Disabling benchmarks unless explicitly enabled

This commit is contained in:
Jay Booth
2026-09-10 11:50:20 +08:00
parent 47205ff58d
commit 05c511a8c9
+4
View File
@@ -764,6 +764,9 @@ int main() {
// RMSE vs the std output. Every type the tiled gate accepts; the repack column is
// n/a where no repack kernel exists for the type. TILED_MM_FORCE is set in main,
// so the tiled column runs at every shape; the shapes vary K, the output rank
/* Benchmarks disabled unless actively uncommented so we don't waste CPU in CI for unrelated changes
const ggml_type bench_types[] = { GGML_TYPE_Q2_K, GGML_TYPE_Q3_K, GGML_TYPE_Q4_K, GGML_TYPE_Q5_K, GGML_TYPE_Q6_K,
GGML_TYPE_IQ4_XS, GGML_TYPE_IQ2_XXS, GGML_TYPE_IQ2_XS, GGML_TYPE_IQ2_S,
GGML_TYPE_IQ3_XXS, GGML_TYPE_IQ3_S, GGML_TYPE_IQ1_S, GGML_TYPE_IQ1_M };
@@ -811,6 +814,7 @@ int main() {
print_mmid_table(mmid_shapes[s].K, mmid_shapes[s].R, mmid_shapes[s].E, mmid_shapes[s].k,
mmid_shapes[s].b_slots, mmid_shapes[s].batch, rows, n_mmid_types);
}
*/
ggml_backend_free(backend);
return n_failed ? 1 : 0;