CUDA: generalize FP16 fattn vec kernel (llama/7061)

opencl : alignment size converted from bits to bytes (llama/7090)

Introduction of CUDA Graphs to LLama.cpp (llama/6766)

metal : use `vm_allocate` instead of `posix_memalign` on macOS (llama/7078)

ggml : introduce bfloat16 support (llama/6412)

metal : fix unused warning

Add an option to build without CUDA VMM (llama/7067)

gguf-split: add --no-tensor-first-split (llama/7072)

CUDA: CUDART < 11.7 workaround for __hmax, __hmax2 (llama/7019)

switch to using localizedDescription (llama/7010)

metal : remove deprecated error code (llama/7008)

metal : log more info on error (llama/6987)

ggml : add Flash Attention (llama/5021)

ggml : fix __MSC_VER -> _MSC_VER (llama/6977)

Fix more int overflow during quant (PPL/CUDA). (llama/6563)

gguf : enforce that tensor names are unique (llama/6905)

add device version in device list (llama/6959)

Reset schedule earlier to allow overlap with ggml graph computation on device (llama/6933)

add basic tensor data validation function (llama/6884)

gguf : fix mismatch between alloc and free functions (llama/6929)

Merge pull request from GHSA-p5mv-gjc5-mwqv

ggml : fix redefinition of vaddvq_f32 for 32-bit ARM (llama/6906)

ggml : fix MIN / MAX macros (llama/6904)

ggml : move 32-bit arm compat in ggml-impl.h (llama/6865)

llamafile : improve sgemm.cpp (llama/6796)

ggml : fix calloc argument ordering. (llama/6820)

ggml : fix ggml_backend_cpu_supports_op() for CPY (llama/0)

ggml : group all experts in a single ggml_mul_mat_id (llama/6505)

ggml : fix llamafile sgemm wdata offsets (llama/6710)

ggml : add llamafile sgemm (llama/6414)

llama : add qwen2moe (llama/6074)

fix mul_mat_id() for new input, make the ut pass (llama/6682)

Added support for GGML_OP_CLAMP in Metal (llama/6662)

fix memcpy() crash, add missed cmd in guide, fix softmax (llama/6622)

CUDA: fix matrix multiplication logic for tests (llama/6667)

metal : unify mul_mv_id kernels (llama/6556)

llama : add gguf_remove_key + remove split meta during quantize (llama/6591)

feat: implemented sigmoid function (ggml/806)

build: fix and ignore msvc warnings (ggml/805)

ggml : expose SSE3 and SSSE3 for MSVC when AVX is available (#2128)

build : improve disabling AVX-512 (#2129)

minor: add CMakeSettings.json to gitignore (#2094)

examples : fix node compilation (#2115)

make : change GNU make default CXX from g++ to c++ (#2100)

Remove unnecessary memory reallocation in fft (#2080)

models : disable old script (#2079)

whisper : more prominent log message for sub-1s audio (#2065)

main : pass nullptr when regex is empty (#2070)

readme : add up-to-date repository for Python bindings (#2063)

release : v1.5.5