metal : fix the crash caused by the lack of residency set support on Intel Macs. (llama/11904)

metal : optimize dequant q6_K kernel (llama/11892)

repo : update links to new url (llama/11886)

vulkan: initial support for IQ1_S and IQ1_M quantizations (llama/11528)

opencl: Fix rope and softmax (llama/11833)

cuda : add ampere to the list of default architectures (llama/11870)

ggml: optimize some vec dot functions for LoongArch ASX (llama/11842)

vulkan: linux builds + small subgroup size fixes (llama/11767)

llamafile: use member variable instead of constant for iq4nlt (llama/11780)

musa: bump MUSA SDK version to rc3.1.1 (llama/11822)

ggml-cpu : add chunking support to mul_mat_id (llama/11666)

ggml : x2 speed for WASM by optimizing SIMD (llama/11453)

HIP: Remove GCN from list of devices that avoid MMQ (llama/11831)

HIP: Switch to std::vector in rocblas version check (llama/11820)

ggml : fix multi-threaded clamp_f32 (llama/11824)

ggml-cpu: Fix duplicate MATMUL_INT8 (llama/11817)

CUDA: fix CUDART_VERSION checks (llama/11821)

Fix #11802: Compile bug - RegQueryValueExA changed to RegQueryValueEx (llama/11803)

CUDA: use arch list for compatibility check (llama/11775)

fix: typos in documentation files (llama/11791)

vulkan: Make Vulkan optional at runtime (ggml/11493). (llama/11494)

vulkan: add environment variable GGML_VK_PREFER_HOST_MEMORY to avoid VRAM allocation (llama/11592)

vulkan: account for lookup tables when checking shared memory size (llama/11502)

ggml: Fix data race in ggml threadpool (llama/11736)

CUDA: fix min. version for movmatrix (llama/11751)

vulkan: print shared memory size (llama/11719)

SYCL: remove XMX info from print devices (llama/11712)

ggml : optimize and build warning fix for LoongArch (llama/11709)

SYCL: Adjust support condition for norm operators (llama/11674)

ggml : fix LoongArch compile error with 128-bit SIMD (llama/11701)

vulkan: optimize coopmat2 iq2/iq3 callbacks (llama/11521)

vulkan: initial support for IQ4_XS quantization (llama/11501)

vulkan: use smaller combined allocations to avoid fragmentation (llama/11551)

metal : avoid breaking build when metal API predates TARGET_OS_VISION (llama/11690)

metal : adjust support conditions for norm operators (llama/11671)

CUDA: support for mat. mul. with ne03 != ne13 (llama/11656)

CUDA: non-contiguous (RMS) norm support (llama/11659)

HIP: force max threads per block to be 1024 (llama/11621)

metal : use residency set for other platforms (llama/11648)

rpc: fix known RCE in rpc-server (ggml/1103)

cmake : fix compile assumptions for power9/etc (#2777)

CUDA: fix Volta FlashAttention logic (llama/11615)

HIP: fix flash_attn_stream_k_fixup warning (llama/11604)

CUDA/HIP: add support for selectable warp size to mmv (llama/11519)

HIP: add GGML_CUDA_CC_IS_* for amd familys as increasing cc archtectures for amd gpus are not supersets of eatch other (llama/11601)

CUDA: use mma PTX instructions for FlashAttention (llama/11583)

`ci`: use sccache on windows instead of ccache (llama/11545)

HIP: require at least HIP 5.5

HIP: Prepare reduction operators for wave 64

CUDA/HIP: add warp_size to cuda_device_info