CUDA: add FP32 FlashAttention vector kernel (llama/7188)

ggml : full ALiBi support (llama/7192)

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

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

ggml : add Flash Attention (llama/5021)

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

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

feat: implemented sigmoid function (ggml/806)

llama : add Command R Plus support (llama/6491)

ggml : mul_mat_id use the same tensor for all the experts (llama/6387)

ggml: bypass code incompatible with CUDA < 11.1 (#2020)

sync : ggml (#2001)

llama : add pipeline parallelism support (llama/6017)

ggml : reuse quantum structs across backends (llama/5943)

1.5 bit: we can do even better (llama/5999)

Better 1.5 bit quantization (llama/5971)

ggml : add ggml-common.h to deduplicate shared code (llama/5940)

ggml : introduce ggml_status (ggml/750)

cuda : fix data race in soft max (llama/5853)

ggml : IQ3_S improvements (llama/5829)

add some new ops, fix some operators and add batch operations to certain operators. (ggml/747)

ggml : make i-quants work with super-blocks of 64 (CPU,Metal) (llama/5760)

IQ4_XS: a 4.25 bpw quantization (llama/5747)

cuda : replace remaining shfl_xor with calls to warp_reduce functions (llama/5744)

Adding IQ2_S and IQ2_M to complete coverage of the 2-3 bit quantization range (llama/5721)

CUDA: fix DEBUG_CUDA_MALLOC (llama/5729)

code : normalize enum names (llama/5697)

IQ3_S: a much better alternative to Q3_K (llama/5676)

Introduce backend GUIDs (ggml/743)

ggml : always define ggml_fp16_t as uint16_t (llama/5666)

sync : llama.cpp (ggml/0)

cuda : ignore peer access already enabled errors (llama/5597)

ci : enable -Werror for CUDA builds (llama/5579)

cuda, metal : fix nans in soft_max (llama/5574)

1.5 bit quantization (llama/5453)

ggml : add ALiBi support for ggml_soft_max_ext (llama/5488)

cuda : print message when initialization fails (llama/5512)

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

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

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

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

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

cuda : fix LLAMA_CUDA_F16 (llama/5262)

llava : add MobileVLM support (llama/5132)

sync : ggml (llama/0)

SOTA 3-bit quants (llama/5196)

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

ggml : add Vulkan backend (llama/2059)

cuda : fix tensor size calculation for non-split buffer (llama/5145)

cuda : fix 2-bit quants on amd hip (llama/5105)