Community
Pull request adds QSA selection without kq_mask to llama.cpp for Qwen models
The llama.cpp pull request removes the kq_mask requirement from QSA selection, which the post says saves gigabytes of VRAM for Qwen Flash Next.
Community
The llama.cpp pull request removes the kq_mask requirement from QSA selection, which the post says saves gigabytes of VRAM for Qwen Flash Next.