ggml-cpu: faster AVX2 variant for IQ1_M (llama/12216)

metal : simplify kernel arguments using a struct (ggml/3229) (llama/12194)

metal : fix default.metallib build (llama/12224)

opencl: Noncontiguous `norm`, `rms_norm`, disable `fp16` for some ops (llama/12217)

cmake : fix undefined reference errors for std::filesystem in ggml (#12092) (llama/12094)

CUDA: fix FA logic for PTX 7.0 and CC >= 7.5 (llama/12222)

HIP/CUDA: set the paramerter value in maintain_cuda_graph instead of replaceing it. (llama/12209)

opencl : fix buffer alignment (llama/12197)

opencl : fix `ulong` kernel args were set from `int` variables (llama/12174)

opencl : fix profile-related errors (llama/12095)

ggml-cpu: Faster IQ1 mul_mat_vec on AVX2 using BMI2 instructions (llama/12154)

SYCL: Disable f16 Unary OPs as not supported by the kernels (llama/12201)

ggml : fix GGMLMetalClass ODR (llama/12200)

ggml : ggml_compute_forward_concat() for arbitrary tensor type (ggml/1118)

vulkan : sync (llama/0)

ggml : portability fixes for VS 2017 (llama/12150)

HIP: implement FlashAttention via rocWMMA for CDNA and RDNA3+ (llama/12032)

ggml : fix kleidiai build (llama/12159)

SYCL: Move CPY kernels to a separate file and add few missing kernels (llama/12133)

ggml-backend : keep paths in native string type when possible (llama/12144)

CUDA: compress mode option and default to size (llama/12029)

ggml : upgrade init_tensor API to return a ggml_status (llama/11854)

vulkan: add specific MMV kernels for IQ2 and IQ3 quants + optimizations (llama/11595)

CUDA: fix logic for V100 + GGML_CUDA_FORCE_MMQ (llama/12098)

ggml: aarch64: implement SVE kernels for q2_k_q8_k vector dot (llama/12064)

CANN: Fix build error with GCC 13 (llama/11990)

vulkan: matmul dequantization improvements (llama/12015)

vulkan: improve im2col (llama/11826)

vulkan: fix assertion when qy_needs_dequant (llama/12068)

ggml-cpu: Fix build with sve (llama/12059)

cuda: unary ops as float + de-duplicate (ggml/1130)

cuda/vulkan: specify fp32-only support for some operations in supports_op (ggml/1129)

cuda/cpu: Increase support for fp16 unary operations (ggml/1125)

whisper : support GGML_BACKEND_DL (#2843)

Support pure float16 add/sub/mul/div operations in the CUDA (and CPU) backend (ggml/1121)

metal : copy kernels for quant to F32/F16 conversions (llama/12017)

opencl: fix for small models (llama/11950)

Optimize mul_mat for Q4_0 on Intel GPU (llama/12035)

SYCL: Fix GGML_SYCL_DEBUG macro (llama/11995)

ggml-cpu: Support s390x SIMD Instruction Set (llama/12019)

CUDA: app option to compile without FlashAttention (llama/12025)

CUDA: optimize FA for GQA + large batches (llama/12014)

cuda: Add Q5_1, Q5_0, Q4_1 and Q4_0 to F32 conversion support. (llama/12000)

CUDA: correct the lowest Maxwell supported by CUDA 12 (llama/11984)

MUSA: support ARM64 and enable dp4a .etc (llama/11843)

ggml-cpu: Add CPU backend support for KleidiAI library (llama/11390)

ggml: aarch64: implement SVE kernels for q3_K_q8_K vector dot (llama/11917)

CUDA: use async data loading for FlashAttention (llama/11894)

vulkan: implement several ops relevant for ggml_opt (llama/11769)

vulkan: support multi/vision rope, and noncontiguous rope (llama/11902)