The Llama.cpp backend can emit the phrase “failed to allocate” when it runs out of memory. Adding this check ensures such messages are correctly classified as out‑of‑memory errors, providing more accurate error handling CPU backends.
e80a865def
