml_fp16_t fp16 = ggml_fp32_to_fp16(x[i]);
|
| 730 |
+
// memcpy(&t, &fp16, sizeof(uint16_t));
|
| 731 |
+
// y[i] = table_gelu_f16[t];
|
| 732 |
// }
|
| 733 |
//}
|
| 734 |
|
| 735 |
+
inline static void ggml_vec_gelu_f32(const int n, float * y, const float * x) {
|
| 736 |
+
for (int i = 0; i < n; ++i) {
|
| 737 |
+
y[i] = ggml_gelu_f32(x[i]);
|
| 738 |
+
}
|
| 739 |
+
}
|
| 740 |
+
|
| 741 |
inline static void ggml_vec_sum_f32 (const int n, float * s, const float * x) { ggml_float sum = 0.0; for (int i = 0; i < n; ++i) sum += x[i]; *s += sum; }
|
| 742 |
inline static void ggml_vec_norm_inv_f32(const int n, float * s, const float * x) { ggml_vec_norm_f32(n, s, x); *s = 1./(*s); }
|
| 743 |
|