llama-bench : add no-kv-offload parameter (ggml-org#4812)

CUDA: fixed redundant value dequantization (ggml-org#4809)

llama : remove unused vars (ggml-org#4796)

llama : remove redundant GQA check (ggml-org#4796)

llama.swiftui : use llama.cpp as SPM package (ggml-org#4804)

llama : print tensor meta for debugging

ggml : use __builtin_amdgcn_sudot4 in __dp4a for gfx11 (ggml-org#4787)

server : fix n_predict check (ggml-org#4798)

llama.swiftui : use correct pointer for llama_token_eos (ggml-org#4797)

cmake : check for openblas64 (ggml-org#4134)

openblas v0.3.22 64-bit pkg-config file is named openblas64.pc
OpenMathLib/OpenBLAS#3790