llama.cpp · b11927

metal : add GGML_OP_TOP_K_MASK (#31230) * metal : add GGML_OP_TOP_K_MASK Port the fused top-k mask from the CUDA backend: a grid-stride kernel fills the output with -inf, then one threadgroup per row copies the mask values of the kept cells over it. Function constants drop the score, blocks and extra paths that a given graph does not use. Assisted-by: dsh:llama.cpp/ggml-org/Qwen3.8-Flash-Next-GGUF (MacBook, M5 Max, 128GB) * metal : clean up top-k mask kernel formatting Assisted-by: dsh:llama.cpp/ggml-org/Qwen3.8-Flash-Next-GGUF (MacBook, M5 Max, 128GB) **Website:** - **Attestations:** - **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b11927/llama-b11927-bin-macos-arm64.tar.gz) - macOS Apple Silicon (arm64, KleidiAI enabled) [DISABLED](

1 个独立来源 · 1 篇报道切换最早在前 ↕

2026-10-27

llama.cpp releases19:08

b11927

metal：添加 GGML_OP_TOP_K_MASK（#31230）* metal：添加 GGML_OP_TOP_K_MASK 将融合后的 top-k 掩码从 CUDA 后端移植过来：一个网格步进内核先用 -inf 填充输出，然后每行一个线程组将保留单元的掩码值复制到其上。函数常量会剔除某个图中未使用的分数、块以及额外路径。协助者：dsh:llama.cpp/ggml-org/Qwen3.8-Flash-Next-GGUF（MacBook，M5 Max，128GB）* metal：清理 top-k 掩码内核的格式协助者：dsh:llama.cpp/ggml-org/Qwen3.8-Flash-Next-GGUF（MacBook，M5 Max，128GB）**网站：** - **证明文件：** - **macOS/iOS：** - [macOS 苹果芯片（arm64）](https://github.com/ggml-org/llama.cpp/releases/download/b11927/llama-b11927-bin-macos-arm64.tar.gz) - macOS 苹果芯片（arm64，已启用 KleidiAI）[已禁用](

事件进展与其他报道