ggml-backend : add device and backend reg interfaces (llama/9707)

Initial cmake support of SYCL for AMD GPUs (llama/9658)

vulkan : do not use tensor->extra (llama/9407)

ggml/ex: calculate accuracy in graph, adapt MNIST (ggml/980)

ggml: refactor cross entropy loss CPU impl. (ggml/976)

scripts : sync ggml-backend.cpp

whisper : fix excessive memory usage (#2443)

examples : update dr_wav.h to newer version (#2449)

talk-llama : sync llama.cpp

metal : reduce command encoding overhead (llama/9698)

sync : ggml

test: fix OPT_STEP_ADAMW for test-backend-ops (ggml/974)

vulkan : mul_mat: fix UB with small warps (ggml/952)

ggml : fix ggml_cast (ggml/973)

ggml: fix gradient allocation logic (ggml/966)

ggml : define missing HWCAP flags (llama/9684)

ggml : add run-time detection of neon, i8mm and sve (llama/9331)

Enable use to the rebar feature to upload buffers to the device. (llama/9251)

mtgpu: enable VMM (llama/9597)

ggml : remove assert for AArch64 GEMV and GEMM Q4 kernels (llama/9217)

cann: fix crash when llama-bench is running on multiple cann devices (llama/9627)

CUDA: remove bad assert (ggml/972)

vulkan : multithread pipeline creation (ggml/963)

vulkan : fix build for GGML_VULKAN_RUN_TESTS, add TFLOPS to log (ggml/961)

vulkan : argsort barriers must be under uniform control flow (ggml/951)

ggml : fix GGML_MAX_N_THREADS + improve formatting (ggml/969)

server : ffmpeg overwrite leftover temp file (#2431)

whisper : add large-v3-turbo (#2440)

tests : remove test-backend-ops (#2434)

ci : disable failing CUDA and Java builds

readme : fix references to download-ggml-model.sh (#2427)

make : remove "talk" target until updated

ggml : add ggml-cpu-impl.h (skip) (#0)

sync : ggml

talk-llama : sync llama.cpp

ggml : add AVX512DQ requirement for AVX512 builds (llama/9622)

log : add CONT level for continuing previous log entry (llama/9610)

threads: fix msvc build without openmp (llama/9615)

cuda: add q8_0->f32 cpy operation (llama/9571)

threads: improve ggml_barrier scaling with large number of threads (llama/9598)

ggml : AVX512 gemm for Q4_0_8_8 (llama/9532)

metal : use F32 prec for K*Q in vec FA (llama/9595)

Revert "[SYCL] fallback mmvq (ggml/9088)" (llama/9579)

musa: enable building fat binaries, enable unified memory, and disable Flash Attention on QY1 (MTT S80) (llama/9526)

Fix merge error in #9454 (llama/9589)

CUDA: enable Gemma FA for HIP/Pascal (llama/9581)

RWKV v6: RWKV_WKV op CUDA implementation (llama/9454)

ggml-alloc : fix list of allocated tensors with GGML_ALLOCATOR_DEBUG (llama/9573)

Update CUDA graph on scale change plus clear nodes/params (llama/9550)

examples : adapt to ggml.h changes (ggml/0)