sync : ggml (#2001)

whisper : improve handling of prompts (#1981)

whisper : improve support for distil-large-v3 (#1982)

ruby : fix build (#1980)

docker : libcuda.so.1 in PATH (#1966)

readme : add Fedora dependencies (#1970)

whisper : token-level timestamps with DTW (#1485)

examples : rename --audio-context to --audio-ctx per help text (#1953)

whisper : set outputs from conv graph (#1959)

alloc : fix allocation data of pre-allocated leafs

cmake : copy ggml-common.h to bin

gitignore : .vimspector.json

talk-llama : sync llama.cpp

sync : ggml

update examples and tests

ggml : add ggml-common.h

ggml : designate enum vals for integer types (llama/6050)

metal : build metallib + fix embed path (llama/6015)

llama : add pipeline parallelism support (llama/6017)

Update get version (llama/6025)

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

ggml : fix UB in IQ2_S and IQ3_S (llama/6012)

sycl : update IQ1_S kernels (WIP - not working!) (llama/5995)

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

ggml, ci : Windows ARM runner and build fixes (llama/5979)

Better 1.5 bit quantization (llama/5971)

Add q3_s and q1_s (llama/5886)

metal : move mm_id indices to shared mem (llama/5982)

ggml : fix unnecessary f32 -> f16 -> f32 casts (mmla) (llama/5951)

ggml : remove old quantization functions (llama/5942)

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

llama : support Mamba Selective State Space Models (llama/5328)

extra : update sync scripts after ggml-common.h

whisper : document whisper_batch.n_seq_id (#1942)

whisper : improve beam search candidate diversity (#1947)

bindings/go : add linker flags to make metal work (#1944)

whisper : make beam candidate sort more stable (#1943)

ggml : try fix 32-bit arm compat (#1938)

talk-llama : use llama_decode instead of llama_eval

talk-llama : sync llama.cpp

talk-llama : sync llama.cpp

sync : ggml

Revert "[SYCL] fix error when set main gpu to non-zero (llama/5901)" (llama/5918)

fix error when set main gpu to non-zero (llama/5901)

ggml : use SYS_get_cpu if SYS_getcpu is not defined (llama/5906)

ggml : use `uint8x16_t` return type for `ggml_vqtbl1q_u8` (llama/5894)

add wait() to make code stable (llama/5895)

quants : use MM256_SET_M128I consistently to fix gcc 7 build (llama/5889)

Vulkan Improvements (llama/5835)

fix mul_mat fault in CI/unit-test (llama/5862)