CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-30 08:42:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

check for nans in imatrix and quantize (#7807)

Browse Source 
* imatrix : detect nan/inf values

* quantize : check imatrix for nan/inf values
This commit is contained in:
slaren
2024-06-07 08:01:29 +02:00
committed by GitHub
parent ee459f40f6
commit c9ee7118d5
2 changed files with 16 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
examples/imatrix/imatrix.cpp
View File

@@ -151,6 +151,10 @@ bool IMatrixCollector::collect_imatrix(struct ggml_tensor * t, bool ask, void *
for (int j = 0; j < (int)src1->ne[0]; ++j) {
e.values[e_start + j] += x[j]*x[j];
e.counts[e_start + j]++;
if (!std::isfinite(e.values[e_start + j])) {
fprintf(stderr, "%f detected in %s\n", e.values[e_start + j], wname.c_str());
exit(1);
}
}
}
}
@@ -183,6 +187,10 @@ bool IMatrixCollector::collect_imatrix(struct ggml_tensor * t, bool ask, void *
for (int j = 0; j < (int)src1->ne[0]; ++j) {
e.values[j] += x[j]*x[j];
e.counts[j]++;
if (!std::isfinite(e.values[j])) {
fprintf(stderr, "%f detected in %s\n", e.values[j], wname.c_str());
exit(1);
}
}
}
if (e.ncall > m_last_call) {

8
llama.cpp
View File

@@ -15237,6 +15237,14 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
if (imatrix_data) {
LLAMA_LOG_INFO("================================ Have weights data with %d entries\n",int(imatrix_data->size()));
qs.has_imatrix = true;
// check imatrix for nans or infs
for (const auto & kv : *imatrix_data) {
for (float f : kv.second) {
if (!std::isfinite(f)) {
throw std::runtime_error(format("imatrix contains non-finite value %f\n", f));
}
}
}
}
}