llama.cpp
23ffc507 - metal : derive matrix types in the remaining kernels

Commit
57 minutes ago
metal : derive matrix types in the remaining kernels Use the value_type_* helpers in the flash-attention, top-k-mask, mul-mv and count-equal kernels to derive the matrix/vector types from the underlying scalar types, and simplify the FA_TYPES macros. Assisted-by: dsh:llama.cpp/ggml-org/DeepSeek-V4.1-Flash-1017-GGUF (Mac Studio, M5 Ultra, 256GB)
Author
Parents
Loading