sync : ggml

ggml-alloc : allocate all leafs as if they were inputs (ggml/731)

talk-llama : sync llama.cpp

sync : ggml

ggml-backend : sync remnant

CUDA: mul_mat_vec_q tiling, refactor mul mat logic (llama/5434)

vulkan: only use M-sized matmul on Apple GPUs (llama/5412)

ggml : fix compile warnings (unused vars) (llama/4966)

ggml : add mmla kernels for quantized GEMM (llama/4966)

metal : use autoreleasepool to avoid memory leaks (llama/5437)

ggml-alloc : v3 (ggml/727)

examples : added audio_ctx argument to main and server (#1857)

metal : option to embed MSL source into compiled binary (#1842)

examples : initialize context params properly (#1852)

talk-llama : sync llama.cpp

sync : ggml

src : relocate new backend sources

ggml : fix `error C2078: too many initializers` for MSVC ARM64 (llama/5404)

CUDA: more warps for mmvq on NVIDIA (llama/5394)

CUDA: fixed mmvq kernel for bs 2,3,4 and -sm row (llama/5386)

Basic Vulkan Multi-GPU implementation (llama/5321)

CUDA: mul_mat_vec_q max. batch size 8 -> 4 (llama/5370)

Slight quantization improvement for Q4_K and Q5_K (llama/5361)

CUDA: mul_mat_vec_q for batch sizes > 1 (llama/5351)

ggml : make use of ggml-quants.h possible in C++ code (llama/5338)

ggml : avoid duplicating function calls using MIN/MAX macros (llama/5325)

iq2_xxs: tune quantization (llama/5320)

cuda : fix LLAMA_CUDA_F16 (llama/5262)

metal : add im2col F32 dst support (llama/5132)

llava : add MobileVLM support (llama/5132)

ggml : limit n_threads to the max n_tasks (llama/5238)

kompute : llama-bench support and ggml_cpu_has_kompute() (llama/5226)

ggml : add abort_callback for cpu backend (ggml/725)

extra : update sync scripts

server : allow CORS request with authorization headers (#1850)

whisper.android : how to build with CLBlast (#1809)

whisper : expose CUDA device setting in public API (#1840)

make : add macOS deployment target option (#1839)

talk-llama : stream response (#1121)

sync : ggml (#0)

ggml : fix IQ3_XXS on Metal (llama/5219)

sync : ggml (llama/0)

Faster AVX2 dot product for IQ2_XS (llama/5187)

SOTA 3-bit quants (llama/5196)

ggml alloc: Fix for null dereference on alloc failure (llama/5200)

Nomic Vulkan backend (llama/4456)

ggml : add max buffer sizes to opencl and metal backends (llama/5181)

metal : free metal objects (llama/5161)

gguf : fix comparison (ggml/715)

`ggml_cuda_cpy` support for 4d tensors and float16->float32 upcasting (ggml/686)