happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Address review comment

This commit is contained in:
Rohan 2025-09-29 16:53:39 +00:00
parent 10955f24d1
commit e211822fcb
1 changed files with 5 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
tests/test-quantize-fns.cpp
View File

@ -176,21 +176,14 @@ int main(int argc, char * argv[]) {
printf("%5s dot product error: %s (%f)\n", ggml_type_name(type), RESULT_STR[failed], vec_dot_error); printf("%5s dot product error: %s (%f)\n", ggml_type_name(type), RESULT_STR[failed], vec_dot_error);
} }
// Test i8mm path (nrc=2) for supported types // Test nrc=2 path for types that support it
if (type == GGML_TYPE_Q4_0 || type == GGML_TYPE_Q4_1 || type == GGML_TYPE_Q8_0 || if (qfns_cpu->nrows == 2) {
type == GGML_TYPE_Q4_K || type == GGML_TYPE_Q6_K) { const float vec_dot_error_nrc2 = dot_product_error(qfns, qfns_cpu, test_size, test_data.data(), test_data2.data(), 2);
#if defined(__ARM_FEATURE_MATMUL_INT8) failed = !(vec_dot_error_nrc2 < max_allowed_error);
const float vec_dot_error_i8mm = dot_product_error(qfns, qfns_cpu, test_size, test_data.data(), test_data2.data(), 2);
failed = !(vec_dot_error_i8mm < max_allowed_error);
num_failed += failed; num_failed += failed;
if (failed || verbose) { if (failed || verbose) {
printf("%5s dot product error (i8mm): %s (%f)\n", ggml_type_name(type), RESULT_STR[failed], vec_dot_error_i8mm); printf("%5s dot product error (nrc=2): %s (%f)\n", ggml_type_name(type), RESULT_STR[failed], vec_dot_error_nrc2);
} }
#else
if (verbose) {
printf("%5s dot product (i8mm): SKIPPED (not supported)\n", ggml_type_name(type));
}
#endif
} }
} }
} }