tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-30 04:45:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

imatrix : don't use FMA explicitly

Browse Source 
This should make comparisons between the formats easier
because this matches the behavior of the previous version.
This commit is contained in:
Francis Couture-Harpin
2025-06-18 16:33:37 -04:00
parent 2c0945027a
commit ba6f6be6ce
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
tools/imatrix/imatrix.cpp
View File

@ -180,7 +180,7 @@ bool IMatrixCollector::collect_imatrix(struct ggml_tensor * t, bool ask, void *
e.counts[ex]++; e.counts[ex]++;
for (int j = 0; j < (int)src1->ne[0]; ++j) { for (int j = 0; j < (int)src1->ne[0]; ++j) {
e.values[e_start + j] = std::fma(x[j], x[j], e.values[e_start + j]); e.values[e_start + j] += x[j] * x[j];
if (!std::isfinite((float)e.values[e_start + j])) { if (!std::isfinite((float)e.values[e_start + j])) {
LOG_ERR("%f detected in %s\n", (float)e.values[e_start + j], wname.c_str()); LOG_ERR("%f detected in %s\n", (float)e.values[e_start + j], wname.c_str());
exit(1); exit(1);
@ -220,7 +220,7 @@ bool IMatrixCollector::collect_imatrix(struct ggml_tensor * t, bool ask, void *
const float * x = (const float *) (data + row * src1->nb[1]); const float * x = (const float *) (data + row * src1->nb[1]);
e.counts[0]++; e.counts[0]++;
for (int j = 0; j < (int)src1->ne[0]; ++j) { for (int j = 0; j < (int)src1->ne[0]; ++j) {
e.values[j] = std::fma(x[j], x[j], e.values[j]); e.values[j] += x[j] * x[j];
if (!std::isfinite((float)e.values[j])) { if (!std::isfinite((float)e.values[j])) {
LOG_ERR("%f detected in %s\n", (float)e.values[j], wname.c_str()); LOG_ERR("%f detected in %s\n", (float)e.values[j], wname.c_str());
exit(1); exit(1);