Qwen3.8-27B MTP Q4S head on Q6 quant is faster than q6, f16 slower = acceptance

Posted 1 hour ago by xlayn
1 points

https://github.com/alainnothere/llama.cpp/commit/77b124a7ab56e0f1c33c025344a9c63465a9a87f

1 comments

Loading..