mirror of
https://github.com/ggerganov/whisper.cpp.git
synced 2025-08-13 22:38:34 +02:00
metal : disable fast math in all quantize kernels (llama/14528)
ggml-ci
This commit is contained in:
@ -109,6 +109,7 @@ void dequantize_q4_0_t4(device const block_q4_0 * xb, short il, thread type4 & r
|
|||||||
}
|
}
|
||||||
|
|
||||||
void quantize_q4_0(device const float * src, device block_q4_0 & dst) {
|
void quantize_q4_0(device const float * src, device block_q4_0 & dst) {
|
||||||
|
#pragma METAL fp math_mode(safe)
|
||||||
float amax = 0.0f; // absolute max
|
float amax = 0.0f; // absolute max
|
||||||
float max = 0.0f;
|
float max = 0.0f;
|
||||||
|
|
||||||
@ -167,6 +168,7 @@ void quantize_q4_1(device const float * src, device block_q4_1 & dst) {
|
|||||||
}
|
}
|
||||||
|
|
||||||
void quantize_q5_0(device const float * src, device block_q5_0 & dst) {
|
void quantize_q5_0(device const float * src, device block_q5_0 & dst) {
|
||||||
|
#pragma METAL fp math_mode(safe)
|
||||||
float amax = 0.0f; // absolute max
|
float amax = 0.0f; // absolute max
|
||||||
float max = 0.0f;
|
float max = 0.0f;
|
||||||
|
|
||||||
@ -461,6 +463,7 @@ void dequantize_q8_0_t4(device const block_q8_0 *xb, short il, thread type4 & re
|
|||||||
}
|
}
|
||||||
|
|
||||||
void quantize_q8_0(device const float * src, device block_q8_0 & dst) {
|
void quantize_q8_0(device const float * src, device block_q8_0 & dst) {
|
||||||
|
#pragma METAL fp math_mode(safe)
|
||||||
float amax = 0.0f; // absolute max
|
float amax = 0.0f; // absolute max
|
||||||
|
|
||||||
for (int j = 0; j < QK8_0; j++) {
|
for (int j = 0; j < QK8_0; j++) {
|
||||||
|
Reference in New Issue
Block a user