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

ggml : fix more imatrix nan cases

Browse source
This commit is contained in:
slaren 2025-02-09 18:15:02 +01:00
parent 19d3c8293b
commit cfb0ae7e4c
1 changed files with 6 additions and 6 deletions
Download patch file Download diff file Expand all files Collapse all files

12
ggml/src/ggml-quants.c
View file

@ -384,7 +384,7 @@ static float make_qx_quants(int n, int nmax, const float * restrict x, int8_t *
float ax = fabsf(x[i]); float ax = fabsf(x[i]);
if (ax > amax) { amax = ax; max = x[i]; } if (ax > amax) { amax = ax; max = x[i]; }
} }
if (amax < GROUP_MAX_EPS) { // all zero if (fabsf(amax) < GROUP_MAX_EPS) { // all zero
for (int i = 0; i < n; ++i) { for (int i = 0; i < n; ++i) {
L[i] = 0; L[i] = 0;
} }
@ -829,7 +829,7 @@ static float make_qp_quants(int n, int nmax, const float * restrict x, uint8_t *
for (int i = 0; i < n; ++i) { for (int i = 0; i < n; ++i) {
max = MAX(max, x[i]); max = MAX(max, x[i]);
} }
if (!max) { // all zero if (fabsf(max) < GROUP_MAX_EPS) { // all zero
for (int i = 0; i < n; ++i) { L[i] = 0; } for (int i = 0; i < n; ++i) { L[i] = 0; }
return 0.f; return 0.f;
} }
@ -3021,7 +3021,7 @@ static void quantize_row_iq2_xxs_impl(const float * restrict x, void * restrict
} }
float max = xval[0]; float max = xval[0];
for (int i = 1; i < 32; ++i) max = MAX(max, xval[i]); for (int i = 1; i < 32; ++i) max = MAX(max, xval[i]);
if (max < GROUP_MAX_EPS) { if (fabsf(max) < GROUP_MAX_EPS) {
scales[ib] = 0; scales[ib] = 0;
memset(L, 0, 32); memset(L, 0, 32);
continue; continue;
@ -3197,7 +3197,7 @@ static void quantize_row_iq2_xs_impl(const float * restrict x, void * restrict v
} }
float max = xval[0]; float max = xval[0];
for (int i = 1; i < 16; ++i) max = MAX(max, xval[i]); for (int i = 1; i < 16; ++i) max = MAX(max, xval[i]);
if (max < GROUP_MAX_EPS) { if (fabsf(max) < GROUP_MAX_EPS) {
scales[ib] = 0; scales[ib] = 0;
memset(L, 0, 16); memset(L, 0, 16);
continue; continue;
@ -3638,7 +3638,7 @@ static void quantize_row_iq3_xxs_impl(int grid_size, const float * restrict x, v
} }
float max = xval[0]; float max = xval[0];
for (int i = 1; i < 32; ++i) max = MAX(max, xval[i]); for (int i = 1; i < 32; ++i) max = MAX(max, xval[i]);
if (max < GROUP_MAX_EPS_IQ3_XXS) { if (fabsf(max) < GROUP_MAX_EPS_IQ3_XXS) {
scales[ib] = 0; scales[ib] = 0;
memset(L, 0, 32); memset(L, 0, 32);
continue; continue;
@ -4808,7 +4808,7 @@ static void quantize_row_iq2_s_impl(const float * restrict x, void * restrict vy
} }
float max = xval[0]; float max = xval[0];
for (int i = 1; i < 16; ++i) max = MAX(max, xval[i]); for (int i = 1; i < 16; ++i) max = MAX(max, xval[i]);
if (max < GROUP_MAX_EPS_IQ2_S) { if (fabsf(max) < GROUP_MAX_EPS_IQ2_S) {
scales[ib] = 0; scales[ib] = 0;
continue; continue;
} }