sync : llama.cpp #1258
ggerganov
merged
19 commits into
master
from
sync-llama.cpp-25-06-01
ggml : add ggml_repeat_4d (llama/13824)
f2df32a4
SYCL: add gelu_erf kernel (llama/13749)
cf971a85
vulkan: use timestamp queries for GGML_VULKAN_PERF (llama/13817)
0c6e4fc1
opencl: mark `mul_mat` `f32f32` as supporting non-contiguous tensors …
78e76028
opencl: add new ops - `argsort`, `div`, `sub`, `addrows`, `sigmoid`, …
8a703636
CANN: Add SOC TYPE printing in cmake configuration (llama/13837)
abd78ce4
CUDA: fix FA tg at long context for CC >= 8.9 (llama/13852)
1d346a5c
ggml: aarch64: Implement SVE F32 kernels for vector functions (llama/…
0508222f
ggml: aarch64: Implement SVE F32 kernels for Mamba Sequential Scan Al…
2efa671d
cmake: Factor out CPU architecture detection (llama/13883)
5c0838a2
arm64: optimize q4_k_q8_k kernel with i8mm (llama/13886)
ec3cddea
cmake: Guard GGML_CPU_ALL_VARIANTS by architecture (llama/13890)
7fe6f98c
SYCL: Add mrope kernel (llama/13755)
d31232fa
cuda : prevent using split buffers with 3d/4d matrices (llama/13919)
e2a95aad
sched : avoid changing cur_copy when a graph is already allocated (ll…
c6689ee5
CUDA: fix typo in FlashAttention code (llama/13926)
ad9d6e65
CUDA: add a prop in ggml_cuda_device_infor for distinguish iGPU or dG…
8720cc16
threading: support for GGML_SCHED_PRIO_LOW, update thread info on Win…
8b09bced
sync : llama.cpp
d0f7473c
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub