llama.cpp

e98c1c18 - test: fix OPT_STEP_ADAMW for test-backend-ops (ggml/974)