Comment by samus

The llama.cpp maintainers working on supporting Qwen3-next are also not enthused by LLM output. They had to go over everything and fix it up.

https://github.com/ggml-org/llama.cpp/pull/16095#issuecommen...