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

ggml : testing GPU FP precision via quantized CPY

Browse source
This commit is contained in:
Georgi Gerganov 2023-12-30 13:22:57 +02:00
parent 24a447e20a
commit f64e4f04e7
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
3 changed files with 23 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

5
ggml-quants.c
View file

@ -1103,6 +1103,8 @@ void dequantize_row_q4_0(const block_q4_0 * restrict x, float * restrict y, int
}
}
#include <stdio.h>
void dequantize_row_q4_1(const block_q4_1 * restrict x, float * restrict y, int k) {
static const int qk = QK4_1;
@ -1110,6 +1112,9 @@ void dequantize_row_q4_1(const block_q4_1 * restrict x, float * restrict y, int
const int nb = k / qk;
printf("d = %9f\n", GGML_FP16_TO_FP32(x[0].d));
printf("m = %9f\n", GGML_FP16_TO_FP32(x[0].m));
for (int i = 0; i < nb; i++) {
const float d = GGML_FP16_TO_FP32(x[i].d);
const float m = GGML_FP16_TO_FP32(x[i].m);