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

ggml : fix Q8_0 and Q8_1 rounding

Browse source
This commit is contained in:
Georgi Gerganov 2023-05-11 20:47:41 +03:00
parent 582a39fff5
commit 6680244838
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
1 changed files with 4 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

8
ggml.c
View file

@ -925,8 +925,8 @@ static void quantize_row_q8_0_reference(const float * restrict x, block_q8_0 * r
const float v0 = x[i*QK8_0 + 2*j + 0]*id;
const float v1 = x[i*QK8_0 + 2*j + 1]*id;
y[i].qs[ j] = v0 + 0.5f;
y[i].qs[QK8_0/2 + j] = v1 + 0.5f;
y[i].qs[ j] = roundf(v0);
y[i].qs[QK8_0/2 + j] = roundf(v1);
}
}
}
@ -1083,8 +1083,8 @@ static void quantize_row_q8_1_reference(const float * restrict x, block_q8_1 * r
const float v0 = x[i*QK8_1 + 2*j + 0]*id;
const float v1 = x[i*QK8_1 + 2*j + 1]*id;
y[i].qs[ j] = v0 + 0.5f;
y[i].qs[QK8_1/2 + j] = v1 + 0.5f;
y[i].qs[ j] = roundf(v0);
y[i].qs[QK8_1/2 + j] = roundf(v1);
sum += y[i].qs[ j];
sum += y[i].qs[QK8_1/2 + j];