metal : small-batch mat-mul kernels (llama/10581)

SYCL: Fix and switch to GGML_LOG system instead of fprintf (llama/10579)

ggml-cpu: replace AArch64 NEON assembly with intrinsics in ggml_gemv_q4_0_4x4_q8_0() (llama/10567)

vulkan: Dynamic subgroup size support for Q6_K mat_vec (llama/10536)

ggml : fix I8MM Q4_1 scaling factor conversion (llama/10562)

ggml-cpu: fix typo in gemv/gemm iq4_nl_4_4 (llama/10580)

sycl : offload of get_rows set to 0 (llama/10432)

sycl : Reroute permuted mul_mats through oneMKL (llama/10408)

CANN: RoPE operator optimization (llama/10563)

vulkan: get the first command buffer submitted sooner (llama/10499)

ggml : remove redundant copyright notice + update authors

ggml : fix row condition for i8mm kernels (llama/10561)

cmake : fix ARM feature detection (llama/10543)

ggml-cpu: support IQ4_NL_4_4 by runtime repack (llama/10541)

kompute : improve backend to pass test_backend_ops (llama/10542)

CANN: Fix SOC_TYPE compile bug (llama/10519)

CANN: ROPE operator optimization (llama/10540)

Add some minimal optimizations for CDNA (llama/10498)

metal : fix group_norm support condition (llama/0)

vulkan: define all quant data structures in types.comp (llama/10440)

vulkan: Handle GPUs with less shared memory (llama/10468)

vulkan: further optimize q5_k mul_mat_vec (llama/10479)

vulkan: skip integer div/mod in get_offsets for batch_idx==0 (llama/10506)

vulkan: optimize Q2_K and Q3_K mul_mat_vec (llama/10459)

mtgpu: Add MUSA_DOCKER_ARCH in Dockerfiles && update cmake and make (llama/10516)

vulkan: fix group_norm (llama/10496)

cmake : enable warnings in llama (llama/10474)

ggml-cpu: cmake add arm64 cpu feature check for macos (llama/10487)

CANN: Improve the Inferencing Performance for Ascend NPU Device (llama/10454)

CANN: RoPE and CANCAT operator optimization (llama/10488)

vulkan: Fix a vulkan-shaders-gen arugment parsing error (llama/10484)

metal : enable mat-vec kernels for bs <= 4 (llama/10491)

llama : accept a list of devices to use to offload a model (llama/10497)

ggml : add support for dynamic loading of backends (llama/10469)

metal : minor code formatting

ggml : do not use ARM features not included in the build (llama/10457)

CANN: Support Ascend310P to accelerate F32 and F16 Model (llama/10216)

cuda : optimize argmax (llama/10441)

vulkan: predicate max operation in soft_max shaders/soft_max (llama/10437)

vulkan: copy iq4_nl LUT into shared memory (llama/10409)

vulkan: further optimize mul_mat_vec using larger loads (llama/10387)

add cmake rvv support (llama/10411)

CUDA: remove unnecessary warp reduce in FA (ggml/1032)

feat: add `GGML_UNARY_OP_ARGMAX` Metal kernel (ggml/1019)

metal : add `GGML_OP_CONV_TRANSPOSE_1D` kernels (ggml/1026)

Do not include arm_neon.h when compiling CUDA code (ggml/1028)

ggml-opt: fix data corruption (ggml/1022)

ruby : Add low-level methods to transcribe (#2585)

models : add `q8_0` models to `download-ggml-model.sh` (#2589)

ruby : Follow source tree change (#2580)